Method and system for determining a long-term brain state change of a patient

Abstract

Systems and methods for determining a long-term brain state change (LTBSC) of a patient. The systems and methods may obtain long-term electroencephalogram (LTEEG) data comprising at least one LTEEG signal associated with a brain of the patient and gathered via an implant device of the patient. A first model trained using first model training data may receive the LTEEG data to generate LTBSC data indicating at least one LTBSC of the patient. In response to generating the LTBSC data, the systems and methods may one or more of: (i) retrain one or more machine learning models using the LTBSC data; (ii) provide the LTBSC data as an input to the one or more machine learning models; or (iii) generate a medical insight of the patient based upon the LTBSC data, and providing, by the one or more processors, the medical insight to a user device.

Description

METHOD AND SYSTEM FOR DETERMINING A LONG-TERM BRAIN STATE CHANGE OF APATIENTTECHNICAL FIELD

[0001] The present disclosure relates to systems and methods for determining a brain state change and, in particular, systems and methods for determining a long-term brain state change of a patient.BACKGROUND OF THE DISCLOSURE

[0002] Epilepsy is considered the world’s most common serious brain disorder, with an estimated 50 million sufferers worldwide and 2.4 million new cases occurring each year.

[0003] Epilepsy is a condition of the brain characterized by epileptic seizures that vary from brief and barely detectable seizures to more conspicuous seizures in which a sufferer vigorously shakes. Epileptic seizures are unprovoked, recurrent and due to unexplained causes.

[0004] Diagnosing epilepsy typically requires detailed study of both clinical observations and electrical and / or other signals in the patient’s brain and / or body. Particularly with respect to studying electrical activity in the patient’s brain (e.g., using electroencephalography to produce an electroencephalogram (EEG)), such study usually requires the patient to be monitored for some period of time. The monitoring of electrical activity in the brain requires the patient to have a number of electrodes placed on the scalp, each of which electrodes is typically connected to a data acquisition unit that samples the signals continuously (e.g., at a high rate) to record the signals for later analysis. Medical personnel monitor the patient to watch for outward signs of epileptic or other events, and review the recorded electrical activity signals to determine whether an event occurred, whether the event was epileptic in nature and, in some cases, the type of epilepsy and / or region(s) of the brain associated with the event. Because the electrodes are wired to the data acquisition unit, and because medical personnel must monitor the patient for outward clinical signs of epileptic or other events, the patient is typically confined to a small area (e.g., a hospital or clinical monitoring room) during the period of monitoring, which is generally short-term in nature due to such restrictions for the patient. Moreover, where the number of electrodes placed on or under the patient’s scalp is significant, the size of the corresponding wire bundle coupling the sensors to the data acquisition unit may be significant, which may generally require the patient to remain generally inactive during the period of monitoring, and may prevent the patient from undertaking normal activities that may be related to the onset of symptoms.

[0005] Advancements in implantable devices which gather EEG, such as the Epi-Minder Minder® system, now provide the ability to gather long-term recordings of EEG (e.g., EEG gatheredover several days, weeks, or months) while having a minimal impact, if any, on the quality of life of the patient in which the device is implanted. Long-term EEG (LTEEG) may provide insights into the health of a patient that short-term EEG simply cannot provide. However, the algorithms and models (hereinafter generally referred to as “models”) that analyze LTEEG data are generally event-driven, detecting singular, seemingly stochastic events that may or may not be connected and / or have clinical symptoms associated with them. Even though the LTEEG can indicate slow, long-term characteristics of the patient’s brain activity, such long-term brain activity is not analyzed by such models. For example, the LTEEG can provide information associated with the “brain state” of the patient, i.e., the functional and physiological condition and / or activity level of the human brain, encompassing complex and dynamic patterns of neural activity (e.g., firing of neurons, communication between different brain regions, the overall cognitive and emotional processes occurring at a given moment, etc.), which can vary based upon factors such as wakefulness, sleep stages, emotional states, cognitive tasks, and external stimuli. Monitoring and understanding brain states provides insight into brain function, behavior, and potential changes over time. Long-term changes in the brain state can reveal trends or profiles in the patient that run independent to the stochastic events, and / or can offer additional information than the stochastic events, as even though stochastic events and long-term brain state changes (LTBSCs) may occur simultaneously, they can operate independently of each other in the sense that longer-term processes continue despite the occurrence of stochastic events. Moreover, long-term biometrics (LTBs), such as biometrics for the face, eyes, voice, may support or improve such insights when analyzed in combination with the long-term brain state changes of the patient. Thus, it is desirable to have systems and methods for determining a long-term brain state change (LTBSC) of a patient.

[0006] Any discussion of documents, acts, materials, devices, articles or the like included in the present background is not to be taken as an admission that any or all of these matters form part of the prior art base or were common general knowledge in the field relevant to the present disclosure as it existed before the priority date of each claim of this application.SUMMARY

[0007] In embodiments, a computer-implemented method for determining a long-term brain state change (LTBSC) of a patient. The computer-implemented method may include (i) obtaining, by one or more processors, long-term electroencephalogram (LTEEG) data comprising at least one LTEEG signal associated with a brain of the patient and gathered via an implant device of the patient; (ii) providing, by the one or more processors, the LTEEG data to a first model trained using first model training data to generate LTBSC data indicating at least one LTBSC of the patient; (iii) in responseto generating the LTBSC data, one or more of: (a) retraining, by the one or more processors, one or more machine learning models using the LTBSC data; (b) providing, by the one or more processors, the LTBSC data as an input to the one or more machine learning models; or (c) generating, by the one or more processors, a medical insight of the patient based upon the LTBSC data; and providing, by the one or more processors, the medical insight to a user device. The method may include additional, fewer, and / or alternate actions, including actions described in the present disclosure.

[0008] In other embodiments, a computer-implemented method for training a first model to determine a long-term brain state change (LTBSC) of a patient. The computer-implemented method may include (i) obtaining, by one or more processors, a first training dataset; (ii) receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; (iii) training, by the one or more processors, a first iteration of the first model using the first training dataset using the feature values defined for the first training dataset to generate primary LTBSC data having associated first error rates; (iv) obtaining, by the one or more processors, a second training dataset; (v) receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and (vi) training, by the one or more processors, a second iteration of the first model using the second training dataset using the feature values defined for the second training dataset to generate secondary LTBSC data having associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates. The method may include additional, fewer, and / or alternate actions, including actions described in the present disclosure.

[0009] In yet other embodiments, a computer-implemented method for training a second model to generate second model output data indicating a seizure of the patient. The method may include (i) obtaining, by one or more processors, a first training dataset; (ii) receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; (iii) training, by the one or more processors, a first iteration of the second model using the first training dataset using the feature values defined for the first training dataset to generate primary second model output data having associated first error rates; (iv) obtaining, by the one or more processors, a second training dataset; (v) receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and (vi) training, by the one or more processors, a second iteration of the second model using the second training dataset using the feature values defined for the second training dataset to generate secondary second model output data having associated second error rates, wherein the second error rates have a reducedoverall error rate compared to an overall error rate of the first error rates. The method may include additional, fewer, and / or alternate actions, including actions described in the present disclosure.

[0010] In still yet other embodiments, a computer-implemented method for training a third model to generate third model output data, the third model output data including a revision of an indication of a seizure in the long-term EEG data of the LTBSC data. The method may include (i) obtaining, by one or more processors, a first training dataset; (ii) receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; (iii) training, by the one or more processors, a first iteration of the third model using the first training dataset using the feature values defined for the first training dataset to generate primary third model output data having associated first error rates; (iv) obtaining, by the one or more processors, a second training dataset; (v) receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and (vi) training, by the one or more processors, a second iteration of the third model using the second training dataset using the feature values defined for the second training dataset to generate secondary third model output data having associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates. The method may include additional, fewer, and / or alternate actions, including actions described in the present disclosure.

[0011] In other embodiments, a computer-implemented method for training a fourth model to generate fourth model output data indicating a prediction of a seizure of the patient. The method may include (i) obtaining, by one or more processors, a first training dataset; (ii) receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; (iii) training, by the one or more processors, a first iteration of the fourth model using the first training dataset using the feature values defined for the first training dataset to generate primary fourth model output data having associated first error rates; (iv) obtaining, by the one or more processors, a second training dataset; (v) receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and (vi) training, by the one or more processors, a second iteration of the fourth model using the second training dataset using the feature values defined for the second training dataset to generate secondary fourth model output data having associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates. The method may include additional, fewer, and / or alternate actions, including actions described in the present disclosure.

[0012] In yet other embodiments, a computer-implemented method for training a fifth model to generate fifth model output data indicating at least one date to implement a medication titration ofthe patient. The method may include (i) obtaining, by one or more processors, a first training dataset; (ii) receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; (iii) training, by the one or more processors, a first iteration of the fifth model using the first training dataset using the feature values defined for the first training dataset to generate primary fifth model output data having associated first error rates; (iv) obtaining, by the one or more processors, a second training dataset; (v) receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and (vi) training, by the one or more processors, a second iteration of the fifth model using the second training dataset using the feature values defined for the second training dataset to generate secondary fifth model output data having associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates. The method may include additional, fewer, and / or alternate actions, including actions described in the present disclosure.

[0013] In still yet other embodiments, a computer-implemented method for training a sixth model to generate sixth model output data indicating an association between at least one environmental characteristic of the patient and the at least one LTBSC of the patient. The method may include (i) obtaining, by one or more processors, a first training dataset; (ii) receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; (iii) training, by the one or more processors, a first iteration of the sixth model using the first training dataset using the feature values defined for the first training dataset to generate primary sixth model output data having associated first error rates; (iv) obtaining, by the one or more processors, a second training dataset; (v) receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and (vi) training, by the one or more processors, a second iteration of the sixth model using the second training dataset using the feature values defined for the second training dataset to generate secondary sixth model output data having associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates. The method may include additional, fewer, and / or alternate actions, including actions described in the present disclosure.

[0014] In other embodiments, a system for determining a long-term brain state change (LTBSC) of a patient. The system may include (i) an implant device comprising a sensor array and a processor device, (a) the sensor array comprising a plurality of electrodes and a local processing device comprising local device communication circuitry, the sensor array being communicatively coupled to the processor device via the local processing device communication circuitry, the sensor array being configured to gather long-term electroencephalogram (LTEEG) data comprising at leastone LTEEG signal associated with a brain of the patient, and provide the LTEEG data to the processing device; and (b) the processor device comprising a microprocessor, a memory, and communication circuitry, the processing device being communicatively coupled to the sensor array via the communication circuitry; (ii) a first model, stored in the memory and configured to be executed by the microprocessor, the first model trained using first model training data and operable to: (a) receive the LTEEG data; (b) determine one or more feature values of the LTEEG data; and (c) based on the one or more feature values, generate LTBSC data indicating at least one LTBSC of the patient; and (iii) in response to generating the LTBSC data, the processor device is configured to one or more of: (a) retrain one or more machine learning models using the LTBSC data; (b) provide the LTBSC data as an input to the one or more machine learning models; or (c) generate a medical insight of the patient based upon the LTBSC data; and provide the medical insight to a user device. The system may include additional, fewer, and / or alternate components, including various components descried in the present disclosure. Moreover, the system may be configured to perform additional, fewer, and / or alternate actions, including various actions described in the present disclosure.

[0015] In yet other embodiments, a first model for determining a long-term brain state change (LTBSC) of a patient. The model may include (i) the first model, stored on one or more memories and configured to be executed by one or more processors, the first model trained using first model training data and operable to: (a) receive long-term electroencephalogram (LTEEG) data comprising at least one LTEEG signal associated with a brain of the patient and gathered via an implant device of the patient; (b) determine one or more feature values of the LTEEG data; and (c) based on the one or more feature values, generate LTBSC data indicating at least one LTBSC of the patient. The model may be configured to perform additional, fewer, and / or alternate actions, including various actions described in the present disclosure.

[0016] In still yet other embodiments, a second model trained using second model training data to generate second model output data indicating a seizure of a patient in long-term brain state change (LTBSC) data. The second model may include (i) the second model stored on one or more memories and configured to be executed by one or more processors, the second model operable to: (a) receive the LTBSC data indicating at least one LTBSC of the patient; (b) determine one or more feature values of the LTBSC data; and (c) based on the one or more feature values, generate the second model output data. The model may be configured to perform additional, fewer, and / or alternate actions, including various actions described in the present disclosure.

[0017] In other embodiments, a third model trained using third model training data to generate third model output data. The third model may include (i) third model, stored on one or morememories and configured to be executed by one or more processors, the third model operable to:(a) receive long-term brain state change (LTBSC) data indicating at least one LTBSC of a patient;(b) determine one or more feature values of the LTBSC data; and (c) based on the one or more feature values, generate the third model output data, wherein: (i) the third model output data includes a revision of an indication of a seizure in LTEEG data of the LTBSC data; (ii) the indication of the seizure is not based upon the at least one LTBSC of the LTBSC data; and (iii) the revision to the indication of the seizure is based upon the at least one LTBSC of the LTBSC data. The model may be configured to perform additional, fewer, and / or alternate actions, including various actions described in the present disclosure.

[0018] In yet other embodiments, a fourth model trained using fourth model training data to generate fourth model output data indicating a prediction of a seizure of a patient. The fourth model may include (i) the fourth model, stored on one or more memories and configured to be executed by one or more processors, the fourth model operable to: (a) receive long-term brain state change (LTBSC) data indicating at least one LTBSC of the patient; (b) determine one or more feature values of the LTBSC data; and (c) based on the one or more feature values, generate the fourth model output data. The model may be configured to perform additional, fewer, and / or alternate actions, including various actions described in the present disclosure.

[0019] In still yet other embodiments, a fifth model trained using fifth model training data to generate fifth model output data indicating at least one date to implement a medication titration of a patient. The fifth model may include (i) the fifth model, stored on one or more memories and configured to be executed by one or more processors, the fifth model operable to: (a) receive longterm brain state change (LTBSC) data indicating at least one LTBSC of the patient; (b) determine one or more feature values of the LTBSC data; and (c) based on the one or more feature values, generate the fifth model output data. The model may be configured to perform additional, fewer, and / or alternate actions, including various actions described in the present disclosure.

[0020] In other embodiments, a sixth model trained using sixth model training data to generate sixth model output data indicating an association between at least one environmental characteristic of a patient and the at least one long-term brain state change (LTBSC) of the patient. The sixth model may include (i) the sixth model, stored on one or more memories and configured to be executed by one or more processors, the sixth model operable to: (a) obtain LTBSC data indicating at least one LTBSC of the patient, and environmental data indicating the at least one environmental characteristic of the patient; (b) determine one or more feature values of the LTBSC data and the environmental data; and (c) based on the one or more feature values, generate the sixth modeloutput data. The model may be configured to perform additional, fewer, and / or alternate actions, including various actions described in the present disclosure.

[0021] In yet other embodiments, a non-transitory computer-readable medium storing processorexecutable instructions that, when executed by one or more processors, cause the one or more processors to at least: (i) obtain long-term electroencephalogram (LTEEG) data comprising at least one LTEEG signal associated with a brain of the patient and gathered via an implant device of the patient; (ii) provide the LTEEG data to a first model trained using first model training data to generate long-term brain state change (LTBSC) data indicating at least one LTBSC of the patient; (iii) in response to generating the LTBSC data, one or more of: (a) retrain one or more machine learning models using the LTBSC data; (b) provide the LTBSC data as an input to the one or more machine learning models; or (c) generate a medical insight of the patient; and provide the medical insight to a user device. The instructions may direct additional, less, or alternate functionality, including that discussed elsewhere herein.

[0022] In accordance with the above, the disclosed systems and methods address the unmet need for a technology that determines a LTBSC of the patient, when developing one or more models which operate on the LTEEG data of the patient to generate the LTBSC data indicating one or more long-term brain state changes (LTBSCs) of the patient. The LTBSCs may indicate, alone, or in conjunction with long-term biometrics (LTBs) of the patient associated with the LTBSCs, longterm trajectories of neurological health which may:

[0023] establish longer-term state / behavior changes in the brain that are more holistic than episodic event tracking (such as seizures in the LTEEG);

[0024] indicate cycles in brain states which can challenge the establishment of causal connections with variations, such as alterations in drug dosage, and indicate relationships with longer-term parameters of the patient, such as lifestyle, weather or seasons;

[0025] reveal risks factors for patients which may not be otherwise revealed by episodic information (e.g., sudden unexplained death by epilepsy); and / or

[0026] reveal new EEG patterns not previously observed by either a human or computer, e.g., extracting new patterns in the LTEEG based upon the detection of transitions to new brain states (along with associated biometrics);

[0027] indicate treatment pathways to target a desired brain state;

[0028] indicate permanent changes in brain function using visualizations and / or explanations based upon LTBSC data; and / or

[0029] indicate similarities (pathology, demographics, drug treatment) with other patients having similar brain state changes, to determine successful treatments methods.

[0030] The LTBSC data, which may generally include the associated LTEEG data and / or longterm biometric (LTB) data of the patient, may identify the long-term changes in the brain’s state and / or biometrics, and may be utilized to train, re-train, and / or fine-tune one or more models to operate on such data when generating an output, such as models otherwise trained to detect or predict health events (e.g., seizures) based on EEG data alone, among others.

[0031] Additional, alternate and / or fewer actions, steps, features and / or functionality may be included in an aspect and / or embodiments, including those described elsewhere herein.BRIEF DESCRIPTION OF THE FIGURES

[0032] Fig. 1 is a block diagram of an example computing environment to determine a LTBSC of the patient, according to the described embodiments.

[0033] Fig. 2 is a block diagram of an example mobile device of the example computer environment of Fig. 1 .

[0034] Fig. 3A is a block diagram of an example implant device system according to the described embodiments.

[0035] Fig. 3B is a block diagram of an example sensor array of the example implant device system of Fig. 3A.

[0036] Fig. 3C depicts an example sensor array including a plurality of electrodes and a local processing device, including an exploded top view of the example electrode device.

[0037] Fig. 3D shows a cross-sectional view of portions of the electrode device of Fig. 3C.

[0038] Fig. 3E is a block diagram depicting in greater detail of an example implant device system implementing static models.

[0039] Fig. 4 is a combined block and logic diagram in which exemplary computer-implemented methods and systems for training a machine learning model are implemented, according to the described embodiments.

[0040] Fig. 5 is a block diagram depicting in greater detail an example implant device system implementing machine learning models.

[0041] Fig. 6 is a block diagrams depicting an embodiment in which evaluative functions take place on an external device, rather than on a local processor device.

[0042] Fig. 7A is a flow chart depicting a method for training a first model to determine a LTBSC of a patient.

[0043] Fig. 7B is a flow chart depicting a method for training a second model to generate second model output data indicating a seizure of the patient.

[0044] Fig. 7C is a flow chart depicting a method for training a third model to generate third model output data, the third model output data including a revision of an indication of a seizure in the LTEEG data of the LTBSC data.

[0045] Fig. 7D is a flow chart depicting a method for training a fourth model to generate fourth model output data indicating a prediction of a seizure of the patient.

[0046] Fig. 7E is a flow chart depicting a method for training a fifth model to generate fifth model output data indicating at least one date to implement a medication titration of the patient.

[0047] Fig. 7F is a flow chart depicting a method for training a sixth model to generate sixth model output data indicating an association between at least one environmental characteristic of the patient and the at least one LTBSC of the patient.

[0048] Fig. 8 is a flow chart depicting a method to determine a LTBSC of the patient.

[0049] Fig. 9 depicts an embodiment of the implant device system in which the sensor array and the processor device are integrated into a single unit.

[0050] Fig. 10A illustrates an example communication scheme between the sensor array and the processor device.

[0051] Fig. 10B illustrates an alternate example of a communication scheme between the sensor array and the processor device.

[0052] Fig. 1 1 A illustrates an example communication scheme between the processor device and external equipment.

[0053] Fig. 1 1 B illustrates an alternate example of a communication scheme between the processor device and external equipment.

[0054] Fig. 1 1 C illustrates yet another example of a communication scheme between the processor device and external equipment.DETAILED DESCRIPTIONOVERVIEW

[0055] The systems and methods disclosed herein generally relate to systems and methods for determining a long-term brain state change of a patient. The LTBSC data (indicating the LTBSCs), which again generally includes the associated LTEEG data and LTB data of the patient if available, provides the ability to derive trajectories of neurological health based on long-term brain activity and biometric trends of the patient. Moreover, information associated with LTBSCs and LTBs of the patient may be used to compliment and / or enhance determinations, insights, predictions and the like otherwise established from the LTEEG data alone, providing a more holistic assessment of the patient’s well-being.

[0056] The disclosed systems and methods may determine LTBSCs of the patient. As further described below, the LTEEG data may comprise at least one LTEEG signal associated with the patient’s brain and gathered via an implant device of the patient. The LTEEG data may be provided to a first model. The first model may be trained, using first model training data, to generate LTBSC data indicating at least one LTBSC of the patient. The first model training data may include, for historical patients (patients for which data have already been collected), historical LTEEG data, historical LTBSC data, and historical LTB data. The first model is trained to determine associations between an historical LTBSC, associated historical LTBs, and at least one historical LTEEG signal corresponding to the historical LTBSC.

[0057] In response to generating the LTBSC data, the systems and methods may generate a medical insight of the patient, and provide the medical insight to a user device (e.g., a user device of the patient, a healthcare provider of the patient, etc.). In at least one aspect, the medical insight may be the (detection of the) LTBSCs, as the LTBSC may be associated with epilepsy, a psychiatric disorder, a stroke, Alzheimer's disease, a sleep disorder, depression, medication efficacy, a medication side effect, or a risk of sudden unexplained death by epilepsy (SUDEP). In at least one aspect, the medical insight may be an association between the LTBSC and the long-term biometrics, e.g., facial long-term biometrics which indicate the LTBSC.

[0058] In response to generating the LTBSC data, the systems and methods may (i) retrain one or more machine-learning models using the LTBSC data and / or (ii) provide the LTBSC data as an input to the one or more machine learning models. In one aspect, the one or more machine learning models may include a second model trained, using second model training data, to generate second model output data indicating a seizure of the patient in the LTEEG of the LTBSC data, based upon receiving the LTBSC data. For example, the second model output data may include a classification of the seizure (e.g., type of seizure, severity, etc.) and / or a confidence metric associated with the indication of the seizure (e.g., a confidence score indicating the level of certainty or reliability associated with the detection and / or classification of the seizure). In such an aspect, retraining the second model using the LTBSC data may include storing the retrained second model in a memory to generate subsequent second model output data using the retrained second model. In such an aspect, providing the LTBSC data as the input to the second model may include, in response to generating the second model output data, providing the second model output data to the user device.

[0059] In one embodiment, the one or more machine learning models may include a third model trained, using third model training data, to generate third model output data, based upon receiving the LTBSC data. The third model output data may include a revision of an indication of a seizure inthe LTEEG data of the LTBSC data, wherein the indication of the seizure is not based upon the least one LTBSC of the LTBSC data (e.g., when the indication generated by a model trained to operate only on EEG data of the patient), and the revision to the indication of the seizure is based upon at least one LTBSC of the LTBSC data. For example, the revision may include a reclassification of the seizure and / or a change in a confidence metric associated with the indication of the seizure, such as revising the classification and confidence metric of the seizure indication described with respect to the second model. In such an aspect, retraining the third model using the LTBSC data may include storing the retrained third model in the memory to generate subsequent third model output data using the retrained third model. In such an aspect, providing the LTBSC data as the input to the third model may include, in response to generating the third model output data, providing the third model output data to the user device (e.g., of the patient, a caregiver of the patient, etc.).

[0060] The one or more machine learning models may include a fourth model trained, using fourth model training data, to generate fourth model output data indicating a prediction of a seizure of the patient, based upon receiving the LTBSC data. For example, the fourth model output data may include at least one date of the prediction of the seizure, and / or a confidence metric associated with the prediction. In such an aspect, retraining the fourth model using the LTBSC data may include storing the retrained fourth model in the memory to generate subsequent fourth model output data using the retrained fourth model. In such an aspect, providing the LTBSC data as the input to the fourth model may include, in response to generating the fourth model output data, providing the fourth model output data to the user device.

[0061] In another embodiment, the one or more machine learning models may include a fifth model trained, using fifth model training data, to generate fifth model output data indicating at least one date to implement a medication titration of the patient, based upon receiving the LTBSC data. In such an aspect, retraining the fifth model using the LTBSC data may include storing the retrained fifth model in the memory to generate subsequent fifth model output data using the retrained fifth model. In such an aspect, providing the LTBSC data as the input to the fifth model may include, in response to generating the fifth model output data, providing the fifth model output data to the user device.

[0062] In still yet another embodiment, the one or more machine learning models may include a sixth model trained, using sixth model training data, to generate sixth model output data indicating an association between at least one environmental characteristic of the patient and at least one LTBSC of the patient, based upon receiving environmental data indicating the at least one environmental characteristic of the patient and the LTBSC data. In one example, the environmentalcharacteristic may be associated with a new medication taken by the patient, and / or a new titration of the medication taken by the patient. In one example, sixth model output data indicates a prediction of a future LTBSC in response to the at least one environmental characteristic, and / or a treatment pathway to a target brain state. In such an aspect, retraining the sixth model using the LTBSC data may include storing the retrained sixth model in the memory to generate subsequent sixth model output data using the retrained sixth model. In such an aspect, providing the LTBSC data as the input to the sixth model may include obtaining the environmental data associated with the LTBSC data, providing the environmental data and the LTBSC data to the sixth model to generate the sixth model output data, and, in response to generating the sixth model output data, providing the sixth model output data to the user device.

[0063] Although epilepsy is disclosed as an example neurological disorder above as well as in other portions of the disclosure, the disclosed embodiments and techniques may be applicable to other neurological disorders and / or other fields of use, such as drug development (e.g., drug efficacy and / or side effects), psychiatry, stroke, Alzheimer's disease, sleep disorders, and depression, to name a few.

[0064] Various aspects of the system and method are described throughout this specification. Unless otherwise specified, aspects of any embodiment that are compatible with another embodiment described herein are considered as contemplated and disclosed embodiments herein, especially as described in the list of aspects at the end of this specification. For example, a feature of a particular embodiment described herein, if that feature would be recognized by a person of ordinary skill in the art to be compatible with the features of a second embodiment described herein, should be considered as a possible feature of the second embodiment. Further, embodiments describing features as optional should be considered as disclosing said embodiments both with and without the optional features, and with various optional features in any combination that would be recognized by a person of ordinary skill in the art as being compatible.

[0065] Throughout the present disclosure, embodiments are described in which various elements are optional - present in some, but not all, embodiments of the system. Where such elements are depicted in the accompanying figures and, specifically, in figures depicting block diagrams, the optional elements are generally depicted in dotted lines to denote their optional nature.EXAMPLE COMPUTER ENVIRONMENT

[0066] Fig. 1 depicts an example computing environment 100 associated with determining aLTBSC of a patient. Although Fig. 1 depicts certain entities, components, equipment, and devices,it should be appreciated that additional or alternate entities, components, equipment, and devices are envisioned.

[0067] As illustrated in Fig. 1 , the computing environment 100 may include in one aspect, one or more servers 105 which may perform the functionalities, such determining the LTBSC of the patient. The server 105 may be part of a cloud network or may otherwise communicate with other hardware or software components within one or more cloud computing environments to send, retrieve, or otherwise analyze data or information described herein. For example, in certain aspects of the present techniques, the computing environment 100 may comprise an on-premise computing environment, a multi-cloud computing environment, a public cloud computing environment, a private cloud computing environment, and / or a hybrid cloud computing environment. For example, an entity (e.g., a business) may host one or more services in a public cloud computing environment (e.g., Alibaba Cloud, Amazon Web Services (AWS), Google Cloud, IBM Cloud, Microsoft Azure, etc.). The public cloud computing environment may be a traditional off-premise cloud (i.e. , not physically hosted at a location owned / controlled by the business). Alternatively, or in addition, aspects of the public cloud may be hosted on-premise at a location owned / controlled by the business. The public cloud may be partitioned using visualization and multi-tenancy techniques and may include one or more infrastructure-as-a-service (laaS) and / or platform-as-a-service (PaaS) services.

[0068] A network 1 10 may comprise any suitable network or networks, including a local area network (LAN), a wide area network (WAN), the Internet, or a combination thereof. For example, the network 110 may include a wireless cellular service (e.g., 4G, 5G, 6G, etc.). Generally, the network 1 10 enables bidirectional communication between the servers 105, a user device 1 15 and / or an implant device 150. In one aspect, the network 1 10 may comprise a cellular base station, such as cell tower(s), communicating to the one or more components of the computing environment 100 via wired / wireless communications based upon any one or more of various mobile phone standards, including NMT, GSM, CDMA, UMTS, LTE, 5G, 6G, or the like. Additionally or alternatively, the network 1 10 may comprise one or more routers, wireless switches, or other such wireless connection points communicating to the components of the computing environment 100 via wireless communications based upon any one or more of various wireless standards, including by non-limiting example, IEEE 802.1 1 a / ac / ax / b / c / g / n (Wi-Fi), Bluetooth, and / or the like.

[0069] Communication circuitry 122 may communicate over the network 1 10 via any suitable wired and / or wireless connection, e.g., using any suitable network interface controller(s) of the communication circuitry 122. The communication circuitry 122 may include one or more transceivers (e.g., WWAN, WLAN, and / or WPAN transceivers) functioning in accordance with IEEEstandards, 3GPP standards, and / or other standards, and that may be used in receipt and transmission of data via external / network ports connected to computer network 1 10.

[0070] The server 105 may include at least one processor 120. The processor 120 may include one or more suitable processors (e.g., central processing units (CPUs) and / or graphics processing units (GPUs)). The processor 120 may be connected to a memory 125 via a computer bus (not depicted) responsible for transmitting electronic data, data packets, or otherwise electronic signals to and from the processor 120 and the memory 125 in order to implement or perform the machine- readable instructions, methods, processes, elements, or limitations, as illustrated, depicted, or described for the various flowcharts, illustrations, diagrams, figures, and / or other disclosure herein. The processor 120 may interface with the memory 125 via a computer bus to execute an operating system and / or computing instructions contained therein, and / or to access other services / aspects. For example, the processor 120 may interface with the memory 125 via the computer bus to create, read, update, delete, or otherwise access or interact with the data stored in the memory 125 and / or a database 126.

[0071] The memory 125 may include one or more forms of volatile and / or non-volatile, fixed and / or removable memory, such as read-only memory (ROM), electronic programmable read-only memory (EPROM), random access memory (RAM), erasable electronic programmable read-only memory (EEPROM), and / or other hard drives, flash memory, MicroSD cards, and others. The memory 125 may store an operating system (e.g., Microsoft Windows, Linux, UNIX, etc.) capable of facilitating the functionalities, applications, methods, or other software as described herein.

[0072] In general, a computer program or computer based product, application, or code (e.g., machine learning (ML) models, or other computing instructions described herein) may be stored on a computer usable storage medium, or tangible, non-transitory computer-readable medium (e.g., standard random access memory (RAM), an optical disc, a universal serial bus (USB) drive, or the like) having such computer-readable program code or computer instructions embodied therein, wherein the computer-readable program code or computer instructions may be installed on or otherwise adapted to be executed by the processor(s) 120 (e.g., working in connection with the respective operating system in memory 125) to facilitate, implement, or perform the machine readable instructions, methods, processes, elements or limitations, as illustrated, depicted, or described for the various flowcharts, illustrations, diagrams, figures, and / or other disclosure herein. In this regard, the program code may be implemented in any desired program language, and may be implemented as machine code, assembly code, byte code, interpretable source code or the like (e.g., via Golang, Python, C, C++, C#, Objective-C, Java, Scala, ActionScript, JavaScript, HTML, CSS, XML, etc.).

[0073] The memory 125 may include at least one database 126. The database 126 may be a relational database, such as Oracle, DB2, MySQL, a NoSQL based database, such as MongoDB, or another suitable database. The database 126 may store data, such as the LTBSC data of multiple patients, data that is used to create, train, operate, and / or fine-tune one or more models, model output data, etc. This may include model training data, such as testing, validation, feedback, and / or other training data. The database 126 may include any other suitable data.

[0074] The memory 125 may include one or more routines or models 128. A model 128, routine, or other element stored in memory may be referred to as receiving an input, producing or storing an output, or executing, the routine, model 128, or other element. The model 128 is, in fact, executing as instructions on the processor 120. Further, those of skill in the art will appreciate that the model 128, routine, or other instructions may be stored in the memory 125 as executable instructions, which instructions the processor 120 may retrieve from the memory 125 and execute. Further, the processor 120 should be understood to retrieve from the memory 125 any data necessary to perform the executed instructions (e.g., data required as an input to the routine or model 128), and to store in the memory 125 the intermediate results and / or output of any executed instructions. The model 128 may include models based on a static algorithm (e.g., a static model), or ML models, such as one or more ML models 130, as further described below.

[0075] The memory 125 may store a plurality of computing modules 135, implemented as respective sets of computer-executable instructions (e.g., one or more source code libraries, trained ML models 130 such as neural networks, convolutional neural networks, etc.), as described herein.

[0076] In one aspect, the computing modules 135 may include an ML module 140. The ML module 140 may include ML training module (MLTM) 142 and / or ML operation module (MLOM) 144. In some embodiments, at least one of a plurality of ML methods and algorithms may be applied by the ML module 140, which may include, but are not limited to: linear or logistic regression, instance-based algorithms, regularization algorithms, decision trees, Bayesian networks, cluster analysis, association rule learning, artificial neural networks, deep learning, combined learning, reinforced learning, dimensionality reduction, and support vector machines. In various embodiments, the implemented ML methods and algorithms are directed toward at least one of a plurality of categorizations of ML, such as supervised learning, unsupervised learning, and reinforcement learning. In one aspect, the ML-based algorithms may be included as a library or package executed on server(s) 105. For example, libraries may include the TensorFlow-based library, the PyTorch library, and / or the scikit-learn Python library.

[0077] In one embodiment, the ML module 140 employs supervised learning, which involves identifying patterns in existing data to make predictions about subsequently received data.Specifically, the ML module is “trained” (e.g., via MLTM 142) using training data, which includes exemplary inputs and associated exemplary outputs. Based upon the training data, the ML module 140 may generate a predictive function which maps outputs to inputs and may utilize the predictive function to generate ML outputs based upon data inputs. The exemplary inputs and exemplary outputs of the training data may include any of the data inputs or ML outputs described above. In the exemplary embodiments, a processing element may be trained by providing it with a large sample of data with known characteristics or features.

[0078] In another embodiment, the ML module 140 may employ unsupervised learning, which involves finding meaningful relationships in unorganized data. Unlike supervised learning, unsupervised learning does not involve user-initiated training based upon exemplary inputs with associated outputs. Rather, in unsupervised learning, the ML module 140 may organize unlabeled data according to a relationship determined by at least one ML method / algorithm employed by the ML module 140. Unorganized data may include any combination of data inputs and / or ML outputs as described above.

[0079] In yet another embodiment, the ML module 140 may employ reinforcement learning, which involves optimizing outputs based upon feedback from a reward signal. Specifically, the ML module 140 may receive a user-defined reward signal definition, receive a data input, utilize a decision-making model to generate the ML output based upon the data input, receive a reward signal based upon the reward signal definition and the ML output, and alter the decision-making model so as to receive a stronger reward signal for subsequently generated ML outputs. Other types of ML may also be employed, including deep or combined learning techniques.

[0080] The MLTM 142 may receive labeled data at an input layer of a model having a networked layer architecture (e.g., an artificial neural network, a convolutional neural network, etc.) for training the ML models 130. The received data may be propagated through one or more connected deep layers of the ML model 130 to establish weights of one or more nodes, or neurons, of the respective layers. Initially, the weights may be initialized to random values, and one or more suitable activation functions may be chosen for the training process. The present techniques may include training a respective output layer of the ML models 130. The output layer may be trained to output a prediction, for example.

[0081] The MLOM 144 may comprise a set of computer-executable instructions implementing ML loading, configuration, initialization and / or operation functionality. The MLOM 144 may include instructions for storing trained ML models 130 (e.g., as ML models 130 in memory 125). As described, once trained, the trained ML models 130 may be operated in inference mode,whereupon when provided with de novo input that the model has not previously been provided, the model may output one or more predictions, classifications, etc., as described herein.

[0082] In operation, the ML model training module 142 may access the database 126 or any other data source for training data suitable to generate the ML models 130. The training data may be sample data with assigned relevant and comprehensive labels (classes or tags) used to fit the parameters (weights) of the ML model 130 with the goal of training it by example. In one aspect, once an appropriate ML model 130 is trained and validated to provide accurate predictions and / or responses, the trained ML model 130 may be loaded into MLOM 144 at runtime to process input data and generate output data.

[0083] While various embodiments, examples, and / or aspects disclosed herein may include training and generating the ML models 130 for the server 105 to load at runtime, it is also contemplated that one or more appropriately trained ML models 130 may already exist (e.g., in memory 125) such that the server 105 may load an existing trained ML model 130 at runtime. It is further contemplated that the server 105 may retrain, fine-tune, update and / or otherwise alter an existing ML model 130 before and / or after loading the model at runtime.

[0084] In one aspect, the computing modules 135 may include an input / output (I / O) module 148, comprising a set of computer-executable instructions implementing communication functions. The I / O module 148 may include a communication component configured to communicate (e.g., send and receive) data via one or more external / network port(s) to one or more networks or local terminals, such as the computer network 1 10 and / or the user device 1 15 (for rendering or visualizing) described herein. In one aspect, the servers 105 may include a client-server platform technology such as ASP.NET, Java J2EE, Ruby on Rails, Node.js, a web service or online API, responsive for receiving and responding to electronic requests.

[0085] I / O module 148 may further include or implement an operator interface configured to present information to an administrator or operator and / or receive inputs from the administrator and / or operator. An operator interface may provide a display screen. The I / O module 148 may facilitate I / O components (e.g., ports, capacitive or resistive touch sensitive input panels, keys, buttons, lights, LEDs), which may be directly accessible via, or attached to, servers 105 or may be indirectly accessible via or attached to the user device 1 15. According to one aspect, an administrator or operator may access the servers 105 via the user device 115 to review information, make changes, input training data, initiate training via the MLTM 142, and / or perform other functions (e.g., operation of the one or more trained models 130 via the MLOM 144).

[0086] The one or more servers 105 may be in communication with at least one user device 1 15, e.g., the user device associated with a patient, doctor, clinician, caregiver, etc. The user device 1 15may comprise one or more computers, which may comprise multiple, redundant, or replicated client computers accessed by one or more users. The user device 1 15 may access services or other components of the computing environment 100 via the network 110, as further described herein. The user device 115 may be any suitable device and include one or more mobile devices, wearables, smart watches, smart contact lenses, smart glasses, AR glasses / headsets, VR glasses / headsets, mixed or extended reality glasses / headsets, displays, display screens, visuals, and / or other electronic or electrical components. The user device 1 15 may include a memory and a processor for, respectively, storing and executing one or more modules, programs, routines, etc. The memory may include one or more suitable storage media such as a magnetic storage device, a solid-state drive, random access memory (RAM), etc. The user device 1 15 may include one or more sensors, such as a microphone, a camera, an infra-red sensor, an accelerometer, etc., one or more of which may be used to generate the facial biometric data. The user device 1 15 may include communication circuitry to access services or other components of the computing environment 100, e.g., via the network 1 10.

[0087] The one or more servers 105 may also be in communication with at least one implant device 150. The implant device 150 may use communication circuitry to access services or other components of the computing environment 100, e.g., via the network 1 10. The implant device 150 may include a memory and a processor for, respectively, storing and executing one or more modules, programs, routines, etc. The memory may include one or more suitable storage media such as a magnetic storage device, a solid-state drive, random access memory (RAM), etc. The implant device may include a sensor array. The sensor array generally provides data in the form of one or more electrical signals, such as LTEEG data in the form of EEG signals, to a processor device of the implant device 150, which receives the signals. The electrical signals may be processed (e.g., via a processor of the implant device 150, or other suitable processor) and / or provided to one or more other devices (e.g., the server 105, the user device 1 15, a processor device, etc.), e.g., to detect LTBSCs indicated by the LTEEG data, detect, classify and / or predict a health event (e.g., seizure) indicated by the EEG data, and / or any other suitable purpose.

[0088] The implant device 150 may gather EEG data, such as LTEEG data, from the brain activity of a patient in whom the sensor array of the implant device 150 is implanted. The implant device 150 via its communication circuitry may transmit (e.g., via the network 1 10) the EEG data to one or more components and / or devices of the computing environment, such as the server 105, the user device 1 15, and / or any other suitable device or component. The server 105, the user device 115 and / or other suitable device may provide the EEG data as an input to one or more models 128, 130, such as an input to the first model, to generate as an output the LTBSC data. The implantdevice 150 may gather the LTEEG data using implanted electrodes over long periods of time (e.g., over weeks, months) in a substantially continuous manner, i.e., largely interrupted and / or having minor breaks in the EEG recording. The LTEEG data may be recorded using a single channel or multiple channels. The LTEEG data may include different frequency bands which may be analyzed (e.g., by the first model) to understand brain activity (e.g., LTBSCs) as it relates to a neurological disorder, such as epilepsy. For example, the frequency bands may represent different patterns of neural oscillations which can provide valuable information for diagnosing and monitoring epilepsy, such as:

[0089] Seizure-Related Low-Frequency Oscillations (1 - 2 Hz): Low-frequency oscillations may be observed during or after seizures, and may be associated with postictal states;

[0090] Delta (0.5 - 4 Hz): Delta waves can be associated with deep sleep and may be observed during slow-wave sleep stages. Abnormal delta activity may be indicative of certain epileptic conditions;

[0091] Theta (4 - 8 Hz): Theta waves may be typically observed in drowsy or relaxed states. Excessive theta activity may be seen in temporal lobe epilepsy;

[0092] Alpha (8 - 13 Hz): Alpha waves may be associated with a relaxed, awake state of an individual having closed eyes. Abnormal alpha activity may indicate focal seizures;

[0093] Beta (13 - 30 Hz): Beta waves may be associated with alertness and cognitive activity. Excessive beta activity may be recorded during epileptic events, especially in the context of complex partial seizures;

[0094] Gamma (30 - 100+ Hz): Gamma waves may be associated with higher cognitive functions, sensory processing, and binding of sensory information, which may suggest a role in epilepsy;

[0095] Mu (8 - 13 Hz): Mu rhythm is a subset of the alpha range, may be associated with motor cortex activity, and may be used to study motor-related seizures and epileptic events involving the motor cortex; and

[0096] Ripples (80 - 250 Hz) and Fast Ripples (250 - 500 Hz): Ripples are high-frequency oscillations, may be associated with the hippocampus, and may be observed during seizures in temporal lobe epilepsy. Fast ripples are very high-frequency oscillations, may be linked to the seizure onset zone in the brain, and may be considered highly specific to epileptic regions.

[0097] In at least one embodiment, the computing environment 100 may be used to determine LTBSCs of a patient. The implant device 150 may gather the LTEEG data of the patient, for example recording the LTEEG over the course of weeks or months. The implant device 150 may provide the LTEEG data to the server 105 via the network 1 10, to be operated on by a first model,such as the models 128, 130. Although in the embodiment being described the server 105 executes the first model, in alternate embodiments the implant device 150, the user device 115, and / or other suitable device and / or processor of the computing environment 100 may execute the first model. In such alternate embodiments, the implant device 150 may provide the LTEEG data to the device / processor executing the first model to operate on the LTEEG data, and / or store the LTEEG data in one or more memories (e.g., the memory 125, the database 126, etc.), which may be accessed by the device / processor executing the first model to operate on the LTEEG data.

[0098] The server 105 may receive the LTEEG data from the implant device 150. The server 105 may load the first model from memory (e.g., from the models 128, 130 in the memory 125) via the ML module 140. The first model may be trained (e.g., via the MLTM 142) using first model training data (e.g., training data stored in the database 126) which includes historical LTEEG data of historical patients, historical LTBSC data of historical patients corresponding to the historical LTEEG data, and / or other suitable training data. The first model may be trained to determine, for the historical patients, associations between an historical LTBSC and at least one historical LTEEG signal corresponding to the historical LTBSC.

[0099] The server 105 may provide the LTEEG data to the first model, via the ML module 140, to generate the LTBSC data. Generating the LTBSC data may include one or more of: (i) analyzing at least one frequency band of the at least one LTEEG signal; (ii) identifying at least one long-duration EEG segment in the at least one LTEEG signal based upon one or more of a root means square, a standard deviation, or a power spectrum density, of the at least one LTEEG signal; (iii) determining one or more feature values (e.g., feature values associated with characteristics of an EEG signal, such as EEG spikes and / or EEG sharps) of the at least one LTEEG segment by applying a principal components analysis algorithm or a support vector machine to the LTEEG data; (iv) determining at least one similar characteristic of the LTEEG data and / or the one or more feature values, across long-duration time points; (v) classifying LTEEG segments having similar characteristics; and / or (vi) determining an LTBSC boundary based upon detecting an edge of a phase transition of the at least one LTEEG signal. The LTBSC data may indicate one or more LTBSCs of the patient, e.g., LTBSCs associated with epilepsy, a psychiatric disorder, a stroke, Alzheimer's disease, a sleep disorder, depression, medication efficacy, and / or a medication side effect. The LTBSC data may include at least a portion of the LTEEG data corresponding to the at least one LTBSC, a classification and / or confidence metric associated with the at least one LTBSC, and / or any other suitable data.

[0100] In at least one aspect, the server 105 may obtain LTB data indicating LTBs of the patient corresponding to the at least one LTBSC, that is LTB data captured during at least a portion of thetime the implant device is recording the LTEEG data. In one example, the patient may capture biometric data on the user device 1 15 (e.g., facial images using a camera of the user device 1 15, audio of the patient’s voice using a microphone of the user device 115, etc.), while the implant device 150 is generating LTEEG data. However, the LTB data may be generated in any suitable manner and / or by any suitable component of the computing environment 100, such as via the microphone of the implant device 150, etc. The LTB data may also be captured when the patient is not experiencing the LTBSC or a seizure, e.g., to determine the baseline biometrics of the patient for comparison to biometrics when they may be experiencing the LTBSC or seizure. The patient’s user device 1 15 may provide the LTB data to one or more models trained and / or configured to operate on the LTB data, such as one or more models executed by the server 105, the user device 115, the implant device 150, and / or any other suitable component. The one or more models, such as the first model, may determine one or more feature values of the LTB data, such as biometrics of the face and / or eyes. In one example, the user device 1 15 provides the LTB data to the server 105 via the network 1 10 for processing by the first model. In such an aspect, the server 105 may provide the LTB data to the first model in a similar manner as that described with respect to proving the LTBSC data to the first model. The first model training data may include historical LTB data of historical patients corresponding to the historical LTBSC data of the historical patients, such that the first model is trained to determine associations between the historical LTBSC and historical longterm biometrics corresponding to the historical LTBSC, and to generate the LTBSC data further based upon the LTB data. In such an aspect, the LTBSC data may include at least a portion of the LTB data, such as the portion of the LTB data corresponding to the LTBSC data (e.g., LTB data indicating patient biometrics before, during and / or after one or more LTBSCs of the patient). As described earlier, the LTB data may improve the operation of one or more models by providing additional context via the LTBs of the state of the patient proximate the time in which the LTEEG is being recorded, one or more LTBSCs are occurring, one or more health events (e.g., seizures) are occurring, etc.

[0101] In response to generating the LTBSC data, the server 105 may generate a medical insight of the patient based upon the LTBSC data, and provide the medical insight to one or more user devices 1 15 (e.g., via the network 1 10), such as the user device 115 of the patient, their medical provider, their caregiver, etc. The medical insight may be the LTBSC data and / or the indication of the LTBSCs of the patient (e.g., how many LTBSCs, when they occurred, etc.), as the LTBSCs may be indicative of one or more neurological conditions, as previously described. The medical insight may be a risk of SUDEP, an association between the at least one LTBSC and the LTBs of the patient, and / or any other suitable medical insight.

[0102] In response to generating the LTBSC data, the server 105 may retrain one or more machine learning models using the LTBSC data, and / or provide the LTBSC data as an input to the one or more machine learning models (e.g., to generate an output therefrom). Retraining the one or more machine learning models may include using the LTBSC data to fine-tune the one or more machine learning models. For example, a base model may be trained to operate on EEG data, such as LTEEG data, as an input to generate as an output, such as a detection of a health event (e.g., seizure) in the EEG data, a prediction of a health event (e.g., a seizure), and / or any other suitable output. As previously described, the LTBSC data (which generally includes the associated LTEEG data, and LTB data if available) may provide additional insight into the health of the patient (e.g., via the associated LTBSCs and LTBs) when analyzed in conjunction with the LTEEG data. Thus, the base model may be fine-tuned to accept the LTBSC data as an input, which includes LTEEG data, an indication of one or more LTBSCs, potentially LTB data, such that the fine-tuned base model provides more accurate, improved and / or altered outputs as a result of the fine-tuning. For example, the base seizure detection-prediction model may receive original LTEEG data and determine a patient experienced three seizures, and the next seizure is predicted in three days. Once the base seizure detection-prediction model is fine-tuned and provided the LTBSC data (which includes the original LTEEG data), the fine-tuned seizure detection-prediction model may generate a revised determination that the patient experienced four seizures, for example when determining the LTEEG data had patterns indicating a fourth seizure which correlated to a LTBSC of the patient, which the base model would otherwise have not detected if the LTBSC data was not provided. Moreover, the fine-tuned seizure detection-prediction model may revise the next seizure prediction to be in five days, based upon patterns in the LTBSCs and LTBs of the patient as indicated by the LTBSC data.

[0103] Any of the data input and / or output by the first model or one or more machine learning models may subsequently be used as training data. Using the example just-described, the finetuned seizure detection-prediction model may be trained using LTBSC data to more accurate forecast a seizure, and also to facilitate faster convergence of the model during training (e.g., less months of EEG data are required to train from).

[0104] In one aspect, the one or more models may include a second model (such as models 128, 130) for detecting a seizure. The second model may be trained using second model training data (e.g., trained by the ML model 140 using training data stored the database 126) to receive the LTBSC data as an input to generate the second model output data, indicating a seizure of the patient, as the output. As may by now be understood, the LTBSC data may generally include the associated LTB data and / or LTEEG data, such that when the second model (or other model)receives as an input the LTBSC data indicating the LTBSCs, the second model (or other model) also receives as the input the LTB data and / or LTEEG data. The second model training data may include historical LTBSC data for a plurality of historical patients, which similarly may include the historical LTEEG data associated with the historical LTBSC data, and also the associated historical LTB data, if available. The second model may be trained to determine, for the historical patients, associations between at least one historical LTBSC and an historical seizure. For example, the second model may be trained to detect LTBSCs and / or LTBs of a patient which correspond to, or are otherwise associated with, a seizure (e.g., a specific brain state change, facial drooping, etc.).

[0105] The second model may generate as an output second model output data indicating one or more seizures of the patient, such as seizure indicated by the LTEEG, and the LTBSCs and / or LTBs. The second model output data may include and / or indicate the detected seizures in the LTEEG associated with the seizure (e.g., an EEG snippet of the LTEEG of the LTBSC data), may correlate the seizures with the LTBSCs (e.g., indicating LTBSCs corresponding to the secures) and / or LTBs (e.g., images showing facial dropping of the patient proximate the seizure) of the patient, etc. The second model output data may include a classification of the seizure (e.g., type of seizure), a confidence metric associated with the seizure (e.g., the confidence that EEG snippet indicates the seizure), and / or any other suitable information. In at least one aspect, if the LTBSC data may indicate that a brain state change coincides with the detection of a seizure, and in response, the second model may operate with a higher sensitivity, for example there may be a higher likelihood that seizures may be indicated in the LTEEG associated with the brain state change which the second model may detect when operating with higher sensitivity. The server 105 may provide the second model output data to one or more user devices 1 15, e.g., via the network 110. For example, the second output data may be provided to the user device of the patient’s doctor, where the LTEEG of the output data is visually displayed on the user device screen of the with indications of the LTEEG patterns associated with the patient’s seizures and LTBSCs, as well as any biometric data (images, sounds) associated therewith. The server 105 may store the second model output data in the memory (e.g., the memory 125, the database 126), for example stored in a medical record of the patient, stored as training data, stored for review by a human reviewer for labeling the second model output data (e.g., for model training), and / or any other suitable purpose.

[0106] In one aspect, the one or more models may include a third model (e.g., the models 128, 130) for generating third model output data based upon receiving the LTBSC data. The second model output data includes a revision of an indication of a seizure in the LTEEG data of the LTBSC data, wherein the indication of the seizure is not based upon the LTBSCs and / or LTBs of theLTBSC data, however the revision to the indication of the seizure is based upon the LTBSCs and / or the LTBs. The third model 413 may be trained using third model training data (e.g., trained by the ML model 140 using training data stored the database 126). The third model training data may include historical LTBSC data of historical patients, such that the third model may be trained to determine, for the historical patients, associations between an historical seizure and at least one historical LTBSC and / or historical LTB. The revision of the seizure indication may include a reclassification of the seizure and / or a change in a confidence metric associated with the seizure as compared to the original, unrevised indication of the seizure. For example, the original seizure indication may indicate two tonic-clonic seizures, having a 25% confidence score and 30% confidence score respectively, indicating the confidence that such seizures actually occurred. The revisions may indicate the 60% and 67% confidence scores respectively, e.g., based upon associated LTBSCs and LTBs, the model may have a higher confidence of the seizures having occurred. As in the case of the second model, and subsequently described models, the server may provide the output of the model, here the second model output data, to one or more user devices 115 and / or store the model out on one or more memories.

[0107] In another aspect, the one or more models may include a fourth model (e.g., the models 128, 130) for predicting a seizure based upon receiving the LTBSC data. The fourth model may be trained using fourth model training data (e.g., trained by the ML model 140 using training data stored the database 126) to receive the LTBSC data as an input and generate fourth model output data indicating the prediction of a seizure of the patient, as the output. The fourth model training data may include, for a plurality of historical patients, historical LTBSC data such that the fourth model may be trained to determine, for the historical patients, associations between historical LTBSCs, historical LTBs, and the timing of historical seizures. For example, patterns and associations in LTBSCs and / or LTBs may provide an indication of the type and frequency of seizures experienced by the patient, such that the fourth model may predict when the seizure may next occur. The fourth model may generate as an output fourth model output data indicating a prediction of the seizure of the patient. The fourth model output data may include one or more dates during which one or more seizures may be most likely to occur, confidence metric(s) associated with the prediction, and / or any other suitable information associated with the prediction of one or more seizures. The server 105 may provide the fourth model output data to one or more user devices 115. For example, the fourth output data may be provided as an alert to the patient’s user device 1 15 indicating the next date they are expected to experience a seizure. The alert and / or fourth model output data may be provided to the patient user device 1 15, and / or any othersuitable user device, when the fourth model output data is generated, proximate the time when the next seizure is predicted, and / or at any other suitable time.

[0108] In still another aspect, the one or more models may include a fifth model (e.g., models 128, 130) to determine one or more times to attempt a medication titration, such as optimal days that may minimize the chance of side-effects from the drug titration. The fifth model may be trained, using fifth model training data (e.g., trained by the ML model 140 using training data stored the database 126), to receive the LTBSC data as an input, and to generate as the output the fifth model output data indicating at least one date to implement a medication titration of the patient. The fifth model training data may include, for a plurality of historical patients, historical LTBSC data such that the fifth model may be trained to determine, for the historical patients, associations between historical LTBSCs, historical LTBs, the timing of historical medication titrations, and associated historical side effects from the historical medication titrations. In one example, the associations in the training data may indicate that specific brain states are more conducive to minimizing the side effects of a titration increase of a medication used to treat seizures. The fifth model may generate as an output fifth model output data indicating at least one date to implement a medication titration of the patient.

[0109] In still yet another aspect, the one or more models may include a sixth model (e.g., models 128, 130) to determine an association between at least one environmental characteristic of the patient and at least one LTBSC of the patient. The sixth model may be trained, using sixth model training data, to generate sixth model output data indicating an association between at least one environmental characteristic of the patient and at least one LTBSC of the patient, based upon receiving environmental data indicating the at least one environmental characteristic of the patient and the LTBSC data. The environmental data may be generated using one or more devices and / or components of the computing environment 100, such as via sensors of user device 1 15, a user interface of the user device, etc., as further described below. The environmental data may indicate one or more characteristics of the patient’s environment (e.g., sounds of the environment) and / or other environmental characteristics (e.g., when the patient takes medication, reports the occurrence of a seizure, etc.). The sixth model training data may include historical LTBSC data of historical patients and historical environmental data of the historical patients, such that the sixth model may be trained to determine, for the historical patients, associations between historical LTBSCs and / or historical LTBs, and historical environmental characteristics. In one example, the environmental characteristic may be associated with a medication, such as a new medication taken by the patient, and / or a new titration of the medication taken by the patient. In at lest one aspect, the environmental data may be obtained in response to a LTBSC of the patient. For example, thesever 105 and / or the models 128, 130, such as the sixth model, may detect the LTBSC of the patient from the LTBSC data. In response, a message may be generated and provided to a user device 1 15, such as a user device of the patient, requesting they provide the environmental data, e.g., by asking questions about the environment of the patient via the message, and receiving responses from the patient via the user device 1 15. In one example, the implant device 150 may generate the environmental data, e.g., via sensors of the implant device 150 and / or user device 115, a user interface of the implant device 150 and / or user device 115, etc. In response to detecting the LTBSC, the server 105 may send a signal to the implant device 150 via the network 110, which results in the smart implant device 150 providing the environmental data to the server 105. In one example, the sixth model output data may indicate a prediction of a future LTBSC in response to the at least one environmental characteristic, and / or a treatment pathway to a target brain state.

[0110] Although the computing environment 100 is shown to include one each of the server 105, the network 110, the user device 1 15, and the implant device 150, it should be understood that different numbers of servers 105, networks 1 10, user devices 115, and / or implant devices 150 may be included in the computing environment 100. In one example, the computing environment 100 may include a plurality of servers 105, thousands of user devices 115, and hundreds of implant devices 150, all of which may be interconnected via the network 110.

[0111] The computing environment 100 may include additional, fewer, and / or alternate components, and may be configured to perform additional, fewer, or alternate actions, including components / actions described herein. Although the computing environment 100 is shown in Fig. 1 as including one instance of various components such as user device 1 15, server 105, network 110, etc., various aspects include the computing environment 100 implementing any suitable number of any of the components shown in Fig. 1 and / or omitting any suitable ones of the components shown in Fig. 1 . For instance, information described as being stored in the database 126 may be stored in the memory 125, and therefore the database 126 may be omitted. Moreover, various aspects include the computing environment 100 including any suitable additional component(s) not shown in Fig. 1 , such as but not limited to the exemplary components described above. Furthermore, it should be appreciated that additional and / or alternative connections between components shown in Fig. 1 may be implemented. As just one example, the user device 115 and the implant device 150 may be connected via a direct communication link (not shown in Fig. 1 ) instead of, or in addition to, via network 1 10.EXAMPLE COMPUTING DEVICE

[0112] Referring now to Fig. 2, in one embodiment the user device 1 15 may include a computing device 215, such as a laptop computer, smartphone, desktop computer, smartwatch, wearable, augmented reality device, virtual reality device, etc. The computing device 215 may include a display 240, a controller 242, one or more sensors 256, communication circuitry 258 such as communication circuitry 122, and a user-input device (not shown).

[0113] The controller 242 may include a program memory 246 (e.g., such as the memory 125 and / or the database 126), a microcontroller / processor / microprocessor (piP) 248 (e.g., such as the processor 120), a random-access memory (RAM) 250 (e.g., such as the memory 125), and / or an input / output (I / O) module 254 (e.g., such as the I / O module 148), all of which may be interconnected via an address / data bus 252. The program memory 246 may include an operating system 260, a data storage 262 (e.g., such as the memory 125 and / or the database 126), a plurality of software applications 264, and / or a plurality of software routines 268. The operating system 260 may be an operating system as previously described with respect to server 105, a mobile operating system such as iOS® or Android™ developed by Apple Inc. and Google Inc., respectively, and / or any other suitable operating system.

[0114] The computing device 215 may include one or more sensors 256. Exemplary sensors 256 may include one or more accelerometers, gyroscopes, inertial measurement units (IMUs), GPS units, proximity sensors, cameras, microphones, as well as any other suitable sensors. Additionally, other types of currently available or later-developed sensors may be included in some embodiments. The sensors 256 may be used to capture the LTB data of the patient (e.g., facial images, audio of the patient’s voice, etc.), data regarding the physical environment, such as the environmental data, and / or any other suitable data. The sensors 256 may be integrated with the computing device 215, and / or communicatively coupled to the computing device 215. The data the sensors 256 generate may be stored in memory, such as the memory 246, 250, provided to one or more devices, such as to the server 105 via the network 110, etc.

[0115] The communication circuitry 258 may communicate with the one or more servers 105, user devices 115, implant devices 150, or another suitable device or devices via any suitable wired or wireless communication protocol network, such as those described with respect to communication circuitry 122, such as a wireless telephony network (e.g., GSM, CDMA, LTE, 5G, 6G, UWB etc.), a Wi-Fi network (802.11 standards), a WiMAX network, a LAN, Ethernet, a Bluetooth network, etc. The user-input device (not shown) may include a “soft” keyboard that is displayed on the display 240 of the computing device 215, an external hardware keyboard communicating via a wired and / or a wireless connection (e.g., a Bluetooth keyboard), an external mouse, a touchscreen, a stylus, and / or any other suitable user-input device.

[0116] The data storage 262 may include data such as profile data including one or more user profiles (e.g., the user profile described herein, such as for the user of the computing device 215), application data for the plurality of applications 264, routine data for the plurality of routines 268, and / or other data necessary to interact (e.g., via network 1 10) with the one or more servers 105, user devices 115, implant devices 150, or another suitable devices. In some embodiments, the controller 242 may also include, or otherwise be communicatively connected to, other data storage mechanisms (e.g., one or more hard disk drives, optical storage drives, solid state storage devices, etc.) that reside within the computing device 215.

[0117] It should be appreciated that although Fig. 2 depicts, for example, one controller 242, one processor 248, one display 240, etc., the computing device 215 may include controllers 242, multiple processors 248, multiple displays 240, etc. Similarly, the memory of the controller 242 may include multiple RAMs 250 and / or multiple program memories 246. Although Fig. 2 depicts the I / O circuit 254 as a single block, the I / O circuit 254 may include a number of different types of I / O circuits. The controller 242 may implement the RAM(s) 250 and / or the program memories 246 as semiconductor memories, magnetically-readable memories, and / or optically-readable memories, for example.

[0118] The one or more processors 248 may be adapted and / or configured to execute any one or more of the plurality of software applications 264 and / or any one or more of the plurality of software routines 268 residing in the program memory 242, in addition to other software applications. One of the plurality of applications 264 may be a client application 266 (e.g., a mobile app) associated with the implant device 150 that may be implemented as a series of machine- readable instructions for performing the various tasks associated with receiving data from, and / or transmitting information to the implant device 150 and / or server 105, displaying such data, etc.

[0119] One of the plurality of applications 264 may be a native application and / or web browser 270, such as Apple’s Safari®, Google Chrome™ mobile web browser, that may be implemented as a series of machine-readable instructions for receiving, interpreting, and / or displaying application screens or web page information from the one or more servers 105 while also receiving inputs from the user. Another application of the plurality of applications may include an embedded web browser 276 that may be implemented as a series of machine-readable instructions for receiving, interpreting, and / or displaying web page information.

[0120] In one aspect, a user may launch a client application 266 from a client device, such as one of the user devices 115, to communicate with the one or more servers 105. In one example, the client application 266 may provide access to (e.g., downloading, viewing, annotating, labeling, providing feedback on, etc.) LTEEG data generated by the implant device 150, predictionsgenerated by one or more models 128, 130, ML model training data, and / or any other suitable data. Additionally, the user may also launch or instantiate any other suitable user interface application (e.g., the native application or web browser 270, and / or any other one of the plurality of software applications 264) to access the one or more servers 105 to realize aspects of the inventive system.

[0121] The computing device 215 may include additional, fewer, and / or alternate components, and may be configured to perform additional, fewer, or alternate actions, including components / actions described herein. Although the computing device 215 is shown in Fig. 2 as including one instance of various components such as display 240, processor 248, etc., various aspects include the computing device 215 implementing any suitable number of any of the components shown in Fig. 2 and / or omitting any suitable ones of the components shown in Fig. 2. Moreover, various aspects include the computing device 215 including any suitable additional component(s) not shown in Fig. 2, such as but not limited to the exemplary components described above. Furthermore, it should be appreciated that additional and / or alternative connections between components shown in Fig. 2 may be implemented.EXAMPLE IMPLANT DEVICE SYSTEM

[0122] Fig. 3A depicts, in its simplest form, a block diagram of the implant device system 300, such as implant device 150. The implant device system 300 includes a sensor array 302, a processor device 304, and a user interface 306. The sensor array 302 generally provides data, in the form of electrical signals, to the processor device 304. The user interface 306 may facilitate self-reporting by the patient of any of various data including events perceived by the patient, as well as medication types, doses, dose times, patient mood, potentially relevant environmental data, and the like, e.g., to generate the environmental data. The user interface 306 may also facilitate programming of the unit (e.g., via signal acquisition settings), calibration of the sensor array 302, providing feedback to the patient (e.g., haptic, vibratory feedback warning of a potential impending health event), etc.

[0123] Turning now to Fig. 3B, the implant device system 300 is presented as a block diagram in greater detail. The sensor array 302 may include a local processing / memory device 344 and a plurality of electrode devices 310A-D, each including an electrode 312. The local processing device / memory 344 may include an electrical amplifier 346 (e.g., pre-amp), a battery 348, battery charging technology 349, a transceiver 350, an analog-to-digital converter (ADC) 352, and a processor 354 to process electrical signals (e.g., EEG signals indicative of brain activity of the patient) received from or transmitted to the electrodes devices 310A-D.

[0124] The local processing device 344 may include a battery 348 and battery charging technology 349 to power the local processing device 344, among other things. The battery charging technology 349 may include a battery charging circuit, for facilitating charging of the battery 348. The battery charging technology 349 may be any known battery charging technology compatible with the arrangement of the sensor array 302. In particular, in embodiments the battery technology 349 may be an inductive charging circuit that facilitates charging through the patient’s skin when the sensor array 302 is disposed beneath the scalp of the patient. In other embodiments, the battery technology 349 may draw energy from the movements of the patient throughout the day by, for example, harnessing the movements of the patient to turn a small generator. In still other embodiments, the battery technology 349 may draw power from the environment in the form of RF signals. In further examples still, the battery technology 349 may draw power from chemical reactions taking place in the environment of the sensor array 302. Of course, more traditional charging methods (e.g., using a wired connection to provide power to the battery technology 349) may also be employed.

[0125] The local processing device 344 may include a transceiver 350, such as a transceiver receiver / transmitter pair that facilitates communication with other various devices (e.g., in a wired and / or wireless manner). The local processing device 344 may include a memory 355 to store EEG data, to store implant device settings (e.g., signal acquisition settings), or any other suitable data. The local processing device 344 may be similar to a processing device of a type commonly used with cochlear implants, although other configurations are possible.

[0126] The data processed and stored by the local processing device 344 may be raw (e.g., unprocessed) EEG signal(s) or partially processed (e.g., partially or fully compressed) EEG signal(s), for example. The EEG signal(s) may be processed and / or otherwise converted to EEG data, such as the LTEEG data, by the local processing device 344, e.g., via the ADC 352, although as used herein the terms EEG signal(s), EEG, EEG data, long-term EEG, and / or long-term EEG data may, in instances, be used interchangeably. The EEG data may be transmitted from the local processing device 344 wirelessly, or via a wire / in a wired manner, to the processor device 304 for further processing and analyzing of the EEG data. The processor device 304 may analyze (e.g., via one or more models, such as a models 128, 130) EEG signal data (or other electrical signals) to determine one or more LTBSCs, etc., among other things. Data may be generated by the processor device 304 on the basis of the analysis, such as the LTBSC data indicating and / or including one or more LTBSCs, the associated LTEEG / LTEEG data, associated LTBs / LTB data, etc.

[0127] By carrying out data analysis externally to the sensor array 302, using the processor device 304, for example, there may be a reduction in power consumption within the sensor array 302, enabling the sensor array 302 to retain a smaller geometrical form. Moreover, the processor device 304 may have significantly higher processing power than would be possible with any processor included in the sensor array 302. The processor device 304 may run software that continuously records electrical data received from the sensor array 302. In other embodiments, the sensor array 302 and / or processor device 304 may store the EEG data to, and / or retrieve the EEG data from, a memory at one or more time, e.g., the memory 125, the database 126, the memory 355, and / or any other suitable memory. In such embodiments, the EEG data may be processed by the processor device 304 at any suitable time, e.g., the processor device 304 may retrieve the LTEEG data of the patient from one or more memories several hours after it is recorded, for processing by one or more models.

[0128] The following description of the sensor array 302 is illustrative in nature. While one of skill in the art would recognize a variety of sensor arrays that may be compatible with the described embodiments, the sensor arrays 302 explicitly described herein may have particular advantages and, in particular, the sensor arrays 302 may include the sensors described in U.S. Patent Application 16 / 125,152 (U.S. Patent Application Publication No. 2019 / 0053730 A1 ) and U.S. Patent Application 16 / 125,148 (U.S. Patent No. 10,568,574) the specifications of each being hereby incorporated herein by reference, for all purposes. Alternative embodiments of the sensor array 102 are contemplated, such as that described in U.S. Patent Application No. 16 / 797,315, entitled “Electrode Device for Monitoring and / or Stimulating Activity in a Subject,” the entirety of which is hereby incorporated by reference herein.

[0129] Turning to Fig. 3C, an electrode device 310 is provided comprising an elongate, implantable body 342 and a plurality of electrodes 312 positioned along the implantable body 342 in the length direction of the implantable body 342. At a proximal end of the implantable body 342, a processing unit 344 is provided for processing electrical signals that can be sent to and / or received from the electrodes 312. Though not required, in some embodiments, an electrical amplifier 346 is positioned in the implantable body 342 between the electrodes 312 and the processing unit 344. In an alternative embodiment, the electrical amplifier 346 may be integrated into the processing unit 344 of the electrode device 310, instead of being positioned in the implantable body 342. A plurality of anchors 313 may be positioned along a length of the implantable body 342, each adjacent a respective one of the electrodes 312. The anchors 313 may be designed to provide stabilization to the electrode device 310 when it is in the implantation position.

[0130] Fig. 3C depicts the electrode device 310 as having four electrodes 312 in the embodiment to a local processing device 344. Of course, in different embodiments, more or fewer numbers of electrodes 312 may be implemented, according to the needs of the electrical signals required for the implementation of the methods described herein. In particular, the electrode device 310 may include, in embodiments, four (4), eight (8), ten (10), twelve (12), sixteen (16), twenty (20), twenty- four (24), or more electrodes 312. In the embodiment depicted, the local processing device 344 and electrode 312 are formed in the electrode device 310 as a one-piece construct. The arrangement is such that the local processing device 344 and the electrodes 312 are permanently fixed together (for the purpose of normal operation and use). There is therefore no requirement or indeed possibility for a user, such as a surgeon, to connect these components of the sensor array 302 together prior to implantation, therefore increasing the strength, cleanliness and ease of use of the sensor array 302.

[0131] The electrodes 310 may be configured to form pairs 341 A, 341 B, e.g., one pair 341 A implanted over the right hemisphere of the brain and one pair 341 B implanted over the left hemisphere of the brain, respectively. For example, the first electrode pair 341 A may be used to monitor electrical activity at right hemisphere of the brain and the second electrode pair 341 B may be used to monitor electrical activity at the left hemisphere of the brain, or vice-versa. Independent electrical activity data may be recorded for each of the right and left hemispheres. The electrode pairs 341 A, 341 B may be positioned away from the subject’s eyes and chewing muscles to avoid introduction of signal artifacts from these locations (e.g., clipping during chewing).

[0132] Fig. 3D provides a cross-sectional view of the implantable body 342 and the electrode 312. The implantable body 342 may have a round, e.g., substantially circular or ovate, cross- sectional profile. Similarly, each of the electrodes 312 may have a round, e.g., substantially circular or ovate, cross-sectional profile. Each of the electrodes 312 may extend circumferentially, completely around a portion of the implantable body 342. By configuring the implantable body 342 and electrodes 312 in this manner, the exact orientation of the implantable body 342 and electrodes 312, when implanted in a subject, is less critical. For example, the electrodes 312 may interact electrically with tissue in substantially any direction. In this regard, the electrodes 312 may be considered to have a 360-degree functionality. The round cross-sectional configuration may also provide for easier insertion of the implantable portions of the electrode device 310 to the target location and with less risk of damaging body tissue. For example, the implantable body 342 may be used with insertion cannulas or sleeves and may have no sharp edges that might otherwise cause trauma to tissue.

[0133] In one embodiment, the implantable body 342 may be formed of an elastomeric material such as medical grade silicone. Each electrode 312 may comprise an annular portion of conductive material that extends circumferentially around a portion of the implantable body 342. More specifically, each electrode 312 may comprise a hollow cylinder of conductive material that extends circumferentially around a portion of the implantable body 342 and, in particular, a portion of the elastomeric material of the implantable body 342. The electrodes 312 may be considered ‘ring’ electrodes.

[0134] In one embodiment, to strengthen the engagement between the electrodes 312 and the implantable body 342, straps 365 may be provided that extend across an outer surface of each electrode 312. In such an embodiment, two straps 365 may be located on substantially opposite sides of each electrode 312 in a direction perpendicular to the direction of elongation of the implantable body 342. The straps 365 may prevent side sections of the implantable body 342 from pulling or breaking away from the electrodes 312 when the implantable body 342 is placed under tension and / or is bent. The straps 365 may similarly be formed of elastomeric material. In alternative embodiments, a different number of straps 365 may be employed (e.g., one, three, four or more straps 365), or the straps 365 may be omitted.

[0135] An electrical connection 340 may extend through the implantable body 342. The electrical connection 340 to the electrodes 312 may comprises relatively fragile platinum wire conductive elements. To reduce the likelihood that the platinum wires will break or snap during bending, flexing and / or stretching of the implantable body 342, the electrical connection 340 may be provided with a wave-like shape and, more specifically, a helical shape, although other non-linear shapes may be used. The helical shape, for example, of the electrical connection 340 enables the electrical connection 340 to stretch, flex and bend in conjunction with the implantable body 342. Bending, flexing, and / or stretching of the implantable body 342 typically occurs during implantation of the implantable body 342 in a subject and upon any removal of the implantable body 342 from the subject after use.

[0136] A reinforcement device 368 may be provided in the electrode device 312, which reinforcement device 368 extends through the implantable body 342 and limits the degree by which the length of the implantable body 342 can extend under tension. The reinforcement device 368 may take the bulk of the strain placed on the electrode device 310 when the electrode device 310 is placed under tension. The reinforcement device 368 may comprise a fiber (e.g., strand, filament, cord or string) of material that is flexible, and which has a high tensile strength. In particular, a fiber of ultra-high-molecular-weight polyethylene (UHMwPE), e.g., Dyneema™, may be provided as the reinforcement device 368 in the present embodiment. The reinforcement device 368 may extendthrough the implantable body 342 in the length direction of the implantable body 342 and may be generally directly encased by the elastomeric material of the implantable body 342.

[0137] The reinforcement device 368 may comprise a variety of different materials in addition to or as an alternative to UHMwPE. The reinforcement device may comprise other plastics and / or non-conductive material such as a poly-paraphenylene terephthalamide, e.g., Kevlar™. In some embodiments, a metal fiber or surgical steel may be used.

[0138] Similar to the electrical connection 340, the reinforcement device 368 may also have a wave-like shape and, more specifically, a helical shape, although other non-linear shapes may be used. The helical shape of the reinforcement device 368 may be different from the helical shape of the electrical connection 340, for example the helical shape of the reinforcement device 368 may have a smaller diameter than the helical shape of the electrical connection 340. Moreover, the helical shape of the reinforcement device 368 may have a greater pitch than the helical shape of the electrical connection 340.

[0139] Turning now to Fig. 3E, the processor device 304 is presented as a block diagram in greater detail. As depicted in Fig. 3E, the implant device system 300 includes, in embodiments, one or more sensors 351 , in addition to the sensor array 302 (e.g., one or more of which may be part of implant device 150), the processor device 304, and the user interface 306. The sensor array 302 and / or the sensors 351 may sense or collect respective data and communicate the respective data to the processor device 304. As may by now be understood, in embodiments, the sensor array 302 may include an array of electrode devices 310 that provide electrical signal data. In particular, the electrode devices 310 may provide electrical signal data indicative of brain activity of the patient (e.g., EEG signal data). While the sensor array 302 may be disposed beneath the scalp of the patient, e.g., on and extending into the cranium so as to facilitate accurate sensing of brain activity, in embodiments it is also contemplated that the sensor array 302 need not be placed beneath the scalp.

[0140] The sensors 351 , such as sensors 256, may include one or more infrared sensors, cameras, microphones, as well as any other suitable sensors. Additionally, other types of currently available or later-developed sensors may be included in some embodiments. The sensors 256 may be used to capture the LTB data of the patient (e.g., facial images), data regarding the physical environment, such as the environmental data, and / or any other suitable data. The sensors 351 may be integrated with the sensory array 302, the electrodes 310, the processor device 304, and / or communicatively coupled to the sensory array 302, the electrodes 310, and / or the processor device 304.

[0141] The sensors 351 may include at least one camera to capture images related to the patient (e.g., facial biometrics) and the patient’s environment. The camera may be, and or include, a CMOS sensor, an infrared sensor, a digital camera, a web camera, and / or any other suitable camera. The camera may be any type of camera suitable for capturing images of the patient, such as black and white images, color images, infrared images, stereo images, video, etc. In embodiments, the camera may be configured to capture facial biometrics of the patient automatically (e.g., according to a schedule, based upon EEG data indicating a health event, based upon recognition of a facial biometric, etc.). The camera may be disposed on the patient and suitable for operation from a portable power source such as a battery, e.g., one or more cameras in a smart watch, headset / head-mounted display (e.g., augmented reality, virtual reality, mixed reality, smart glasses), and / or other wearable camera. In some embodiments, the camera may be integrated with, and / or communicatively coupled to, one or more devices, such as the user device 115, the computing device 215, and / or the processor device 304.

[0142] The sensors 351 , such as the camera, may generate LTB data indicating biometrics of the patient, such as facial biometrics via images or video of the patient’s face showing facial expressions, eye movements, etc. The facial biometrics, individually or in combination with data from the sensor array 302 (e.g., EEG data), self-reported data received via the user interface 306, and / or other suitable data, may assist algorithms executing within the processor device 304 in determining whether the patient has experienced a health event of interest (e.g., seizure, medication side effect, etc.) and, if so, classifying the health event. For example, the camera may generate facial biometric data comprising images which depict a drooping eye of the patient, that may indicate that patient is experiencing a seizure. Such an indication, in conjunction with electrical signals detected by the sensor array 302, may provide corroboration that the patient has, in fact, experienced a health event of interest.

[0143] The sensors 351 may include at least one microphone to detect sound related to the patient and the patient’s environment, e.g., sounds included in the environmental data and / or LTB data. The microphone 351 may be any type of microphone suitable for disposal on the patient and suitable for operation from a portable power source such as a battery. In particular, the microphone 351 may be a piezoelectric microphone, a MEMS microphone, or a fiber optic microphone. The microphone 351 may be disposed at any of a variety of positions on the patient including, but not limited to, the patient’s head, arm, torso, leg, hand, or neck. In some embodiments, the microphone 351 may be integrated with the sensor array 302 and placed on or beneath the scalp of the patient with the sensor array 302, while in others the microphone 351 may be integrated with the processor device 304, and still in others the microphone 351 may be distinct from both the sensor array 302and the processor device 304. In embodiments, especially those in which the patient’s voice is the primary sensing target for the microphone 351 , the microphone 351 senses sound via bone conduction. Of course, while depicted in the accompanying figures as a single microphone, the microphone 351 may be one or more microphones, disposed as an array in a particular position on the patient, or as distinct units on a variety of positions on the patient. In embodiments implementing multiple microphones, the multiple microphones may be of the same type, or may be different, depending on the location of each on the patient, the environment in which each is disposed (e.g., sub-scalp vs. not), the location of each in the hardware (e.g., separate from other devices or integrated within the processor device 304, for example), etc. Each may have the same or different directionality and / or sensitivity characteristics as the others, depending on the placement of the microphone and the noises or vibrations the microphone is intended to detect.

[0144] The sensors 351 , such as the microphone, may detect the patient’s voice, in embodiments, with the goal of determining one or more of: vocal biometrics, pauses in vocalization; stutters; periods of extended silence; abnormal vocalization; and / or other vocal abnormalities that, individually or in combination with data from the sensor array 302, and / or self-reported data received via the user interface 306, may assist algorithms executing within the processor device 304 in determining whether the patient has experienced an health event of interest (e.g., seizure, LTBSC, etc.) and, if so, classifying the health event. In embodiments, the microphone may record one or more samples of the user’s voice when biometric images of the patient (e.g., of the face, eyes, etc.), and / or EEG of the patient, are captured. The vocal biomarkers of the voice samples can then be correlated to facial biometrics and / or the EEG. In embodiments, the microphone may also detect other noises in the patient’s environment that may be indicative that the patient experienced a health event of interest. For example, the microphone may detect the sound of glass breaking, that may indicate that the patient has dropped a glass. Such an indication, in conjunction with electrical signals detected by the sensor array 302, may provide corroboration that the patient has, in fact, experienced a health event of interest.

[0145] In some embodiments, the sensors 351 may neither be necessary nor required in order to generate data associate with the detection and / or prediction of a health event, generate environmental data, generate LTB data, and / or any other data, (e.g., in situations only requiring LTBSC data to generate an model output), and thus are considered optional. Accordingly, the sensors 351 are depicted with dotted lines to denote that they are optional.

[0146] Together, the sensor array 302 and, if present, the sensor(s) 351 , may provide data from which biomarker data related to the patient(s) may be extracted. As used herein, the term “biomarker” refers to a broad subcategory of objective indications of medical state, observable fromoutside the patient, that may be measured accurately and reproducibly. Biomarkers differ from symptoms, which are generally perceived by patients themselves. The system 100 may be configured to determine a variety of biomarkers depending on the inclusion and / or placement of the various sensor devices (i.e., the sensor array 302 and, if present, the sensors(s) 351 ). By way of example, and not limitation, mood disruption biomarker data may be determined from microphone data collected by one or more microphones; speech production biomarker data may be determined from microphone data collected by one or more microphones; epileptiform activity biomarker data may be determined from EEG data received from one or more electrode devices 310 in the sensor array 302; jaw movement biomarker data may be determined from a combination of electromyography data and microphone data collected by one or more devices (e.g., electrode devices 310) disposed on the patient; vomiting biomarker data may be determined from a combination of electromyography data and / or microphone data collected by one or more devices (e.g., electrode devices 310, sensors 351 ) disposed on the patient; sleep biomarker data may be determined from EEG data received from one or more electrode devices 310 in the sensor array 302, etc.

[0147] Turning to the processor device 304, the processor device 304 may include communication circuitry 356, sensors 351 , a microprocessor 358, and a memory device 360. The microprocessor 358 may be any known microprocessor configurable to execute the routines (e.g., for detecting and classifying health events of interest, generating signal acquisition settings, and / or filtering health event data), including, by way of example and not limitation, general purpose microprocessors (GPUs), RISC microprocessors, digital signal processors (DSPs), application specific integrated circuits (ASICs), and field-programmable gate arrays (FPGAs).

[0148] The communication circuitry 356, such as communication circuitry 122, 258, may be any transceiver and / or receiver / transmitter pair that facilitates communication with the various devices from which the processor device 304 receives data and / or transmits data, such as the server 105, user device 1 15, or any other suitable device. The communication circuitry 356 is communicatively coupled, in a wired or wireless manner, to each of the sensor array 302, the sensors 351 , and the user interface 306. Additionally, the communication circuitry 356 is coupled to the microprocessor 358, which, in addition to executing various routines and instructions for performing analysis (e.g., of LTEEG data), may also facilitate storage in the memory 360 of data received, via the communication circuity 356, from the sensor array 302, the sensors 351 , and / or the user interface 306.

[0149] The sensors 351 , e.g., via the microprocessor 358 and / or communication circuitry 356, may generate and store sensor data in memory such as the memory 360, provide the sensor data to another device, such as the server 105 and / or user device 1 15, etc.

[0150] The memory 360 may include both volatile memory (e.g., random access memory (RAM)) and non-volatile memory, in the form of magnetic and / or solid-state media. In addition to an operating system (not shown), the memory 360 may store sensor array data 364 received from the sensor array 302. The data stored in the memory 360 may be stored with corresponding metadata, labels, time stamps, etc.

[0151] In some embodiments, the memory 360 may include one or more models 370. The models 370 may include a first model 370A which operates on the LTEEG data 372, to generate the LTBSC data 375A. The LTEEG data 372 may include substantially continuous EEG data of the patient over more than a week of time, or any other amount of time which may be suitable to detect a LTBSC, e.g., several days, several weeks, several months, etc. The LTBSC data 375A may include and / or indicate a classification associated with the at least one LTBSC (e.g., classifications indicating a type of LTBSC), a confidence metric (e.g., a measure indicating the level of certainty or reliability associated with the detection and / or classification of the LTBSC), and / or any other suitable data associated with the LTBSCs, as previously described. Some example LTBSC classifications include:

[0152] Cognitive Decline: Some individuals with epilepsy, especially those with chronic, uncontrolled seizures, may experience cognitive decline over time. This can include difficulties with memory, attention, language, and problem-solving skills. The degree of cognitive decline can vary widely among individuals and may be influenced by the location and frequency of seizures, the age of onset, and the specific epilepsy syndrome;

[0153] Mood Disorders: There is a higher prevalence of mood disorders, including depression and anxiety, among people with epilepsy compared to the general population. These conditions can be a reaction to the challenges of living with epilepsy, but they may also be related to the underlying brain changes caused by the disorder and recurrent seizures;

[0154] Structural Changes: Chronic epilepsy can lead to structural changes in the brain, including atrophy (shrinkage) of certain brain regions and alterations in brain networks. These changes can be detected using neuroimaging techniques such as MRI (Magnetic Resonance Imaging). The hippocampus, a region critical for memory formation, is particularly susceptible to damage in temporal lobe epilepsy, a common form of the disorder;

[0155] Alterations in Brain Connectivity: Beyond structural changes, epilepsy can lead to alterations in how different regions of the brain communicate with each other. These changes in brain connectivity can impact a wide range of functions, from cognitive abilities to emotional regulation; and

[0156] Increased Risk of SUDEP: While not a change in brain state per se, it's important to note that individuals with epilepsy, especially those with poorly controlled seizures, have an increased risk of SUDEP. This risk may be linked to the cumulative impact of seizures on brain function and heart and respiratory regulation.

[0157] The models 370 may include a second model 370B which operates on the LTBSC data 375A (e.g., the LTBSC data 375A output by the first model 370A) to generate second model output data 375B. The second model output data 375C may indicate a seizure of the patient associated with the LTBSC data 375A. For example, the LTBSC data 375A may include the LTEEG data 372 of the patient which, once processed by the second model 370B, indicates one or more seizures, such as seizures associated with the LTBSCs and / or LTBs of the LTBSC data 375A. The second model output data 375C may include the LTEEG data 372 indicating the seizure (e.g., an EEG snippet of the seizure), a classification of the seizure, a confidence metric associated with the classification and / or indication of the seizure, LTB data 375A associated with the seizure (e.g., when the LTBSC data includes corresponding LTB data), and / or any other suitable data associated with detecting the seizure, as previously described etc.

[0158] The models 370 may include a third model 370C which operates on the LTBSC data 375A wherein (i) the third model output data includes a revision of an indication of a seizure in the LTEEG data of the LTBSC data; (ii) the indication of the seizure is not based upon the at least one LTBSC of the LTBSC data; and (iii) the revision to the indication of the seizure is based upon the at least one LTBSC of the LTBSC data. The third model 370C may generate third model output data 275C. The third model output data 375C may include one or more revisions of an indication of a seizure in the second model output data 375B, and / or any other suitable data associated with revising the indication of a seizure, as previously described.

[0159] The models 370 may include a fourth model 370D which operates on the LTBSC data 375A to generate fourth model output data 375D indicating a prediction of a seizure of the patient based upon receiving the LTBSC data. The fourth model output data 375D may include at least one date of the prediction of the seizure, a confidence metric associated with the prediction and / or any other suitable data associated with predicting the seizure, as previously described etc.

[0160] The models 370 may include a fifth model 370E which operates on the LTBSC data 375A to generate fifth model output data 375E indicating at least one date to implement a medicationtitration of the patient, and / or any other suitable data associated with implement a medication titration of the patient, as previously described etc.

[0161] The models 370 may include a sixth model 370F which operates on environmental data 374 indicating the at least one environmental characteristic of the patient, and the LTBSC data 375A. The sixth model 370F may generate sixth model may generate sixth model output data 375F indicating an association between at least one environmental characteristic of the patient and the at least one LTBSC of the patient, and / or any other suitable data.

[0162] Of course, it should be understood that wherever a routine, model, or other element stored in memory is referred to as receiving an input, producing or storing an output, or executing, the routine, model, or other element, it is, in fact, executing as instructions on the microprocessor 358. Further, those of skill in the art will appreciate that the model or routine or other instructions would be stored in the memory 360 as executable instructions, which instructions the microprocessor 358 would retrieve from the memory 360 and execute. Further, the microprocessor 358 should be understood to retrieve from the memory 360 any data necessary to perform the executed instructions (e.g., data required as an input to the routine or model), and to store in the memory 360 the intermediate results and / or output of any executed instructions.EXAMPLE ML MODEL TRAINING

[0163] Fig. 4 schematically illustrates how a server such as the server 105, a user device, such as the user device 1 15, a computing device such as the computing device 215, and / or other suitable device, may execute an ML module 405 to train one or more ML models 410 using training data 420 to operate on at least one input 430 to generate at least one output 450.

[0164] In one aspect, the ML module 405 may include one or more hardware and / or software components such as the ML module 140, the MLTM 142, and / or the MLOM 144 of server 105. The ML module 405 may obtain, create, train, retrain, fine-tune, retrieve, load, operate and / or save one or more ML models 410, such as ML model(s) 130.

[0165] The ML module 405 may use training data 420 to generate, train, retrain and / or fine-tune the one or more ML models 410. For the purposes of discussion herein, the term training data may be used interchangeably with the term historical training data, and more generally data may at times be used interchangeably with the term historical data (e.g., data that is already in existence may be considered historical relative to data that is yet-to-be generated). The server may store the training data 420 in a memory at one or more times (e.g., to update the training data 420), such as the memory 125 and / or the database 126 of the server 105, data storage 262 of computing device 215, and / or any other suitable memory of any other suitable device. In one aspect, the training data420 may include one or more training datasets, such as training datasets 420A, 420B, 420C, etc. The one or more training datasets may include any suitable training data 420, such as one or more types of training data described herein.

[0166] In one aspect, one or more sets of training data may be used to train iterations of an ML model 410 such as ML models iterations 410A, 41 OB, 410C, etc. The iterative ML models may be trained to provide one or more outputs associated with each iterative ML model, such as output 440A associated with first iterative ML model 410A, output 440B associated with second iterative ML model 41 OB, output 440C associated with third iterative ML model 410C, and so on. While three sets of training data 420A-420C, three iterative ML models 41 OA-410C, and three sets of outputs 440A-440C are depicted in Fig. 4, this is merely an example for ease of illustration only and any number and / or variety of training data 420, models 410, and / or output 440 may be implemented by the disclosed techniques.

[0167] The ML module 405 may implement a supervised or unsupervised machine learning program or algorithm, as previously described. In some aspects, a training dataset may include a set of feature values. In embodiments in which the ML module 405 implements unsupervised learning algorithms, the ML module 405 may find its own structure in unlabeled feature values of a training dataset. In embodiments in which the ML module 405 implements supervised learning algorithms, the ML module 405 (e.g., via MLTM 142) may store one or more routines that facilitate the labeling of the feature values (e.g., by clinician reviewing the feature values and / or the training dataset, such as training datasets 420A, 420B, 420C, etc.) to create one or more label attributes associated with the feature values. In one example, the LTEEG data may include features associated with LTBSCs of the patient, seizures of the patients, and / or any other suitable features. In one example, the LTB data may include features associated with the LTBs of the patient (e.g., biometrics of the face and / or eyes), with LTBSCs of the patient (e.g., LTBSCs associated with LTBs), with seizures of the patient (e.g., seizures associated with LTBs), and / or any other suitable feature. In one example, the LTBSC data, which may include the LTEEG data and / or the LTB data, may include any of the just-described features, as well as any other suitable features. The ML module 405 may use both the feature values and the label attributes to discover rules, relationships, or other “models” that map the features to the labels by, for example, determining and / or assigning weights or other metrics. The ML module 405 may output the set of rules, relationships, or other models as a trained ML model, such as first ML model 410A.

[0168] In one aspect, the first trained ML model 410A may operate on the first training dataset 420A and / or the feature values extracted therefrom, or on a portion of the first training dataset 420A and / or a portion of the feature values extracted therefrom that were reserved for validating the firsttrained ML model 41 OA, in order to provide results, such as first trained ML model output 440A, for comparison and / or analysis by a trained professional in order to validate the output 440A of the first ML model 41 OA.

[0169] In embodiments, the first trained ML model 41 OA may be trained by a first set of hardware, such as the server 105, and stored to a memory such as memory 125 and / or database 126. The model 41 OA may be subsequently trained and / or operated by the first set of hardware, or may be provided (or otherwise received - e.g., via portable storage media) to second set of data collection hardware for further training, execution, etc., such as the implant device system 300 depicted in any of Figs. 3A-3E. The second set of data collection hardware may implement the first trained ML model 41 OA to provide an associated output 440A based on data that was collected the data collection hardware and not part of the first training dataset 420A or, alternatively, may simply collect additional data for use by the ML module 405 to iterate the first trained ML model 410A.

[0170] As should be understood, the training data 420 may include additional training datasets, such as a second training dataset 420B. The ML module 405 may implement a supervised or unsupervised machine learning program or algorithm, as described above, for iterating the first trained ML model 410A, which may have a first error rate associated with its first / primary output 440A, to create a second trained ML model 410B, which may have a second error rate, reduced from the first error rate, associated with the second / secondary trained ML model output 440B. The second trained ML model 410B may use the second training dataset 420B and / or the feature values extracted therefrom, or on a portion of the second training dataset 420B and / or a portion of the feature values extracted for validating the second trained ML model 410B, in order to provide an output 440B for comparison and / or analysis by a trained professional in order to validate the output of the second trained ML model 410B. An error rate of the output 440B by the second trained ML model 410B may be reduced relative to an error rate of the output 440A by the first trained ML model 410A. The second trained ML model 410B (e.g., due to its reduced relative error rate) may be stored to a memory such as memory 125, database 126 or other suitable memory, as a new model, and / or replacing a previous model, and / or otherwise provided to one or more systems, such as implant device system 300.

[0171] Once trained, the ML model 410 may perform operations on one or more data inputs 430 to produce a desired data output 440. In one aspect, the ML model 410 may be loaded at runtime from a database (e.g., loaded by ML engine 405 from the database 126) to process the input data 430. The ML engine 405 may obtain the input data 430 (e.g., from the database 126), and provide the input data 430 for the trained ML model 410 to operate on, and generate the output 440.

[0172] The ML module 405 may use training data 420 to retrain and / or fine-tune one or more ML models 410. For the purposes of discussion herein, the terms training, retraining and / or fine-tuning may at times be used interchangeably. In one aspect, an ML model 410 may be retrained based upon updated training data 420. Retraining the model may provide for improved operation of the model, e.g., retraining the fourth model 414 to improve the accuracy of the prediction of a seizure. In one aspect, a model may be fine-tuned. Fine-tuning may include adding and / or adjusting the parameters (e.g., weights, layers) of a previously trained ML model based upon data (e.g., specific data, new data, etc.). For example, a model (e.g., the second mode 412) may be trained to detect a seizure based upon receiving EEG data (e.g., short-term EEG) of a patient. The model may be fine-tuned through a training process to accept LTBSC data as an input data to improve the seizure detection capabilities of the model.FIRST MODEL

[0173] In one aspect, the models 410 may include a first model 41 1 . The first model 41 1 may be trained via the ML module 405 using training data 420 (referred to as first model training data) to receive the LTEEG data as an input 430 and to generate the LTBSC data as the output 440. In at least one aspect, the first model 411 may also receive LTB data as an input 430. The LTB data may correspond to the LTEEG data in that the LTB data captures long-term biometrics of the patient during at least a portion of the time when the LTEEG data is recorded. The first model training data 420 may include, for a plurality of historical patients, historical LTEEG data, historical LTB data, and / or historical LTBSC, and / or any other suitable training data 420. The first model 411 may be trained to determine, for the historical patients, associations between patterns and / or features in historical LTEEG data and historical LTBs (if provided) corresponding to historical LTBSCs, and / or any other suitable associations or patterns. In one example, patterns of phase changes, sharps and / or spikes in one or more EEG signals and / or frequency bands of the LTEEG may be indicative of the LTBSC. In one example, biometrics of the eyes and / or face of the LTB data may be indicative of the LTBSC. As previously described, generally the LTBSC data may include the LTEEG data associated with the LTBSCs (e.g., the unprocessed EEG, the EEG after performing one or more transformations, such as RMS, of one or more EEG signals) and the associated LTB data. The LTBSC data may include a classification of the LTBSC, a confidence metric associated with the LTBSC, and / or any other suitable data.

[0174] In one example of the first model 411 , an EEG device of the patient (e.g., implant device 150), may capture LTEEG data (e.g., via a sensor array such as sensor 302) over the course of six months. The EEG device may provide the LTEEG data to the first model 370A executing on the processor device, such as processor device 304, in a wired or wireless manner (e.g., via a databus, communication circuitry 356, etc.). LTB data associated with the LTEEG data (e.g., generated at one or more times during which the LTEEG is being recorded) may be generated, e.g., by the patient via user device such as user device 1 15. The LTB data may comprise multiple facial images, audio voice recordings, and / or other LTBs. The patient may generate the LTB data over the same sixth months that the LTEEG data is recorded, via a mobile app associated with the EEG device and executing on their smartphone, such as client application 266 of computing device 215. The smartphone may provide the LTB data to the processor device via a network such as network 110. The processor device may provide the LTEEG data and the LTB data as an input 430 to the first model 411 to generate the LTBSC data as the output 440. The first model 411 may generate as an output 440 LTBSC data 441 which identifies one or more LTBSCs of the patient, as previously described. For example, the LTBSC data 441 may include and / or indicate the LTEEG associated with the LTBSCs, a classification of the LTBSCs, a confidence metric associated with the LTBSCs (e.g., the confidence that EEG indicates the LTBSC), the LTB data associated with the LTBSC (e.g., audio-visual data of the patient proximate the LTBSC), and / or any other suitable information. The LTBSC data 441 may indicate the patient experienced two LTBSCs during the six months’ time, may include EEG snippets visually indicating the LTBSCs in the EEG signals, as well as facial images depicting a change in long-term facial biometrics associated with the LTBSCs, as well and any other suitable information. The processor device may provide the LTBSC data 441 to the smartphone of the patient. The mobile application executing on the smartphone and associated with the implant device system may receive the LTBSC data 441 and display the associated EEG snippets and facial images associated with the LTBSCs, along with other information about the LTBSCs provided in the first mode output data (e.g., the dates, classifications., confidence metrics of the LTBSCs, etc.). The processor device may also store the first model output data in a memory, such as the memory 304 of the processor device, a database such as the database 126 (e.g., as training data, in a patient profile, etc.), and / or any other suitable memory.SECOND MODEL

[0175] In one aspect, the models 410 may include a second model 412 for detecting a seizure. The second model 412 may be trained using training data 420 (referred to as second model training data) to receive the LTBSC data as an input 430 and to generate, as the output 440, the second model output data 442 indicating a seizure of the patient. The second model training data 420 may include historical LTBSC data of historical patients, and / or any other suitable training data 420. The second model 412 may be trained to determine, for the historical patients, associations between at least one historical LTBSC and an historical seizure, and / or any other suitable associations or patterns. For example, patterns of sharps, spikes, or other signal characteristics in the LTEEG, thetransition to a new brain state, and biometrics indicating a change the patient face may be indicative of a seizure. The second model 412 may generate as an output 440 second model output data 442 indicating a seizure of the patient. The second model output data 442 may include and / or indicate the EEG associated with the seizure (e.g., an EEG snippet of the LTEEG of the LTBSC), a classification of the seizure (e.g., type of seizure), a confidence metric associated with the seizure (e.g., the confidence that EEG snippet indicates the seizure), the long-term biometric data associated with the seizure (e.g., images showing facial drooping of the patient proximate the seizure), and / or any other suitable information. In at least one aspect, if the LTBSC data indicates that a brain state change may coincide with the detection of a seizure, the second model 412 may subsequently operate with a higher sensitivity based upon the LTBSC, as there may be a higher likelihood that seizures may be indicated in the associated EEG, as previously described.

[0176] Continuing with the above example, the patient’s smartphone mobile app may be in communication with a server (e.g., server 105) associated with the mobile app and the patient’s implant device. The server may store the second model 412 and the patient’s LTBSC data in a database (e.g., the database 126), and periodically execute the second model 412 to determine whether the patient is experiencing a seizure, and if so, generate and provide an alert (e.g.,, an alert included in, or comprised of, the second model output data) to the mobile app of the smartphone indicting the seizure. The server may retrieve the patient’s LTBSC data from the memory, load the second model 412 from memory for execution via the ML module 405, and provide the LTBSC data to the second model 412. The second model 412 may operate on the LTBSC data to determine the patient is experiencing a seizure, and generate second model output data 442 indicating the seizure. The second model output data 442 may indicate there is a 95% chance a gran-mal seizure, and also include the LTEEG snippet associated with the seizure. In this example, the LTBSC data used by the second model 412 to detect the seizure may not include LTB data. The server may generate an alert associated with the second model output data, for example an alert warning the patient they may be experiencing a seizure and including the information from the second model output data, and also requesting to capture biometric facial images of the patient, if possible. The server may provide include the alert in the second model output data 442, and provide the second model output data 442 to the smartphone of the patient via the mobile app, as well as a computing device of a caretaker designated by the patient to receive alerts when a seizure is occurring. The server may store the second model output data 442 in the database.THIRD MODEL

[0177] In one aspect, the models 410 may include a third model 413 for revising information associated with the detection of a seizure in the LTEEG data of the LTBSC data, wherein theindication of the seizure is not based upon the at least one LTBSC of the LTBSC data, and the revision to the indication of the seizure is based upon the at least one LTBSC of the LTBSC data, as previously described. The third model 413 may be trained using training data 420 (referred to as third model training data) to receive as an input 430 the LTBSC data which includes an indication of a seizure in the LTBSC data, and to generate as an output 440 third model output data 443. The third model training data 420 may include historical LTBSC data of historical patients, and / or any other suitable training data. The third model 413 may be trained to determine, for the historical patients, associations between an historical seizure and at least one historical LTBSC, and / or any other suitable patterns or associations. The third model 413 may generate as an output 440 third model output data 443 indicating a revision of the seizure of the patient. The revision may include a reclassification of the seizure and / or a change in a confidence metric associated with the indication of the seizure, an identification of an EEG pattern which may be incorrectly classified as a seizure (e.g., for a human reviewer to review the determination of the seizure in light of the LTBSC data), and / or any other suitable information.

[0178] In one example, a human reviewer of the LTEEG data of the LTBSC data labels the LTBSC data to identify the occurrence of two clonic seizures. The reviewer identifies the seizures before any LTBSCs associated with the LTEEG data are determined, such that the identification of the seizures by the reviewer is not based upon the LTBSCs of the LTBSC data, and / or associated LTBs. The LTBSC data, now indicating the two seizures, the LTBSCs and the LTBs of the patient, may be provided to the third model 413. The third model may detect patterns in the LTBSCs and LTBs which indicates three seizures occurred rather than two, and accordingly revise the indications of the seizures to generate the third model output data 443.FOURTH MODEL

[0179] In one aspect, the models 410 may include a fourth model 414 for predicting a seizure. The fourth model 414 may be trained using training data 420 (referred to as fourth model training data) to receive the LTBSC data as an input 430, and to generate as the output 440 the fourth model output data 444, indicating a prediction of a seizure of the patient. In at least one aspect, the fourth model 414 may also receive LTB data as the input 430, such as LTB data included in, and corresponding to, the LTBSC data. The fourth model training data 420 may include, for a plurality of historical patients, historical LTBSC data indicating historical seizures, and / or any other suitable training data. The fourth model 414 may be trained to determine, for the historical patients, associations between an historical LTBSC and historical long-term biometrics, with historical seizures of historical patients, and / or any other suitable patterns or associations. For example, patterns and associations in LTBSCs and long-term biometrics may provide an indication of when aseizure may next occur. The fourth model 414 may generate as an output 440 fourth model output data 444 indicating a prediction of the seizure of the patient. The fourth model output data 444 may include includes a date of the prediction of the seizure, and / or a confidence metric associated with the prediction, and / or any other suitable information associated with the predictions of a seizure.

[0180] In one example, a server may store the fourth model 414 and LTBSC data of the patient in a memory. The server may provide the LTBSC data to the fourth model 414 to generate the fourth model output data 444 predicting the patient’s next date for a seizure. The server may provide the fourth model output data 444 to a smartphone of the patient, to be processed by a mobile app and displayed on a screen of the smartphone, such that the mobile app may display date of the seizure prediction.FIFTH MODEL

[0181] In one aspect, the models 410 may include a fifth model 415 to determine one or more times to attempt a medication titration, such as optimal days which may minimize the chance of side-effects from the drug titration. The fifth model 415 may be trained, using training data 420 (referred to as fifth model training data), to receive the LTBSC data as an input 430, and to generate as the output 440 the fifth model output data 445 indicating at least one date to implement a medication titration of the patient. The fifth model training data 420 may include historical LTBSC data of historical patients, and / or any other suitable training data. The fifth model 415 may be trained to determine, for the historical patients, associations between at least one historical LTBSC, timing of an historical medication titration, and associated historical side effects from the historical medication titration and / or any other suitable patterns or associations. The fifth model 415 may generate as an output 440 fifth model output data 445 indicating at least one date to implement a medication titration of the patient.

[0182] In one example, a server may store the fifth model 415 and LTBSC data of the patient. The server may provide the LTBSC data to the fifth model 415 to generate the fifth model output data 445 predicting the patient’s next date for a seizure. The server may provide the fifth model output data 445 to a smartphone of the patient, to be processed by a mobile app and display the one or more optimal dates to attempt a medication titration on the screen of the smartphone.SIXTH MODEL

[0183] In one aspect, the models 410 may include a sixth model 416 to determine associations between environmental characteristics and the LTBSCs of the patient. The sixth model 416 may be trained using training data 420 (referred to as sixth model training data) to receive the LTBSC data, and environmental data associated with the LTBSC data, as an input 430, and to generate, as the output 440, the sixth model output data 446 indicating an association between at least oneenvironmental characteristic and the LTBSC of the patient. In at least one aspect, the environmental data is received in response to a LTBSC of the patient, as previously described. The sixth model training data 420 may include historical LTBSC data of historical patients and historical environmental data of the historical patients, and / or any other suitable training data. The sixth model 416 may be trained to determine, for the historical patients, associations between at least one historical LTBSC and at least one historical environmental characteristic, and / or any other suitable patterns or associations. The sixth model 415 may generate as an output 440 sixth model output data 446 indicating associations between environmental characteristics of the patient and the LTBSCs of the patient. The sixth model output data 446 may indicate a cause of the LTBSC, a prediction of a future LTBSC in response to the at least one environmental characteristic, and / or a treatment pathway to target a desired brain state, and / or any other suitable information.

[0184] In one example, the environmental data may be generated by the implant device of the patient, e.g., via the user interface 306, etc., as previously described, and / or any other suitable device, such as via the user device 1 15. In one example, the environmental data may be generated by a user device, such as a computing device of the user which runs a program for the user to provide environmental data, such as when the patient takes, however, the environmental data may be generated in any suitable manner. The mobile app, may provide the environmental data to the sixth model 415 tunning on a server. The sixth model may receive the environmental data, as well as LTBSC data already generated and existing on the server, to generate the sixth model output data 446. The sixth model output data 446 may indicate an association between when the patient takes their medication, and their LTBSC. The sixth model output data 446 may further indicate a prediction of a future LTBSC (e.g., when the LTBSC will occur). The server may provide the sixth model output data 446 to the user’s mobile device, so they may review the association and prediction.

[0185] While various ML models (e.g., models 411 -416) have been described with respect to Fig. 4 and more generally throughout the present disclosure, a single ML model may provide at least some functionality associated with one or more separately described ML models. For example, a third model may provide the functionality of the first and third models. Additionally, the functionality described with respect to a single ML model may be carried out by one or more separate ML models. Moreover, one or more ML models may have additional functionality provided by the training data which may not be expressly described. Furthermore, although the each of the models 41 1 -416 are described as generating specific outputs based upon receiving specific input data, the models 41 1 -416 may receive other input data and produced other outputs. For example, during a seizure, the patient may indicate that they are experiencing a seizure via the user interface 306 ofthe implant device, and as a result the implant device may generate patient reported seizure data. This data may be provided to the second model 411 as an input, and may be operated on by the second model in addition to the LTBSC data to generate the second model output data indicating a seizure of the patient. The considerations just described likewise apply to static models, which may not be ML models.

[0186] Fig. 5 is a block diagram of an example implant device system 500 similar to the system 300 of Fig. 3E, but which includes one or more trained ML models 380 (e.g., such as ML model(s) 130, 410-416) instead of the model 370 based on a static algorithm (e.g., a static model). That is, Fig. 5 corresponds generally to Fig. 3E, with the only difference between the implant device system 300 and the implant device system 500 in respective figures being the inclusion of the trained ML models 380, including models 380A-380F such as models 41 1 -416 respectively, rather than the model 370A-370F based on a static algorithm. The implant device system 500, as depicted in Fig. 5, is the same in all respects as in Fig. 3E, above, except that the trained ML models 370 are created using ML algorithms to search for patterns in training data and, upon implementation in the processor device 304, the trained models 370 may receive an input and determine (e.g., from one or more feature values) an output, as previously described.

[0187] Fig. 6 is a block diagram depicting another example embodiment, in which generating the LTBSC data 375A, the second model output data 375B, the third model output data 375C, the fourth model output data 375D, the fifth model output data 375E, and / or the sixth model output data 375F may take place on a device other than the processor device 304 and, specifically, on an external device 504. In the embodiments depicted in Fig. 6, it is contemplated that the one or more models may be either the static models 370 or the trained ML models 380 and, as a result, Fig. 6 illustrates an alternate embodiment of Fig. 3E and of Fig. 5. In the embodiments contemplated within Fig. 6, the processor device 304 generally collects the LTEEG data 372 from the sensor array 302. The LTEEG data 372 is stored in the memory 360 of the processor device 304. While the processor device 304 may be equipped to perform the modeling - that is may have stored in the memory 360 the models 370 or 380, and may be configured to operate on the input data to generate the output data associated with the models 370, 380 - in the embodiments contemplated by Fig. 6, this functionality is optional. Instead, the microprocessor 358 may be configured to communicate with the external device 504 such that the external device 504 may perform the analysis.

[0188] The external device 504 may be a server such as server 105, a user device such as user device 1 15, a computing device such as computing device 215, a workstation, a cloud computing platform, or any other suitable device, configured to receive data from the one or more processordevices 304 associated with one or more respective patients. The external device 504 may include communication circuitry 556, coupled to a processor 558 that, in turn, is coupled to a memory 560. The processor 558 may be any known processor configurable to execute the routines necessary for generating the LTBSC data 375A, the second model output data 375B, the third model output data 375C, the fourth model output data 375D, the fifth model output data 375E, and / or the sixth model output data 375F, including, by way of example and not limitation, general purpose microprocessors (GPUs), RISC microprocessors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or any other suitable processor, such as the processor 120, 248, and the like.

[0189] The communication circuitry 556 may be any transceiver and / or receiver / transmitter pair that facilitates communication with the various devices from or to which the processor device 504 receives data and / or transmits data, such as communication circuitry 122. The communication circuitry 556 is coupled to the processor 558, which, in addition to executing various routines and instructions for performing analysis, may also facilitate storage in the memory 560 of data received, via the communication circuity 556, from the processor devices 304 of the one or more patients.

[0190] The memory 560 may include one or more memories, such as the memory 125 and / or the database 126. The memory 560 may include both volatile memory (e.g., random access memory (RAM)) and non-volatile memory, in the form of magnetic and / or solid-state media. In addition to an operating system (not shown), the memory 560 may store data received from the processor devices 304, such as received data 581 which may include the LTEEG data 372, and / or any other suitable data.

[0191] Like the processor device 304, the external device 504 may have, stored in its memory 560, one or more static models 370 and / or trained ML models 380, such as the first models 370A, 380A, the second models 370B, 380B, the third models 370C, 380C, the fourth models 370D, 380D, the fifth models 370D, 380D, and the sixth models 370E, 380E. The processor 558 may execute one or more of the models 370, 380, receiving as inputs one or more of the LTEEG data 372, the environmental data 374, the LTBSC data 375A, the LTB data 375AA, and / or any other suitable data, as previously described with respect to the static models 370 of Fig. 3E and / or ML models 380 of Fig. 5.

[0192] The embodiments depicted in Fig. 6 also contemplate that, even in embodiments in which the processor device 304 executes the models 370 or 380 to produce results 362A, 362B, 362C, the processor device 304 may communicate the outputs / results 375A-375F, as well as the data 372, 374, 375AA upon which the outputs / results are based, to the external device 504. The external device 504 may receive such data, and may store the data, for later viewing or analysis bythe patient(s), physicians, or others, etc. In embodiments in which the external device 504 performs analysis of data of multiple patients, or for which the external device 504 receives from multiple processor devices 304 data of multiple patients, the external device 504 may store the received data 581 , the LTBSC data 375A, the second model output data 375B, the third model output data 375C, the fourth model output data 375D, the fifth model output data 375E, and / or the sixth model output data 375F, and / or any other suitable data for each patient separately in the memory 560.EXAMPLE METHOD FOR TRAINING A FIRST MODEL TO GENERATE LTBSC DATA

[0193] Fig. 7A is a flow chart depicting a computer-implemented method 700 for training a first model to generate LTBSC data indicating at least one LTBSC of a patient. The first model, including the first iteration, second iteration, or other iterations of the first model, may be trained to determine, for the historical patients, associations between an historical LTBSC and at least one historical LTEEG signal corresponding to the historical LTBSC.

[0194] The method 700 may include obtaining, by one or more processors, a first training dataset (block 701 ).

[0195] The method 700 may include receiving as feature values a selection of one or more attributes of the first training dataset (block 702). The one or more attributes may comprise one or more characteristics of an electroencephalogram (EEG) signal. A training dataset may comprise historical long-term EEG (LTEEG) data of historical patients and historical LTBSC data of historical patients corresponding to the historical LTEEG data.

[0196] The method 700 may include training, by the one or more processors, a first iteration of the first model using the first training dataset (block 703) using the feature values defined for the first training dataset to generate first primary first model output data (block 704). The primary first model output data may have associated first error rates.

[0197] The method 700 may include obtaining, by one or more processors, a second training dataset (block 705). The method 700 may include receiving as feature values the selection of one or more attributes of the second training dataset (block 706).

[0198] The method 700 may include training, by the one or more processors, a second iteration of the first model using the second training dataset (block 707) using the feature values defined for the second training dataset to generate secondary first model output data (block 708). The secondary first model output data may have associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

[0199] In one aspect of the method 700, the LTBSC data may include at least a portion of the LTEEG data corresponding to the at least one LTBSC.

[0200] In one aspect of the method 700, the training dataset may further comprise historical LTB data of historical patients corresponding to the historical LTBSC data of the historical patients; and the first model may be further trained to determine, for the historical patients, associations between the historical LTBSC and historical long-term biometrics corresponding to the historical LTBSC.

[0201] In one aspect of the method 700, the LTBSC data may include at least a portion of the LTB data corresponding to the LTBSC data.

[0202] In one aspect, the method 700 may further include obtaining, by the one or more processors, LTEEG data associated with one or more patients; and based upon the LTEEG data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

[0203] In one aspect, the method 700 may further include receiving one or more labels for the first training dataset, wherein training the first iteration of the first model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the first LTBSC data; and receiving one or more labels for the second training dataset, wherein training the second iteration of the first model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the second LTBSC data.EXAMPLE METHOD FOR TRAINING A SECOND MODEL TO GENERATE SECOND MODEL OUTPUT DATA

[0204] Fig. 7B is a flow chart depicting a computer-implemented method 710 for training a second model to generate second model output data indicating a seizure of the patient. The second model, including the first iteration, second iteration, or other iterations of the second model, may be trained to determine, for the historical patients, associations between at least one historical LTBSC and an historical seizure.

[0205] The method 710 may include obtaining, by one or more processors, a first training dataset (block 71 1 ). a training dataset may comprise historical LTBSC data of historical patients.

[0206] The method 710 may include receiving as feature values a selection of one or more attributes of the first training dataset (block 712). The one or more attributes may comprise one or more characteristics of an EEG signal (e.g., epileptiform activity, spikes, sharps, etc.).

[0207] The method 710 may include training, by the one or more processors, a first iteration of the second model using the first training dataset (block 713) using the feature values defined for the first training dataset to generate first primary second model output data (block 714). The primary second model output data may have associated first error rates.

[0208] The method 710 may include obtaining, by one or more processors, a second training dataset (block 715). The method 710 may include receiving as feature values the selection of one or more attributes of the second training dataset (block 716).

[0209] The method 710 may include training, by the one or more processors, a second iteration of the second model using the second training dataset (block 717) using the feature values defined for the second training dataset to generate secondary second model output data (block 718). The secondary second model output data may have associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates. The second model output data may include at least a portion of LTBSC data corresponding to the seizure.

[0210] In one aspect of the method 710, the training dataset may further comprise historical longterm biometric (LTB) data corresponding to the historical LTBSC data of the historical patients; and the second model may be further trained to determine, for the historical patients, associations between the historical LTBs, and the corresponding historical LTBSCs and / or historical seizures; and

[0211] In one aspect of the method 710, the LTBSC data may include at least a portion of the LTB data corresponding to the LTBSC data.

[0212] In one aspect, the method 710 may include obtaining, by the one or more processors, LTBSC data associated with one or more patients; and based upon the LTBSC data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

[0213] In one aspect, the method 710 may include receiving one or more labels for the first training dataset, wherein training the first iteration of the second model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the primary second model output data; and receiving one or more labels for the second training dataset, wherein training the second iteration of the second model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the secondary second model output data.EXAMPLE METHOD FOR TRAINING A THIRD MODEL TO GENERATE THIRD MODEL OUTPUT DATA

[0214] Fig. 7C is a flow chart depicting a computer-implemented method 720 for training a third model to generate third model output data, the third model output data including a revision of an indication of a seizure in the long-term EEG data of the LTBSC data. The third model, including the first iteration, second iteration, or other iterations of the third model, may be trained to determine, forthe historical patients, associations between an historical seizure and at least one historical LTBSC. The revision may include a reclassification of the seizure and / or a change in a confidence metric associated with the indication of the seizure.

[0215] The method 720 may include obtaining, by one or more processors, a first training dataset (block 721 ). A training dataset may comprise historical LTBSC data of historical patients.

[0216] The method 720 may include receiving as feature values a selection of one or more attributes of the first training dataset (block 722). The one or more attributes may comprise one or more characteristics of an EEG signal.

[0217] The method 720 may include training, by the one or more processors, a first iteration of the third model using the first training dataset (block 723) using the feature values defined for the first training dataset to generate first primary third model output data (block 724). The primary third model output data may have associated first error rates.

[0218] The method 720 may include obtaining, by one or more processors, a second training dataset (block 725). The method 720 may include receiving as feature values the selection of one or more attributes of the second training dataset (block 726).

[0219] The method 720 may include training, by the one or more processors, a second iteration of the third model using the second training dataset (block 727) using the feature values defined for the second training dataset to generate secondary third model output data (block 728). The secondary third model output data may have associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

[0220] In one aspect of the method 720, the training dataset may further comprise historical longterm biometric (LTB) data corresponding to the historical LTBSC data of the historical patients; and the third model may be further trained to determine, for the historical patients, associations between the historical LTBs, and the corresponding historical LTBSCs and / or historical indications of seizures.

[0221] In one aspect of the method 720, the LTBSC data may include at least a portion of the LTB data corresponding to the LTBSC data.

[0222] In one aspect, the method 720 further include obtaining, by the one or more processors, the LTBSC data indicating seizures of patients; and based upon the LTBSC data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

[0223] In one aspect, the method 720 may include receiving one or more labels for the first training dataset, wherein training the first iteration of the third model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the primary third model output data; and receiving one or more labels for the second training dataset, whereintraining the second iteration of the third model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the secondary third model output data.EXAMPLE METHOD FOR TRAINING A FOURTH MODEL TO GENERATE FOURTH MODEL OUTPUT DATA

[0224] Fig. 7D is a flow chart depicting a computer-implemented method 730 for training a fourth model to generate fourth model output data indicating a prediction of a seizure of the patient. The fourth model, including the first iteration, second iteration, or other iterations of the fourth model, may be trained to determine, for the historical patients, associations between at least one historical LTBSC and timing of an historical seizure. The fourth model output data may include a date of the prediction of the seizure, and / or a confidence metric associated with the prediction.

[0225] The method 730 may include obtaining, by one or more processors, a first training dataset (block 731 ). A training dataset may comprise historical LTBSC data of historical patients.

[0226] The method 730 may include receiving as feature values a selection of one or more attributes of the first training dataset (block 732). The one or more attributes may comprise one or more characteristics of an EEG signal.

[0227] The method 730 may include training, by the one or more processors, a first iteration of the fourth model using the first training dataset (block 733) using the feature values defined for the first training dataset to generate first primary fourth model output data (block 734). The primary fourth model output data may have associated first error rates.

[0228] The method 730 may include obtaining, by one or more processors, a second training dataset (block 735). The method 730 may include receiving as feature values the selection of one or more attributes of the second training dataset (block 736). In one aspect of the method 730 the first training dataset does not include long-term based brain state change (LTBSC) data, and the second training dataset includes the LTBSC data, the first and second training datasets being used to fine-tune the fourth model.

[0229] The method 730 may include training, by the one or more processors, a second iteration of the fourth model using the second training dataset (block 737) using the feature values defined for the second training dataset to generate secondary fourth model output data (block 738). The secondary fourth model output data may have associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

[0230] In one aspect of the method 730, the training dataset may further comprise historical longterm biometric (LTB) data corresponding to the historical LTBSC data of the historical patients; andthe fourth model may be further trained to determine, for the historical patients, associations between the historical LTBs, and the corresponding historical LTBSCs and / or historical predictions of seizures.

[0231] In one aspect of the method 730, the LTBSC data may include at least a portion of the LTB data corresponding to the LTBSC data.

[0232] In one aspect, the method 730 may include obtaining, by the one or more processors, LTBSC data associated with one or more patients; and based upon the LTBSC data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

[0233] In one aspect, the method 730 may include receiving one or more labels for the first training dataset, wherein training the first iteration of the fourth model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the primary fourth model output data; and receiving one or more labels for the second training dataset, wherein training the second iteration of the fourth model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the secondary fourth model output data.EXAMPLE METHOD FOR TRAINING A FIFTH MODEL TO GENERATE FIFTH MODEL OUTPUT DATA

[0234] Fig. 7E is a flow chart depicting a computer-implemented method 740 for training a fifth model to generate fifth model output data indicating at least one date to implement a medication titration of the patient. The fifth model, including the first iteration, second iteration, or other iterations of the fifth model, may be trained determine, for the historical patients, associations between at least one historical LTBSC, timing of an historical medication titration, and associated historical side effects from the historical medication titration.

[0235] The method 740 may include obtaining, by one or more processors, a first training dataset (block 741 ). A training dataset may comprise historical LTBSC data of historical patients.

[0236] The method 740 may include receiving as feature values a selection of one or more attributes of the first training dataset (block 742). The one or more attributes may comprise one or more characteristics of an EEG signal.

[0237] The method 740 may include training, by the one or more processors, a first iteration of the fifth model using the first training dataset (block 743) using the feature values defined for the first training dataset to generate first primary fifth model output data (block 744). The primary fifth model output data may have associated first error rates.

[0238] The method 740 may include obtaining, by one or more processors, a second training dataset (block 745). The method 740 may include receiving as feature values the selection of one or more attributes of the second training dataset (block 746).

[0239] The method 740 may include training, by the one or more processors, a second iteration of the fifth model using the second training dataset (block 747) using the feature values defined for the second training dataset to generate secondary fifth model output data (block 748). The secondary fifth model output data may have associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

[0240] In one aspect of the method 740, the training dataset may further comprise historical longterm biometric (LTB) data corresponding to the historical LTBSC data of the historical patients; and the fifth model may be further trained to determine, for the historical patients, associations between the historical LTBs, and one or more of the corresponding historical LTBSCs, historical times in which medication titrations were implemented, and associated historical side effects.

[0241] In one aspect of the method 740, the LTBSC data may include at least a portion of the LTB data corresponding to the LTBSC data.

[0242] In one aspect, the method 740 may further include obtaining, by the one or more processors, the LTBSC data; and based upon the LTBSC data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

[0243] In one aspect, the method 740 may further include receiving one or more labels for the first training dataset, wherein training the first iteration of the fifth model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the primary fifth model output data; and receiving one or more labels for the second training dataset, wherein training the second iteration of the fifth model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the secondary fifth model output data.EXAMPLE METHOD FOR TRAINING A SIXTH MODEL TO GENERATE SIXTH MODEL OUTPUT DATA

[0244] Fig. 7F is a flow chart depicting a computer-implemented method 750 for training a sixth model to generate sixth model output data indicating an association between at least one environmental characteristic of the patient and the at least one LTBSC of the patient. The sixth model, including the first iteration, second iteration, or other iterations of the sixth model, may be trained to determine, for the historical patients, associations between at least one historical LTBSC and at least one historical environmental characteristic.

[0245] The method 750 may include obtaining, by one or more processors, a first training dataset (block 750). A training dataset may comprise historical LTBSC data of historical patients and historical environmental data of the historical patients.

[0246] The method 750 may include receiving as feature values a selection of one or more attributes of the first training dataset (block 751 ). The one or more attributes may comprise one or more characteristics of an EEG signal.

[0247] The method 750 may include training, by the one or more processors, a first iteration of the sixth model using the first training dataset (block 752) using the feature values defined for the first training dataset to generate first primary sixth model output data (block 754). The primary sixth model output data may have associated first error rates.

[0248] The method 750 may include obtaining, by one or more processors, a second training dataset (block 755). The method 750 may include receiving as feature values the selection of one or more attributes of the second training dataset (block 756).

[0249] The method 750 may include training, by the one or more processors, a second iteration of the sixth model using the second training dataset (block 757) using the feature values defined for the second training dataset to generate secondary sixth model output data (block 758). The secondary sixth model output data may have associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

[0250] In one aspect of the method 750, the training dataset may further comprise historical longterm biometric (LTB) data corresponding to the historical LTBSC data of the historical patients; and the sixth model may be further trained to determine, for the historical patients, associations between the historical LTBs, and the corresponding historical LTBSCs and / or historical environmental characteristics.

[0251] In one aspect of the method 750, the LTBSC data may include at least a portion of the LTB data corresponding to the LTBSC data.

[0252] In one aspect, the method 750 may further include obtaining, by the one or more processors, the LTBSC data; and based upon the LTBSC data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

[0253] In one aspect, the method 750 may further include receiving one or more labels for the first training dataset, wherein training the first iteration of the sixth model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the primary sixth model output data; and receiving one or more labels for the second training dataset, wherein training the second iteration of the sixth model using the second training dataset furthercomprises using the one or more labels for the second training dataset to generate the secondary sixth model output data.EXAMPLE METHOD FOR DETERMINING A LTBSC OF A PATIENT

[0254] Fig. 8 is a flow chart depicting a computer-implemented method 800 for determining a LTBSC of a patient. The method 800 may include obtaining, by one or more processors, LTEEG data comprising at least one LTEEG signal associated with a brain of the patient and gathered via an implant device of the patient (block 810). The LTEEG data may include at least one substantially continuous LTEEG signal of the patient gathered over more than a week of time.

[0255] The method 800 may include providing, by the one or more processors, the LTEEG data to a first model trained using first model training data to generate LTBSC data indicating at least one LTBSC of the patient (block 820). The first model training data may include historical LTEEG data of historical patients and historical LTBSC data of historical patients corresponding to the historical LTEEG data. The first model may be trained to determine, for the historical patients, associations between an historical LTBSC and at least one historical LTEEG signal corresponding to the historical LTBSC. The LTBSC data may include at least a portion of the LTEEG data corresponding to the at least one LTBSC, and / or a classification and / or confidence metric associated with the at least one LTBSC. The at least one LTBSC may be associated with one or more of epilepsy, a psychiatric disorder, a stroke, Alzheimer's disease, a sleep disorder, depression, medication efficacy, or a medication side effect.

[0256] The method 800 may include, in response to generating the LTBSC data, one or more of: (i) retraining, by the one or more processors, one or more machine learning models using the LTBSC data; (ii) providing, by the one or more processors, the LTBSC data as an input to the one or more machine learning models; or (iii) generating, by the one or more processors, a medical insight of the patient based upon the LTBSC data; and providing, by the one or more processors, the medical insight to a user device (block 830).

[0257] In at least one aspect of the method 800, generating the LTBSC data may include one or more of: (i) analyzing, by the one or more processors, at least one frequency band of the at least one LTEEG signal; (ii) identifying, by the one or more processors, at least one LTEEG segment in the at least one LTEEG signal based upon one or more of a root means square, a standard deviation, or a power spectrum density, of the at least one LTEEG signal; (iii) determining, by the one or more processors, one or more feature values of the at least one LTEEG segment by applying a principal components analysis algorithm or a support vector machine to the LTEEG data; (iv) determining, by the one or more processors, at least one similar characteristic of the LTEEGdata and / or the one or more feature values, across long-duration time points; (v) classifying, by the one or more processors, LTEEG segments having similar characteristics; or (vi) determining, by the one or more processors, an LTBSC boundary based upon detecting an edge of a phase transition of the at least one LTEEG signal. The one or more feature values may be associated with one or more characteristics of an EEG signal, such as EEG spikes and / or EEG sharps. The at least one frequency band may be selected from the group consisting of 1 to 2 hertz (Hz), 0.5 to 4 Hz, 4 to 8 Hz, 8 to 13 Hz, 13 to 30 Hz, 30 to 100 Hz, 80 to 250 Hz, and 250 to 500 Hz.

[0258] In one aspect, the method 800 may include obtaining, by the one or more processors, long-term biometric (LTB) data indicating long term biometrics of the patient corresponding to the at least one LTBSC; and providing, by the one or more processors, the LTB data to the first model, wherein generating the LTBSC data is further based upon the LTB data. The first model may determine one or more feature values of the LTB data, such as biometrics of the face and / or eyes. The LTBs may indicate one or more of a voice biomarker, eye attentiveness, or a facial change. In such an aspect: (i) the first model training data may further include historical LTB data of historical patients corresponding to the historical LTBSC data of the historical patients; (ii) the first model may be further trained to determine, for the historical patients, associations between the historical LTBSC and historical long-term biometrics corresponding to the historical LTBSC; and / or (iii) the LTBSC data may include at least a portion of the LTB data corresponding to the LTBSC data. Retraining the one or more one machine learning models using the LTBSC data may include fine- tuning the one or more machine learning models. The medical insight may be one or more of: (i) an indication of the at least one LTBSC; (ii) a risk of sudden unexplained death by epilepsy; or (iii) an association between the at least one LTBSC and the long-term biometrics.

[0259] In one aspect, the method 800 may further include (i) obtaining, by the one or more processors, additional first model training data; (ii) retraining, by the one or more processors, the first model using the additional first model training data; and (iii) storing, by the one or more processors, the retrained first model on a memory to generate subsequent LTBSC data using the retrained first model.

[0260] In one aspect of the method 800, (i) the one or more machine learning models may include a second model trained using second model training data to generate second model output data indicating a seizure of the patient in the LTEEG of the LTBSC data, based upon receiving the LTBSC data; (ii) retraining the second model using the LTBSC data may include storing, by the one or more processors, the retrained second model in the memory to generate subsequent second model output data using the retrained second model; and (iii) providing the LTBSC data as the input to the second model may include, in response to generating the second model output data,providing, by the one or more processors, the second model output data to the user device. In such an aspect, the second model training data may include historical LTBSC data of historical patients, and / or the second model may be trained to determine, for the historical patients, associations between at least one historical LTBSC and an historical seizure. The second model output data may include a classification of the seizure and / or a confidence metric associated with the indication of the seizure. In such an aspect, the method 800 may further include (i) obtaining, by the one or more processors, additional second model training data; (ii) retraining, by the one or more processors, the second model using the additional second model training data; and (iii) storing, by the one or more processors, the retrained second model in the memory to generate subsequent second model output data using the retrained second model.

[0261] In one aspect of the method 800, (i) the one or more machine learning models may include a third model trained using third model training data to generate third model output data, based upon receiving the LTBSC data, wherein: (a) the third model output data may include a revision of an indication of a seizure in the LTEEG data of the LTBSC data; (b) the indication of the seizure is not based upon the at least one LTBSC of the LTBSC data; and (c) the revision to the indication of the seizure is based upon the at least one LTBSC of the LTBSC data; (ii) retraining the third model using the LTBSC data may include storing, by the one or more processors, the retrained third model in the memory to generate subsequent third model output data using the retrained third model; and (iii) providing the LTBSC data as the input to the third model may include, in response to generating the third model output data, providing, by the one or more processors, the third model output data to the user device. The third model training data may include historical LTBSC data of historical patients, and / or the third model may be trained to determine, for the historical patients, associations between an historical seizure and at least one historical LTBSC. The revision may include a reclassification of the seizure and / or a change in a confidence metric associated with the indication of the seizure. In such an aspect, the method 880 may include (i) obtaining, by the one or more processors, additional third model training data; (ii) retraining, by the one or more processors, the third model using the additional third model training data; and (iii) storing, by the one or more processors, the retrained third model in the memory to generate subsequent third model output data using the retrained third model.

[0262] In one aspect of the method 800, (i) the one or more machine learning models may include a fourth model trained using fourth model training data to generate fourth model output data indicating a prediction of a seizure of the patient, based upon receiving the LTBSC data; (ii) retraining the fourth model using the LTBSC data may include storing, by the one or more processors, the retrained fourth model in the memory to generate subsequent fourth model outputdata using the retrained fourth model; and (iii) providing the LTBSC data as the input to the fourth model may include, in response to generating the fourth model output data, providing, by the one or more processors, the fourth model output data to the user device. The fourth model training data may include historical LTBSC data of historical patients, and / or the fourth model may be trained to determine, for the historical patients, associations between at least one historical LTBSC and timing of an historical seizure. The fourth model output data may include at least one date of the prediction of the seizure, and / or a confidence metric associated with the prediction. In such an aspect, the method 800 may include (i) obtaining, by the one or more processors, additional fourth model training data; (ii) retraining, by the one or more processors, the fourth model using the additional fourth model training data; and (iii) storing, by the one or more processors, the retrained fourth model in the memory to generate subsequent fourth model output data using the retrained fourth model.

[0263] In one aspect of the method 800, (i) the one or more machine learning models may include a fifth model trained using fifth model training data to generate fifth model output data indicating at least one date to implement a medication titration of the patient, based upon receiving the LTBSC data; (ii) retraining the fifth model using the LTBSC data may include storing, by the one or more processors, the retrained fifth model in the memory to generate subsequent fifth model output data using the retrained fifth model; and (iii) providing the LTBSC data as the input to the fifth model may include, in response to generating the fifth model output data, providing, by the one or more processors, the fifth model output data to the user device. The fifth model training data may include historical LTBSC data of historical patients, and / or the fifth model may be trained to determine, for the historical patients, associations between at least one historical LTBSC, timing of an historical medication titration, and associated historical side effects from the historical medication titration. In such an aspect, the method 800 may further include (i) obtaining, by the one or more processors, additional fifth model training data; (ii) retraining, by the one or more processors, the fifth model using the additional fifth model training data; and (iii) storing, by the one or more processors, the retrained fifth model in the memory to generate subsequent fifth model output data using the retrained fifth model.

[0264] In one aspect of the method 800, (i) the one or more machine learning models may include a sixth model trained using sixth model training data to generate sixth model output data indicating an association between at least one environmental characteristic of the patient and the at least one LTBSC of the patient, based upon receiving environmental data indicating the at least one environmental characteristic of the patient and the LTBSC data; (ii) retraining the sixth model using the LTBSC data may include storing, by the one or more processors, the retrained sixth model inthe memory to generate subsequent sixth model output data using the retrained sixth model; and (iii) providing the LTBSC data as the input to the sixth model may include (a) obtaining, by the one or more processors, the environmental data associated with the LTBSC data; (b) providing, by the one or more processors, the environmental data and the LTBSC data to the sixth model to generate the sixth model output data; and (c) in response to generating the sixth model output data, providing, by the one or more processors, the sixth model output data to the user device. The sixth model training data may include historical LTBSC data of historical patients and historical environmental data of the historical patients, and / or the sixth model may be trained to determine, for the historical patients, associations between at least one historical LTBSC and at least one historical environmental characteristic.

[0265] In one aspect of the method 800, the at least one environmental characteristic may be associated with a medication. The association with the medication may include a new medication taken by the patient, and / or a new titration of the medication taken by the patient. The environmental data may be obtained in response to a LTBSC of the patient; and the sixth model output data may indicate a cause of the LTBSC. The sixth model output data may indicate a prediction of a future LTBSC in response to the at least one environmental characteristic, and / or a treatment pathway to a target brain state. The environmental data may be generated by an implant device of the patient. The LTBSC data may include LTB data corresponding to the at least one LTBSC. In such an aspect, the method 800 may include (i) obtaining, by the one or more processors, additional sixth model training data; (ii) retraining, by the one or more processors, the sixth model using the additional sixth model training data; and (iii) storing, by the one or more processors, the retrained sixth model in the memory to generate subsequent sixth model output data using the retrained sixth model.

[0266] As may by now be understood, the presently disclosed method and system are amenable to a variety of embodiments, some of which have already been described herein.

[0267] Fig. 9 depicts an alternate embodiment, in which the sensor array 302 and the processor device 304 are integrated into a single unit 910. The combined unit 910 includes the battery 348 and battery charging technology 349 for powering the unit 910, as previously described. The battery charging technology 349 may include a battery charging circuit, for facilitating charging of the battery 348. The battery charging technology 349 may be any known battery charging technology compatible with the arrangement of the sensor array 302. In particular, in embodiments the battery technology 349 may be an inductive charging circuit that facilitates charging through the patient’s skin when the sensor array 302 is disposed beneath the scalp of the patient. In other embodiments, the battery technology 349 may draw energy from the movements of the patientthroughout the day by, for example, harnessing the movements of the patient to turn a small generator. In still other embodiments, the battery technology 349 may draw power from the environment in the form of RF signals. In further examples still, the battery technology 349 may draw power from chemical reactions taking place in the environment of the sensor array 302. Of course, more traditional charging methods (e.g., using a wired connection to provide power to the battery technology 349) may also be employed.

[0268] The single unit 910 includes electrode devices 310. As in previously described embodiments, the unit 910 may include one or more sensors 351 . Additionally, the unit 910 may, as previously described, be communicatively coupled to one or more sensors 351 , that are external to the unit 910. The amplifier 346 and analog-to-digital converter (ADC) 352 are also included. The microprocessor 358, memory 360, and communication circuitry 356 function as described throughout. The single unit 910 may be connected to external equipment 905. The external equipment 905 may be one or more servers, such as server 105, or other suitable external equipment, that may receive and store the data for individual patients and / or communicate the data for the patients to the respective medical personnel or physicians diagnosing and / or treating the patients.

[0269] Various communication schemes are contemplated, as well. Figs. 10A and 10B illustrate possible communication schemes between the sensor array 302 and the processor device 304 and, in particular, Fig. 10A illustrates a wireless connection 1002 between the sensor array 302 and the processor device 304 (i.e. , between the communication circuitry of the sensor array 302 and the communication circuitry 356 of the processor device 304). The wireless connection 1002 may be any known type of wireless connection, including a Bluetooth® connection (e.g., low-energy Bluetooth), a wireless internet connect (e.g., IEEE 802.11 , known as “Wi-Fi”), a near-field communication connection, or similar. Fig. 10B illustrates a wired connection 1004 between the sensor array 302 and the processor device 304. The wired connection may be a serial connection, for example.

[0270] The sensor array 302 may communicate data to the processor device 304 as the data are acquired by the sensor array 302 or periodically. For example, the sensor array 302 may store, in the memory 355 of the local processing unit 344, data as it is acquired from the electrode devices 310, and may periodically (e.g., every second, every minute, every half hour, every hour, every day, when the memory 355 is full, etc.) transmit the data to the processor device 304. In other embodiments, the sensor array 302 may store data until the processor device 304 is coupled to the sensor array 302 (e.g., via wireless or wired connection). The sensor array 302 may also store the data until the processor device 304 requests the transmission of the data from the sensor array 302to the processor device 304. In these manners, the sensor array 302 may be optimized, for example, to preserve battery life, etc.

[0271] Figs. 11 A-1 1 C illustrate possible communication schemes between the processor device 304 and external equipment or servers 905, regardless of whether or not the processor device 304 is integrated with the sensor array 302 (e.g., as in Fig. 9). In Fig. 11 A, for example, the processor device 304 may be coupled by a wireless communication connection to a mobile device 915, such as user device 1 15 or computing device 215, which may, in turn, be coupled to the external equipment 905 by, for example, the Internet. In Fig. 1 1 B, the processor device 304 is coupled to one or more intermediary devices 920 (e.g., a mobile telephony base station, a wireless router, etc.), which in turn provides connectivity to the external equipment 905 via the Internet. In Fig. 1 1 C, the processor device 304 is itself a mobile device, such as a mobile telephony device, which may be coupled by one or more intermediary devices 920 to the external equipment 905 by way of the Internet.

[0272] The following list of aspects reflects a variety of the embodiments explicitly contemplated by the present disclosure. Those of ordinary skill in the art will readily appreciate that the aspects below are neither limiting of the embodiments disclosed herein, nor exhaustive of all of the embodiments conceivable from the disclosure above, but are instead meant to be exemplary in nature.

[0273] 1 . A computer-implemented method for determining a long-term brain state change (LTBSC) of a patient, the method comprising: obtaining, by one or more processors, long-term electroencephalogram (LTEEG) data comprising at least one LTEEG signal associated with a brain of the patient and gathered via an implant device of the patient; providing, by the one or more processors, the LTEEG data to a first model trained using first model training data to generate LTBSC data indicating at least one LTBSC of the patient; in response to generating the LTBSC data, one or more of: (i) retraining, by the one or more processors, one or more machine learning models using the LTBSC data; (ii) providing, by the one or more processors, the LTBSC data as an input to the one or more machine learning models; or (iii) generating, by the one or more processors, a medical insight of the patient based upon the LTBSC data; and providing, by the one or more processors, the medical insight to a user device.

[0274] 2. The computer-implemented method according to aspect 1 , wherein the LTEEG data includes at least one substantially continuous LTEEG signal of the patient gathered over more than a week of time.

[0275] 3. The computer-implemented method according to aspect 1 or 2, wherein: the first model training data comprises historical LTEEG data of historical patients and historical LTBSC data of historical patients corresponding to the historical LTEEG data.

[0276] 4. The computer-implemented method according to any one of the preceding aspects, wherein: the first model is trained to determine, for the historical patients, associations between an historical LTBSC and at least one historical LTEEG signal corresponding to the historical LTBSC.

[0277] 5. The computer-implemented method according to any one of the preceding aspects, wherein the LTBSC data includes at least a portion of the LTEEG data corresponding to the at least one LTBSC.

[0278] 6. The computer-implemented method according to any one of the preceding aspects, wherein the LTBSC data includes a classification and / or confidence metric associated with the at least one LTBSC.

[0279] 7. The computer-implemented method according to any one of the preceding aspects, further comprising: obtaining, by the one or more processors, long-term biometric (LTB) data indicating long-term biometrics of the patient corresponding to the at least one LTBSC; and providing, by the one or more processors, the LTB data to the first model, wherein generating the LTBSC data is further based upon the LTB data.

[0280] 8. The computer-implemented method according to aspect 7, wherein the long-term biometrics indicate one or more of a voice biomarker, eye attentiveness, or a facial change.

[0281] 9. The computer-implemented method according to aspect 7 or aspect 8, wherein: the first model training data further comprises historical LTB data of historical patients corresponding to the historical LTBSC data of the historical patients; and the first model is further trained to determine, for the historical patients, associations between the historical LTBSC and historical longterm biometrics corresponding to the historical LTBSC.

[0282] 10. The computer-implemented method according to any one of aspects 7 to 9, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

[0283] 1 1 . The computer-implemented method according to any one of the preceding aspects, wherein generating the LTBSC data includes one or more of: analyzing, by the one or more processors, at least one frequency band of the at least one LTEEG signal; identifying, by the one or more processors, at least one LTEEG segment in the at least one LTEEG signal based upon one or more of a root means square, a standard deviation, or a power spectrum density, of the at least one LTEEG signal; determining, by the one or more processors, one or more feature values of the atleast one LTEEG segment by applying a principal components analysis algorithm or a support vector machine to the LTEEG data; determining, by the one or more processors, at least one similar characteristic of the LTEEG data and / or the one or more feature values, across long- duration time points; classifying, by the one or more processors, LTEEG segments having similar characteristics; or determining, by the one or more processors, an LTBSC boundary based upon detecting an edge of a phase transition of the at least one LTEEG signal.

[0284] 12. The computer-implemented method according to aspect 1 1 , wherein the one or more feature values are associated with one or more characteristics of an electroencephalogram (EEG) signal.

[0285] 13. The computer-implemented method according to aspect 1 1 or 12, wherein the at least one frequency band is selected from the group consisting of 1 to 2 hertz (Hz), 0.5 to 4 Hz, 4 to 8 Hz, 8 to 13 Hz, 13 to 30 Hz, 30 to 100 Hz, 80 to 250 Hz, and 250 to 500 Hz.

[0286] 14. The computer-implemented method according to any one of the preceding aspects, wherein the at least one LTBSC is associated with one or more of epilepsy, a psychiatric disorder, a stroke, Alzheimer's disease, a sleep disorder, depression, medication efficacy, or a medication side effect.

[0287] 15. The computer-implemented method according to any one of the preceding aspects, wherein retraining the one or more one machine learning models using the LTBSC data includes fine-tuning the one or more machine learning models.

[0288] 16. The computer-implemented method according to any one of the preceding aspects, wherein the medical insight is one or more of: an indication of the at least one LTBSC; a risk of sudden unexplained death by epilepsy; or an association between the at least one LTBSC and the long-term biometrics.

[0289] 17. The computer-implemented method according to any one of the preceding aspects, further comprising: obtaining, by the one or more processors, additional first model training data; retraining, by the one or more processors, the first model using the additional first model training data; and storing, by the one or more processors, the retrained first model on a memory to generate subsequent LTBSC data using the retrained first model.

[0290] 18. The computer-implemented method according to any one of the preceding aspects, wherein: the one or more machine learning models include a second model trained using second model training data to generate second model output data indicating a seizure of the patient in the LTEEG of the LTBSC data, based upon receiving the LTBSC data; retraining the second modelusing the LTBSC data further comprises storing, by the one or more processors, the retrained second model in the memory to generate subsequent second model output data using the retrained second model; and providing the LTBSC data as the input to the second model further comprises, in response to generating the second model output data, providing, by the one or more processors, the second model output data to the user device.

[0291] 19. The computer-implemented according to aspect 18, wherein the second model training data comprises historical LTBSC data of historical patients.

[0292] 20. The computer-implemented method according to aspect 19, wherein the second model is trained to determine, for the historical patients, associations between at least one historical LTBSC and an historical seizure.

[0293] 21 . The computer-implemented method according to any one of aspects 18 to 20, wherein the second model output data includes a classification of the seizure and / or a confidence metric associated with the indication of the seizure.

[0294] 22. The computer-implemented method according to any one of aspects 18 to 21 , wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0295] 23. The computer-implemented method according to any one of aspects 18 to 22, further comprising: obtaining, by the one or more processors, additional second model training data; retraining, by the one or more processors, the second model using the additional second model training data; and storing, by the one or more processors, the retrained second model in the memory to generate subsequent second model output data using the retrained second model.

[0296] 24. The computer-implemented method according to any one of aspects 1 to 17, wherein: the one or more machine learning models include a third model trained using third model training data to generate third model output data, based upon receiving the LTBSC data, wherein: the third model output data includes a revision of an indication of a seizure in the LTEEG data of the LTBSC data; the indication of the seizure is not based upon the at least one LTBSC of the LTBSC data; and the revision to the indication of the seizure is based upon the at least one LTBSC of the LTBSC data; retraining the third model using the LTBSC data further comprises storing, by the one or more processors, the retrained third model in the memory to generate subsequent third model output data using the retrained third model; and providing the LTBSC data as the input to the third model further comprises, in response to generating the third model output data, providing, by the one or more processors, the third model output data to the user device.

[0297] 25. The computer-implemented method according to aspect 24, wherein the third model training data comprises historical LTBSC data of historical patients.

[0298] 26. The computer-implemented method according to aspect 25, wherein the third model is trained to determine, for the historical patients, associations between an historical seizure and at least one historical LTBSC.

[0299] 27. The computer-implemented method according to any one of aspects 24 to 26, wherein the revision includes a reclassification of the seizure and / or a change in a confidence metric associated with the indication of the seizure.

[0300] 28. The computer-implemented method according to any one of aspects 24 to 27, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0301] 29. The computer-implemented method according to any one of aspects 24 to 28, further comprising: obtaining, by the one or more processors, additional third model training data; retraining, by the one or more processors, the third model using the additional third model training data; and storing, by the one or more processors, the retrained third model in the memory to generate subsequent third model output data using the retrained third model.

[0302] 30. The computer-implemented method according to any one of aspects 1 to 17, wherein: the one or more machine learning models include a fourth model trained using fourth model training data to generate fourth model output data indicating a prediction of a seizure of the patient, based upon receiving the LTBSC data; retraining the fourth model using the LTBSC data further comprises storing, by the one or more processors, the retrained fourth model in the memory to generate subsequent fourth model output data using the retrained fourth model; and providing the LTBSC data as the input to the fourth model further comprises, in response to generating the fourth model output data, providing, by the one or more processors, the fourth model output data to the user device.

[0303] 31 . The computer-implemented method according to aspect 30, wherein the fourth model training data comprises historical LTBSC data of historical patients.

[0304] 32. The computer-implemented method according to aspect 31 , wherein the fourth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and timing of an historical seizure.

[0305] 33. The computer-implemented method according to any one of aspects 30 to 32, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0306] 34. The computer-implemented method according to any one of aspects 30 to 33, wherein the fourth model output data includes at least one date of the prediction of the seizure, and / or a confidence metric associated with the prediction.

[0307] 35. The computer-implemented method according to any one of aspects 30 to 34, further comprising: obtaining, by the one or more processors, additional fourth model training data; retraining, by the one or more processors, the fourth model using the additional fourth model training data; and storing, by the one or more processors, the retrained fourth model in the memory to generate subsequent fourth model output data using the retrained fourth model.

[0308] 36. The computer-implemented method according to any one of aspects 1 to 17, wherein: the one or more machine learning models include a fifth model trained using fifth model training data to generate fifth model output data indicating at least one date to implement a medication titration of the patient, based upon receiving the LTBSC data; retraining the fifth model using the LTBSC data further comprises storing, by the one or more processors, the retrained fifth model in the memory to generate subsequent fifth model output data using the retrained fifth model; and providing the LTBSC data as the input to the fifth model further comprises, in response to generating the fifth model output data, providing, by the one or more processors, the fifth model output data to the user device.

[0309] 37. The computer-implemented method according to aspect 36, wherein the fifth model training data comprises historical LTBSC data of historical patients.

[0310] 38. The computer-implemented method according to aspect 37, wherein the fifth model is trained to determine, for the historical patients, associations between at least one historical LTBSC, timing of an historical medication titration, and associated historical side effects from the historical medication titration.

[0311] 39. The computer-implemented method according to any one of aspects 36 to 38, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0312] 40. The computer-implemented method according to any one of aspects 36 to 39, further comprising: obtaining, by the one or more processors, additional fifth model training data; retraining, by the one or more processors, the fifth model using the additional fifth model training data; and storing, by the one or more processors, the retrained fifth model in the memory to generate subsequent fifth model output data using the retrained fifth model.

[0313] 41 . The computer-implemented method according to any one of aspects 1 to 17, wherein: the one or more machine learning models include a sixth model trained using sixth modeltraining data to generate sixth model output data indicating an association between at least one environmental characteristic of the patient and the at least one LTBSC of the patient, based upon receiving environmental data indicating the at least one environmental characteristic of the patient and the LTBSC data; retraining the sixth model using the LTBSC data further comprises storing, by the one or more processors, the retrained sixth model in the memory to generate subsequent sixth model output data using the retrained sixth model; and providing the LTBSC data as the input to the sixth model further comprises: obtaining, by the one or more processors, the environmental data associated with the LTBSC data; providing, by the one or more processors, the environmental data and the LTBSC data to the sixth model to generate the sixth model output data; and in response to generating the sixth model output data, providing, by the one or more processors, the sixth model output data to the user device.

[0314] 42. The computer-implemented method according to aspect 41 , wherein the sixth model training data comprises historical LTBSC data of historical patients and historical environmental data of the historical patients.

[0315] 43. The computer-implemented method according to aspect 42, wherein the sixth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and at least one historical environmental characteristic.

[0316] 44. The computer-implemented method according to any one of aspects 41 to 43, wherein the at least one environmental characteristic is associated with a medication.

[0317] 45. The computer-implemented method according to aspect 44, wherein the association with the medication includes a new medication taken by the patient, and / or a new titration of the medication taken by the patient.

[0318] 46. The computer-implemented method according to any one of aspects 41 to 45, wherein: the environmental data is obtained in response to a LTBSC of the patient; and the sixth model output data indicates a cause of the LTBSC.

[0319] 47. The computer-implemented method according to any one of aspects 41 to 46, wherein the sixth model output data indicates a prediction of a future LTBSC in response to the at least one environmental characteristic, and / or a treatment pathway to a target brain state.

[0320] 48. The computer-implemented method according to any one of aspects 41 to 47, wherein the environmental data is generated by an implant device of the patient.

[0321] 49. The computer-implemented method according to any one of aspects 41 to 48, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0322] 50. The computer-implemented method according to any one of aspects 41 to 48, further comprising: obtaining, by the one or more processors, additional sixth model training data; retraining, by the one or more processors, the sixth model using the additional sixth model training data; and storing, by the one or more processors, the retrained sixth model in the memory to generate subsequent sixth model output data using the retrained sixth model.

[0323] 51 . A system for determining a long-term brain state change (LTBSC) of a patient, the system comprising: an implant device comprising a sensor array and a processor device, the sensor array comprising a plurality of electrodes and a local processing device comprising local device communication circuitry, the sensor array being communicatively coupled to the processor device via the local processing device communication circuitry, the sensor array being configured to gather long-term electroencephalogram (LTEEG) data comprising at least one LTEEG signal associated with a brain of the patient, and provide the LTEEG data to the processing device; and the processor device comprising a microprocessor, a memory, and communication circuitry, the processing device being communicatively coupled to the sensor array via the communication circuitry; a first model, stored in the memory and configured to be executed by the microprocessor, the first model trained using first model training data and operable to: receive the LTEEG data; determine one or more feature values of the LTEEG data; and based on the one or more feature values, generate LTBSC data indicating at least one LTBSC of the patient; and in response to generating the LTBSC data, the processor device is configured to one or more of: (i) retrain one or more machine learning models using the LTBSC data; (ii) provide the LTBSC data as an input to the one or more machine learning models; or (iii) generate a medical insight of the patient based upon the LTBSC data; and provide the medical insight to a user device.

[0324] 52. The system according to aspect 51 , wherein the processor device is a wearable processor device.

[0325] 53. The system according to aspect 51 , wherein the processor device is embedded in the sensor array or disposed adjacent to the sensor array.

[0326] 54. The system according to aspect 51 , wherein the processor device communicates wirelessly with a server.

[0327] 55. The system according to aspect 51 , wherein the processor device communicates wirelessly with the user device.

[0328] 56. The system according to aspect 55, wherein the user device is configured to communicate data from the processor device to one or more servers via an Internet.

[0329] 57. The system according to aspect 51 , wherein the processor device is the user device.

[0330] 58. The system according to aspect 51 , wherein the processor device is a server.

[0331] 59. The system according to aspect 51 , wherein the processor device is remote from the sensor array and is communicatively coupled to the sensor array via an Internet.

[0332] 60. The system according to any one of aspects 51 to 59, wherein the plurality of electrodes provide one or more electrical signals indicating a presence or absence of a biomarker.

[0333] 61 . The system according to aspect 60, wherein the biomarker includes epileptiform activity biomarker data determined from the LTEEG data received from one or more sensor arrays.

[0334] 62. The system according to any one of aspects 51 to 61 , wherein the sensor array comprises a wireless transceiver.

[0335] 63. The system according to any one of aspects 51 to 62, wherein the LTEEG data includes at least one substantially continuous LTEEG signal of the patient gathered over more than a week of time.

[0336] 64. The system according to any one of aspects 51 to 63, wherein:

[0337] the first model training data comprises historical LTEEG data and historical LTBSC data of historical patients corresponding to the historical LTEEG data.

[0338] 65. The system according to aspect 64, wherein: the first model is trained to determine, for the historical patients, associations between an historical LTBSC and at least one historical LTEEG signal corresponding to the historical LTBSC.

[0339] 66. The system according to any one of aspects 51 to 65, wherein the LTBSC data includes at least a portion of the LTEEG data corresponding to the at least one LTBSC.

[0340] 67. The system according to any one of aspects 51 to 66, wherein the LTBSC data includes a classification and / or confidence metric associated with the at least one LTBSC.

[0341] 68. The system according to any one of aspects 51 to 67, wherein the one or more feature values of the LTEEG data include one or more characteristics of an electroencephalogram (EEG) signal.

[0342] 69. The system according to any one of aspects 51 to 68, wherein: the processor device is further configured to obtain long-term biometric (LTB) data indicating long-term biometrics of the patient corresponding to the at least one LTBSC; and the first model, when executed by themicroprocessor, is further operable to: receive the LTB data; and determine one or more feature values of the LTB data, wherein generating the LTBSC data is further based upon the LTB data.

[0343] 70. The system according to aspect 69, wherein the long-term biometrics indicate one or more of a voice biomarker, eye attentiveness, or a facial change.

[0344] 71 . The system according to aspects 69 or 70, wherein the one or more feature values of the LTB data include biometrics of the face and / or eyes.

[0345] 72. The system according to any one of aspects 69 to 71 , wherein: the first model training data further comprises historical LTB data of historical patients corresponding to the historical LTBSC data of the historical patients; and the first model is further trained to determine, for the historical patients, associations between the historical LTBSC and historical long-term biometrics corresponding to the historical LTBSC.

[0346] 73. The system according to any one of aspects 69 to 72, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

[0347] 74. The system according to any one of aspects 51 to 73, wherein to generate the LTBSC, the first model, when executed by the microprocessor, is further operable to: analyze at least one frequency band of the at least one LTEEG signal; identify at least one LTEEG segment in the at least one LTEEG signal based upon one or more of a root means square, a standard deviation, or a power spectrum density, of the at least one LTEEG signal; determine one or more feature values of the at least one LTEEG segment by applying a principal components analysis algorithm or a support vector machine to the LTEEG data; determine at least one similar characteristic of the LTEEG data and / or the one or more feature values, across long-duration time points; classify LTEEG segments having similar characteristics; or determine an LTBSC boundary based upon detecting an edge of a phase transition of the at least one LTEEG signal.

[0348] 75. The system according to aspect 74, wherein the one or more feature values of the LTEEG segment include one or more characteristics of an electroencephalogram (EEG) signal.

[0349] 76. The system according to aspects 74 or 75, wherein the at least one frequency band is selected from the group consisting of 1 to 2 hertz (Hz), 0.5 to 4 Hz, 4 to 8 Hz, 8 to 13 Hz, 13 to 30 Hz, 30 to 100 Hz, 80 to 250 Hz, and 250 to 500 Hz.

[0350] 77. The system according to any one of aspects 51 to 76, wherein the at least one LTBSC is associated with one or more of epilepsy, a psychiatric disorder, a stroke, Alzheimer's disease, a sleep disorder, depression, medication efficacy, or a medication side effect.

[0351] 78. The system according to any one of aspects 51 to 77, wherein retraining the one or more one machine learning models using the LTBSC data includes fine-tuning the one or more machine learning models.

[0352] 79. The system according to any one of aspects 51 to 58, wherein the medical insight is one or more of: an indication of the at least one LTBSC; a risk of sudden unexplained death by epilepsy; or an association between the at least one LTBSC and the long-term biometrics.

[0353] 80. The system according to any one of aspects 51 to 79, wherein the processor device is further configured to: receive additional first model training data; retrain the first model using the additional first model training data; and store the retrained first model in the memory to generate subsequent LTBSC data using the retrained first model.

[0354] 81 . The system according to any one of aspects 51 to 80, wherein: the one or more machine leaning models include a second model, stored in the memory and configured to be executed by the microprocessor, the second model trained using second model training data and operable to generate second model output data indicating a seizure of the patient in the LTEEG of the LTBSC data, based upon receiving the LTBSC data; to retrain the second model using the LTBSC data, the processor device is further configured to store the retrained second model in the memory to generate subsequent second model output data using the retrained second model; and to provide the LTBSC data as the input to the second model, the processor device is further configured to, in response to the second model generating the second model output data, provide the second model output data to the user device.

[0355] 82. The system according to aspect 81 , wherein the second model training is trained using second model training data comprising historical LTBSC data of historical patients.

[0356] 83. The system according to aspect 82, wherein the second model is trained to determine, for the historical patients, associations between at least one historical LTBSC and an historical seizure.

[0357] 84. The system according to any one of aspects 81 to 83, wherein the second model output data includes a classification of the seizure and / or a confidence metric associated with the indication of the seizure.

[0358] 85. The system according to any one of aspects 51 to 84, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0359] 86. The system according to any one of aspects 51 to 55, wherein the processor is further configured to: obtain additional second model training data; retrain the second model usingthe additional second model training data; and store the retrained second model in the memory to generate subsequent second model output data using the retrained second model.

[0360] 87. The system according to any one of aspects 51 to 80, wherein: the one or more machine leaning models include third model, stored in the memory and configured to be executed by the microprocessor, the third model trained using third model training data and operable to generate third model output data, based upon receiving the LTBSC data, wherein: the third model output data includes a revision of an indication of a seizure in the LTEEG data of the LTBSC data; the indication of the seizure is not based upon the at least one LTBSC of the LTBSC data; and the revision to the indication of the seizure is based upon the at least one LTBSC of the LTBSC data; to retrain the third model using the LTBSC data, the processor device is further configured to store the retrained third model in the memory to generate subsequent third model output data using the retrained third model; and to provide the LTBSC data as the input to the third model, the processor device is further configured to, in response to generating the third model output data, provide the third model output data to the user device.

[0361] 88. The system according to aspect 87, wherein the third model is trained using third model training data comprising historical LTBSC data of historical patients.

[0362] 89. The system according to aspect 88, wherein the third model is trained to determine, for the historical patients, associations between an historical seizure and at least one historical LTBSC.

[0363] 90. The system according to any one of aspects 87 to 89, wherein the revision includes a reclassification of the seizure and / or a change in a confidence metric associated with the indication of the seizure.

[0364] 91 . The system according to any one of aspects 87 to 90, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0365] 92. The system according to any one of aspects 87 to 91 , wherein the processor device is further configured to: receive additional third model training data; retrain the third model using the additional third model training data; and store the retrained third model in the memory to generate subsequent third model output data using the retrained third model.

[0366] 93. The system any one of aspects 51 to 80, wherein: the one or more machine leaning models include a fourth model, stored in the memory and configured to be executed by the microprocessor, the fourth model trained using fourth model training data and operable to generate fourth model output data indicating a prediction of a seizure of the patient, based upon receiving theLTBSC data; to retraining the fourth model using the LTBSC data, the processor device is further configured to store the retrained fourth model in the memory to generate subsequent fourth model output data using the retrained fourth model; and to provide the LTBSC data as the input to the fourth model, the processor device is further configured to, in response to the fourth model generating the fourth model output data, provide the fourth model output data to the user device.

[0367] 94. The system according to aspect 93, wherein the fourth model is trained using fourth model training data comprising historical LTBSC data of historical patients.

[0368] 95. The system according to aspect 94, wherein the fourth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and timing of an historical seizure.

[0369] 96. The system according to any one of aspects 93 to 95, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0370] 97. The system according to any one of aspects 93 to 96, wherein the fourth model output data includes at least one date of the prediction of the seizure, and / or a confidence metric associated with the prediction.

[0371] 98. The system according to any one of aspects 93 to 97, wherein the processor device is further configured to: receive additional fourth model training data; retrain the fourth model using the additional fourth model training data; and store the retrained fourth model in the memory to generate subsequent fourth model output data using the retrained fourth model.

[0372] 99. The system any one of aspects 51 to 80, wherein the one or more machine leaning models include a fifth model, stored in the memory and configured to be executed by the microprocessor, the fifth model trained using fifth model training data and operable to generate fifth model output data indicating at least one date to implement a medication titration of the patient based upon receiving the LTBSC data; to retrain the fifth model using the LTBSC data, the processor device is further configured to store the retrained fifth model in the memory to generate subsequent fifth model output data using the retrained fifth model; and to provide the LTBSC data as the input to the fifth model, the processor device is further configured to, in response to the fifth model generating the fifth model output data, provide the fifth model output data to the user device.

[0373] 100. The system according to aspect 99, wherein the fifth model training data comprises historical LTBSC data of historical patients.

[0374] 101 . The system according to aspect 100, wherein the fifth model is trained to determine, for the historical patients, associations between at least one historical LTBSC, timing of an historical medication titration, and associated historical side effects from the historical medication titration.

[0375] 102. The system according to any one of aspects 99 to 101 , wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0376] 103. The system according to any one of aspects 99 to 102, wherein the processor device is further configured to: receive additional fifth model training data; retrain the fifth model using the additional fifth model training data; and store the retrained fifth model in the memory to generate subsequent fifth model output data using the retrained fifth model.

[0377] 104. The system according to any one of aspects 51 to 80, wherein: the one or more machine leaning models include a sixth model, stored in the memory and configured to be executed by the microprocessor, the sixth model trained using sixth model training data and operable to generate sixth model output data indicating an association between at least one environmental characteristic of the patient and the at least one LTBSC of the patient, based upon receiving environmental data indicating the at least one environmental characteristic of the patient and the LTBSC data; to retrain the sixth model using the LTBSC data, the processor device is further configured to store the retrained sixth model in the memory to generate subsequent sixth model output data using the retrained sixth model; and to provide the LTBSC data as the input to the sixth model, the processor device is further configured: obtain the environmental data associated with the LTBSC data; provide the environmental data and the LTBSC data to the sixth model to generate the sixth model output data; and in response to generating the sixth model output data by the sixth model, provide the sixth model output data to the user device.

[0378] 105. The system according to aspect 104, wherein the sixth model training data comprises historical LTBSC data of historical patients and historical environmental data of the historical patients.

[0379] 106. The system according to aspect 105, wherein the sixth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and at least one historical environmental characteristic.

[0380] 107. The system according to any one of aspects 104 to 106, wherein the at least one environmental characteristic is associated with a medication.

[0381] 108. The system according to aspect 107, wherein the association with the medication includes a new medication taken by the patient, and / or a new titration of the medication taken by the patient.

[0382] 109. The system according to any one of aspects 104 to 108, wherein: the environmental data is obtained in response to a LTBSC of the patient; and the sixth model output data indicates a cause of the LTBSC.

[0383] 1 10. The system according to any one of aspects 104 to 109, wherein the sixth model output data indicates a prediction of a future LTBSC in response to the at least one environmental characteristic, and / or a treatment pathway to a target brain state.

[0384] 1 1 1. The system according to any one of aspects 104 to 110, wherein the environmental data is generated by the implant device of the patient.

[0385] 1 12. The system according to any one of aspects 104 to 11 1 , wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0386] 1 13. The system according to any one of aspects 104 to 112, wherein the processor device is further configured to: receive additional sixth model training data; retrain the sixth model using the additional sixth model training data; and store the retrained sixth model in the memory to generate subsequent sixth model output data using the retrained sixth model.

[0387] 1 14. A first model for determining a long-term brain state change (LTBSC) of a patient, the first model comprising: the first model, stored on one or more memories and configured to be executed by one or more processors, the first model trained using first model training data and operable to: receive long-term electroencephalogram (LTEEG) data comprising at least one LTEEG signal associated with a brain of the patient and gathered via an implant device of the patient; determine one or more feature values of the LTEEG data; and based on the one or more feature values, generate LTBSC data indicating at least one LTBSC of the patient.

[0388] 1 15. The first model of aspect 1 14, wherein the first model is a machine learning model.

[0389] 1 16. The first model of aspect 1 14, wherein the first model is based upon a static algorithm.

[0390] 1 17. The first model of any one of aspects 1 14 to 1 16, wherein the LTEEG data includes at least one substantially continuous LTEEG signal of the patient gathered over more than a week of time.

[0391] 1 18. The first model of any one of aspects 1 14 to 1 17, wherein the first model training data comprises historical LTEEG data of historical patients and historical LTBSC data of historical patients corresponding to the historical LTEEG data.

[0392] 1 19. The first model of aspect 1 18, wherein the first model is trained to determine, for the historical patients, associations between an historical LTBSC and at least one historical LTEEG signal corresponding to the historical LTBSC.

[0393] 120. The first model of any one of aspects 1 14 to 1 19, wherein the LTBSC data includes at least a portion of the LTEEG data corresponding to the at least one LTBSC.

[0394] 121 . The first model of any one of aspects 1 14 to 120, wherein the LTBSC data includes a classification and / or confidence metric associated with the at least one LTBSC.

[0395] 122. The first model of any one of aspects 1 14 to 121 , the first model further configured, when executed by the one or more processors, to be operable to: obtain long-term biometric (LTB) data indicating long-term biometrics of the patient corresponding to the at least one LTBSC; and determine one or more feature values of the LTB data, wherein generating the LTBSC data is further based upon the LTB data.

[0396] 123. The first model of aspect 122, wherein the long-term biometrics indicate one or more of a voice biomarker, eye attentiveness, or a facial change.

[0397] 124. The first model of aspects 122 or 123, wherein: the first model training data further comprises historical LTB data of historical patients corresponding to the historical LTBSC data of the historical patients; and the first model is further trained to determine, for the historical patients, associations between the historical LTBSC and historical long-term biometrics corresponding to the historical LTBSC.

[0398] 125. The first model of any one of aspects 122 to 124, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

[0399] 126. The first model of any one of aspects 1 14 to 125, wherein the one or more feature values are associated with one or more characteristics of an electroencephalogram (EEG) signal.

[0400] 127. The first model of any one of aspects 1 14 to 126, wherein the at least one LTBSC is associated with one or more of epilepsy, a psychiatric disorder, a stroke, Alzheimer's disease, a sleep disorder, depression, medication efficacy, or a medication side effect.

[0401] 128. A second model trained using second model training data to generate second model output data indicating a seizure of a patient in long-term brain state change (LTBSC) data, thesecond model stored on one or more memories and configured to be executed by one or more processors, the second model operable to: receive the LTBSC data indicating at least one LTBSC of the patient; determine one or more feature values of the LTBSC data; and based on the one or more feature values, generate the second model output data.

[0402] 129. The second model of aspect 128, wherein the second model is a machine learning model.

[0403] 130. The second model of aspect 128, wherein the second model is based upon a static algorithm.

[0404] 131 . The second model of any one of aspects 128 to 130, wherein the second model training data comprises historical LTBSC data of historical patients.

[0405] 132. The second model of aspect 131 , wherein the second model is trained to determine, for the historical patients, associations between at least one historical LTBSC and an historical seizure.

[0406] 133. The second model of any one of aspects 128 to 132, wherein the second model output data includes a classification of the seizure and / or a confidence metric associated with the indication of the seizure.

[0407] 134. The second model of any one of aspects 128 to 133, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0408] 135. A fourth model trained using fourth model training data to generate fourth model output data indicating a prediction of a seizure of a patient, the fourth model, stored on one or more memories and configured to be executed by one or more processors, the fourth model operable to: receive long-term brain state change (LTBSC) data indicating at least one LTBSC of the patient; determine one or more feature values of the LTBSC data; and based on the one or more feature values, generate the fourth model output data.

[0409] 136. The fourth model of aspect 135, wherein the fourth model is a machine learning model.

[0410] 137. The fourth model of aspect 135, wherein the fourth model is based upon a static algorithm.

[0411] 138. The fourth model of any one of aspects 135 to 137, wherein the fourth model training data comprises historical LTBSC data of historical patients.

[0412] 139. The fourth model of aspect 138, wherein the fourth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and timing of an historical seizure.

[0413] 140. The fourth model of any one of aspects 135 to 139, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0414] 141 . The fourth model of any one of aspects 135 to 140, wherein the fourth model output data includes a date of the prediction of the seizure, and / or a confidence metric associated with the prediction.

[0415] 142. A third model trained using third model training data to generate third model output data, the third model, stored on one or more memories and configured to be executed by one or more processors, the third model operable to: receive long-term brain state change (LTBSC) data indicating at least one LTBSC of a patient; determine one or more feature values of the LTBSC data; and based on the one or more feature values, generate the third model output data, wherein: the third model output data includes a revision of an indication of a seizure in LTEEG data of the LTBSC data; the indication of the seizure is not based upon the at least one LTBSC of the LTBSC data; and the revision to the indication of the seizure is based upon the at least one LTBSC of the LTBSC data.

[0416] 143. The third model of aspect 142, wherein the third model is a machine learning model.

[0417] 144. The third model of aspect 142, wherein the third model is based upon a static algorithm.

[0418] 145. The third model of any one of aspects 142 to 144, wherein the third model training data comprises historical LTBSC data of historical patients.

[0419] 146. The third model of aspect 145, wherein the third model is trained to determine, for the historical patients, associations between an historical seizure and at least one historical LTBSC.

[0420] 147. The third model of any one of aspects 142 to 146, wherein the revision includes a reclassification of the seizure and / or a change in a confidence metric associated with the indication of the seizure.

[0421] 148. The third model of any one of aspects 142 to 147, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0422] 149. A fifth model trained using fifth model training data to generate fifth model output data indicating at least one date to implement a medication titration of a patient, the fifth model,stored on one or more memories and configured to be executed by one or more processors, the fifth model operable to: receive long-term brain state change (LTBSC) data indicating at least one LTBSC of the patient; determine one or more feature values of the LTBSC data; and based on the one or more feature values, generate the fifth model output data.

[0423] 150. The fifth model of aspect 149, wherein the fifth model is a machine learning model.

[0424] 151 . The fifth model of aspect 149, wherein the fifth model is based upon a static algorithm.

[0425] 152. The fifth model of any one of aspects 149 to 151 , wherein the fifth model training data comprises historical LTBSC data of historical patients.

[0426] 153. The fifth model of aspect 152, wherein the fifth model is trained to determine, for the historical patients, associations between at least one historical LTBSC, timing of an historical medication titration, and associated historical side effects from the historical medication titration.

[0427] 154. The fifth model of any one of aspects 149 to 153, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0428] 155. A sixth model trained using sixth model training data to generate sixth model output data indicating an association between at least one environmental characteristic of a patient and the at least one long-term brain state change (LTBSC) of the patient, the sixth model, stored on one or more memories and configured to be executed by one or more processors, the sixth model operable to: obtain LTBSC data indicating at least one LTBSC of the patient, and environmental data indicating the at least one environmental characteristic of the patient; determine one or more feature values of the LTBSC data and the environmental data; and based on the one or more feature values, generate the sixth model output data.

[0429] 156. The sixth model of aspect 155, wherein the sixth model is a machine learning model.

[0430] 157. The sixth model of aspect 155, wherein the sixth model is based upon a static algorithm.

[0431] 158. The sixth model of any one of aspects 155 to 157, wherein the sixth model training data comprises historical LTBSC data of historical patients and historical environmental data of the historical patients.

[0432] 159. The sixth model of aspect 158, wherein the sixth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and at least one historical environmental characteristic.

[0433] 160. The sixth model of any one of aspects 155 to 159, wherein the at least one environmental characteristic is associated with a medication.

[0434] 161 . The sixth model of aspect 160, wherein the association with the medication includes a new medication taken by a patient, and / or a new titration of the medication taken by the patient.

[0435] 162. The sixth model of any one of aspects 155 to 161 , wherein: the environmental data is obtained in response to a LTBSC of the patient; and the sixth model output data indicates a cause of the LTBSC.

[0436] 163. The sixth model of any one of aspects 155 to 162, wherein the sixth model output data indicates a prediction of a future LTBSC in response to the at least one environmental characteristic, and / or a treatment pathway to a target brain state.

[0437] 164. The sixth model of any one of aspects 155 to 163, wherein the environmental data is generated by an implant device of the patient.

[0438] 165. The sixth model of any one of aspects 155 to 164, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0439] 166. A computer-implemented method for training a first model to determine a long-term brain state change (LTBSC) of a patient, the method comprising: obtaining, by one or more processors, a first training dataset; receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; training, by the one or more processors, a first iteration of the first model using the first training dataset using the feature values defined for the first training dataset to generate primary LTBSC data having associated first error rates; obtaining, by the one or more processors, a second training dataset; receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and training, by the one or more processors, a second iteration of the first model using the second training dataset using the feature values defined for the second training dataset to generate secondary LTBSC data having associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

[0440] 167. The computer-implemented method according to aspect 166, wherein the one or more attributes comprise one or more characteristics of an electroencephalogram (EEG) signal.

[0441] 168. The computer-implemented method according to aspects 166 or 167, wherein a training dataset comprises historical long-term EEG (LTEEG) data of historical patients and historical LTBSC data of historical patients corresponding to the historical LTEEG data.

[0442] 169. The computer-implemented method according to aspect 168, wherein the first model is trained to determine, for the historical patients, associations between an historical LTBSC and at least one historical LTEEG signal corresponding to the historical LTBSC.

[0443] 170. The computer-implemented method according to aspects 168 or 169, wherein the LTBSC data includes at least a portion of the LTEEG data corresponding to the at least one LTBSC.

[0444] 171. The computer-implemented method according to any one of aspects 166 to 170, wherein: the training dataset further comprises historical LTB data of historical patients corresponding to the historical LTBSC data of the historical patients; and the first model is further trained to determine, for the historical patients, associations between the historical LTBSC and historical long-term biometrics corresponding to the historical LTBSC.

[0445] 112.. The computer-implemented method according to aspect 171 , wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

[0446] 173. The computer-implemented method according to any one of aspects 166 to 172, comprising: obtaining, by the one or more processors, LTEEG data associated with one or more patients; and based upon the LTEEG data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

[0447] 174. The computer-implemented method according to any one of aspects 166 to 173, further comprising: receiving one or more labels for the first training dataset, wherein training the first iteration of the first model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the first LTBSC data; and receiving one or more labels for the second training dataset, wherein training the second iteration of the first model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the second LTBSC data.

[0448] 175. A computer-implemented method for training a second model to generate second model output data indicating a seizure of the patient, the method comprising: obtaining, by one or more processors, a first training dataset; receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; training, by the one or more processors, a first iteration of the second model using the first training dataset using the feature values defined for the first training dataset to generate primary second model output data having associated first error rates; obtaining, by the one or more processors, a second training dataset; receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and training, by the one or more processors, a seconditeration of the second model using the second training dataset using the feature values defined for the second training dataset to generate secondary second model output data having associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

[0449] 176. The computer-implemented method according to aspect 175, wherein the one or more attributes comprise one or more characteristics of an electroencephalogram (EEG) signal.

[0450] 177. The computer-implemented method according to aspects 175 or 176, wherein a training dataset comprises historical LTBSC data of historical patients.

[0451] 178. The computer-implemented method according to aspect 177, wherein the second model is trained to determine, for the historical patients, associations between at least one historical LTBSC and an historical seizure.

[0452] 179. The computer-implemented method according to any one of aspects 175 to 178, wherein the second model output data includes at least a portion of LTBSC data corresponding to the seizure.

[0453] 180. The computer-implemented method according to any one of aspects 175 to 179, wherein: the training dataset further comprises historical long-term biometric (LTB) data corresponding to the historical LTBSC data of the historical patients; and the second model is further trained to determine, for the historical patients, associations between the historical LTBs, and the corresponding historical LTBSCs and / or historical seizures.

[0454] 181 . The computer-implemented method according to aspect 180, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

[0455] 182. The computer-implemented method according to any one of aspects 175 to 181 , comprising: obtaining, by the one or more processors, LTBSC data associated with one or more patients; and based upon the LTBSC data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

[0456] 183. The computer-implemented method according to any one of aspects 175 to 182, further comprising: receiving one or more labels for the first training dataset, wherein training the first iteration of the second model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the primary second model output data; and receiving one or more labels for the second training dataset, wherein training the second iteration of the second model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the secondary second model output data.

[0457] 184. A computer-implemented method according to any one of aspects 175 to 183, wherein: the first training dataset does not include long-term based brain state change (LTBSC) data, and the second training dataset includes the LTBSC data, the first and second training datasets being used to fine-tune the second model.

[0458] 185. A computer-implemented method for training a fourth model to generate fourth model output data indicating a prediction of a seizure of the patient, the method comprising: obtaining, by one or more processors, a first training dataset; receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; training, by the one or more processors, a first iteration of the fourth model using the first training dataset using the feature values defined for the first training dataset to generate primary fourth model output data having associated first error rates; obtaining, by the one or more processors, a second training dataset; receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and training, by the one or more processors, a second iteration of the fourth model using the second training dataset using the feature values defined for the second training dataset to generate secondary fourth model output data having associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

[0459] 186. The computer-implemented method according to aspect 185, wherein the one or more attributes comprise one or more characteristics of an electroencephalogram (EEG) signal.

[0460] 187. The computer-implemented method according to aspects 185 or 186, wherein a training dataset comprises historical LTBSC data of historical patients.

[0461] 188. The computer-implemented method according to aspect 187, wherein the fourth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and timing of an historical seizure.

[0462] 189. The computer-implemented method according to any one of aspects 185 to 188, wherein the fourth model output data includes a date of the prediction of the seizure, and / or a confidence metric associated with the prediction.

[0463] 190. The computer-implemented method according to any one of aspects 185 to 189, wherein: the training dataset further comprises historical long-term biometric (LTB) data corresponding to the historical LTBSC data of the historical patients; and the fourth model is further trained to determine, for the historical patients, associations between the historical LTBs, and the corresponding historical LTBSCs and / or historical predictions of seizures.

[0464] 191 . The computer-implemented method according to aspect 190, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

[0465] 192. The computer-implemented method according to any one of aspects 185 to 191 , comprising: obtaining, by the one or more processors, LTBSC data associated with one or more patients; and based upon the LTBSC data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

[0466] 193. The computer-implemented method according to any one of aspects 185 to 192, further comprising: receiving one or more labels for the first training dataset, wherein training the first iteration of the fourth model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the primary fourth model output data; and receiving one or more labels for the second training dataset, wherein training the second iteration of the fourth model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the secondary fourth model output data.

[0467] 194. A computer-implemented method according to any one of aspects 185 to 193, wherein: the first training dataset does not include long-term based brain state change (LTBSC) data, and the second training dataset includes the LTBSC data, the first and second training datasets being used to fine-tune the fourth model.

[0468] 195. A computer-implemented method for training a third model to generate third model output data, the third model output data including a revision of an indication of a seizure in the longterm EEG data of the LTBSC data, the method comprising: obtaining, by one or more processors, a first training dataset; receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; training, by the one or more processors, a first iteration of the third model using the first training dataset using the feature values defined for the first training dataset to generate primary third model output data having associated first error rates; obtaining, by the one or more processors, a second training dataset; receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and training, by the one or more processors, a second iteration of the third model using the second training dataset using the feature values defined for the second training dataset to generate secondary third model output data having associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

[0469] 196. The computer-implemented method according to aspect 195, wherein the one or more attributes comprise one or more characteristics of an electroencephalogram (EEG) signal.

[0470] 197. The computer-implemented method according to aspects 195 or 196, wherein a training dataset comprises historical LTBSC data of historical patients.

[0471] 198. The computer-implemented method according to aspect 197, wherein the third model is trained to determine, for the historical patients, associations between an historical seizure and at least one historical LTBSC.

[0472] 199. The computer-implemented method according to any one of aspects 195 to 198, wherein the revision includes a reclassification of the seizure and / or a change in a confidence metric associated with the indication of the seizure.

[0473] 200. The computer-implemented method according to any one of aspects 195 to 199, wherein: the training dataset further comprises historical long-term biometric (LTB) data corresponding to the historical LTBSC data of the historical patients; and the third model is further trained to determine, for the historical patients, associations between the historical LTBs, and the corresponding historical LTBSCs and / or historical indications of seizures.

[0474] 201 . The computer-implemented method according to aspect 200, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

[0475] 202. The computer-implemented method according to any one of aspects 195 to 201 , comprising: obtaining, by the one or more processors, the LTBSC data indicating seizures of patients; and based upon the LTBSC data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

[0476] 203. The computer-implemented method according to any one of aspects 195 to 202, further comprising: receiving one or more labels for the first training dataset, wherein training the first iteration of the third model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the primary third model output data; and receiving one or more labels for the second training dataset, wherein training the second iteration of the third model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the secondary third model output data.

[0477] 204. A computer-implemented method for training a fifth model to generate fifth model output data indicating at least one date to implement a medication titration of the patient, the method comprising: obtaining, by one or more processors, a first training dataset; receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; training, by the one or more processors, a first iteration of the fifth model using the first training dataset using the feature values defined for the first training dataset to generateprimary fifth model output data having associated first error rates; obtaining, by the one or more processors, a second training dataset; receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and training, by the one or more processors, a second iteration of the fifth model using the second training dataset using the feature values defined for the second training dataset to generate secondary fifth model output data having associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

[0478] 205. The computer-implemented method according to aspect 204, wherein the one or more attributes comprise one or more characteristics of an electroencephalogram (EEG) signal.

[0479] 206. The computer-implemented method according to aspects 204 or 205, wherein a training dataset comprises historical LTBSC data of historical patients.

[0480] 207. The computer-implemented method according to aspect 206, wherein the fifth model is trained to determine, for the historical patients, associations between at least one historical LTBSC, timing of an historical medication titration, and associated historical side effects from the historical medication titration.

[0481] 208. The computer-implemented method according to any one of aspects 204 to 207, wherein: the training dataset further comprises historical long-term biometric (LTB) data corresponding to the historical LTBSC data of the historical patients; and the fifth model is further trained to determine, for the historical patients, associations between the historical LTBs, and one or more of the corresponding historical LTBSCs, historical times in which medication titrations were implemented, and associated historical side effects.

[0482] 209. The computer-implemented method according to aspect 208, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

[0483] 210. The computer-implemented method according to any one of aspects 204 to 209, comprising: obtaining, by the one or more processors, the LTBSC data; and based upon the LTBSC data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

[0484] 21 1 . The computer-implemented method according to any one of aspects 204 to 210, further comprising: receiving one or more labels for the first training dataset, wherein training the first iteration of the fifth model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the primary fifth model output data; and receiving one or more labels for the second training dataset, wherein training the second iteration ofthe fifth model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the secondary fifth model output data.

[0485] 212. A computer-implemented method for training a sixth model to generate sixth model output data indicating an association between at least one environmental characteristic of the patient and the at least one LTBSC of the patient, the method comprising: obtaining, by one or more processors, a first training dataset; receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; training, by the one or more processors, a first iteration of the sixth model using the first training dataset using the feature values defined for the first training dataset to generate primary sixth model output data having associated first error rates; obtaining, by the one or more processors, a second training dataset; receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and training, by the one or more processors, a second iteration of the sixth model using the second training dataset using the feature values defined for the second training dataset to generate secondary sixth model output data having associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

[0486] 213. The computer-implemented method according to aspect 212, wherein the one or more attributes comprise one or more characteristics of an electroencephalogram (EEG) signal.

[0487] 214. The computer-implemented method according to aspects 212 or 213, wherein a training dataset comprises historical LTBSC data of historical patients and historical environmental data of the historical patients.

[0488] 215. The computer-implemented method according to aspect 214, wherein the sixth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and at least one historical environmental characteristic.

[0489] 216. The computer-implemented method according to any one of aspects 212 to 215, wherein: the training dataset further comprises historical long-term biometric (LTB) data corresponding to the historical LTBSC data of the historical patients; and the sixth model is further trained to determine, for the historical patients, associations between the historical LTBs, and the corresponding historical LTBSCs and / or historical environmental characteristics.

[0490] 217. The computer-implemented method according to aspect 216, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

[0491] 218. The computer-implemented method according to any one of aspects 212 to 217, comprising: obtaining, by the one or more processors, the LTBSC data; and based upon the LTBSC data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

[0492] 219. The computer-implemented method according to any one of aspects 212 to 218, further comprising: receiving one or more labels for the first training dataset, wherein training the first iteration of the sixth model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the primary sixth model output data; and receiving one or more labels for the second training dataset, wherein training the second iteration of the sixth model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the secondary sixth model output data.

[0493] 220. A non-transitory computer-readable medium storing processor-executable instructions that, when executed by one or more processors, cause the one or more processors to at least: obtain long-term electroencephalogram (LTEEG) data comprising at least one LTEEG signal associated with a brain of the patient and gathered via an implant device of the patient; provide the LTEEG data to a first model trained using first model training data to generate long-term brain state change (LTBSC) data indicating at least one LTBSC of the patient; in response to generating the LTBSC data, one or more of: (i) retrain one or more machine learning models using the LTBSC data; (i) provide the LTBSC data as an input to the one or more machine learning models; or (iii) generate a medical insight of the patient; and provide the medical insight to a user device.

[0494] 221 . The non-transitory computer-readable medium of aspect 220, wherein the LTEEG data includes at least one substantially continuous LTEEG signal of the patient gathered over more than a week of time.

[0495] 222. The non-transitory computer-readable medium of aspects 220 or 221 , wherein:

[0496] the first model training data comprises historical LTEEG data of historical patients and historical LTBSC data of historical patients corresponding to the historical LTEEG data.

[0497] 223. The non-transitory computer-readable medium of aspect 222, wherein: the first model is trained to determine, for the historical patients, associations between an historical LTBSC and at least one historical LTEEG signal corresponding to the historical LTBSC.

[0498] 224. The non-transitory computer-readable medium of any one of aspects 220 to 223, wherein the LTBSC data includes at least a portion of the LTEEG data corresponding to the at least one LTBSC.

[0499] 225. The non-transitory computer-readable medium of any one of aspects 220 to 224, wherein the LTBSC data includes a classification and / or confidence metric associated with the at least one LTBSC.

[0500] 226. The non-transitory computer-readable medium of any one of aspects 220 to 225, further comprising instructions that, when executed, cause the one or more processors to: obtain long-term biometric (LTB) data indicating long-term biometrics of the patient corresponding to the at least one LTBSC; and provide the LTB data to the first model, wherein generating the LTBSC data is further based upon the LTB data.

[0501] 227. The non-transitory computer-readable medium of aspect 226, wherein the long-term biometrics indicate one or more of a voice biomarker, eye attentiveness, or a facial change.

[0502] 228. The non-transitory computer-readable medium of aspects 226 or 227, wherein: the first model training data further comprises historical LTB data of historical patients corresponding to the historical LTBSC data of the historical patients; and the first model is further trained to determine, for the historical patients, associations between the historical LTBSC and historical longterm biometrics corresponding to the historical LTBSC.

[0503] 229. The non-transitory computer-readable medium of any one of aspects 226 to 228, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

[0504] 230. The non-transitory computer-readable medium of any one of aspects 220 to 229, wherein generating the LTBSC data includes one or more of: analyze at least one frequency band of the at least one LTEEG signal; identify at least one LTEEG segment in the at least one LTEEG signal based upon one or more of a root means square, a standard deviation, or a power spectrum density, of the at least one LTEEG signal; determine one or more feature values of the at least one LTEEG segment by applying a principal components analysis algorithm or a support vector machine to the LTEEG data; determine at least one similar characteristic of the LTEEG data and / or the one or more feature values, across long-duration time points; classify LTEEG segments having similar characteristics; or determine an LTBSC boundary based upon detecting an edge of a phase transition of the at least one LTEEG signal.

[0505] 231 . The non-transitory computer-readable medium of aspect 230, wherein the one or more feature values are associated with one or more characteristics of an electroencephalogram (EEG) signal.

[0506] 232. The non-transitory computer-readable medium of aspects 230 or 231 , wherein the at least one frequency band is selected from the group consisting of 1 to 2 hertz (Hz), 0.5 to 4 Hz, 4 to 8 Hz, 8 to 13 Hz, 13 to 30 Hz, 30 to 100 Hz, 80 to 250 Hz, and 250 to 500 Hz.

[0507] 233. The non-transitory computer-readable medium of any one of aspects 220 to 232, wherein the at least one LTBSC is associated with one or more of epilepsy, a psychiatric disorder, a stroke, Alzheimer's disease, a sleep disorder, depression, medication efficacy, or a medication side effect.

[0508] 234. The non-transitory computer-readable medium of any one of aspects 220 to 233, wherein retraining the one or more one machine learning models using the LTBSC data includes fine-tuning the one or more machine learning models.

[0509] 235. The non-transitory computer-readable medium of any one of aspects 220 to 234, wherein the medical insight is one or more of: an indication of the at least one LTBSC; a risk of sudden unexplained death by epilepsy; or an association between the at least one LTBSC and the long-term biometrics.

[0510] 236. The non-transitory computer-readable medium of any one of aspects 220 to 235, further comprising instructions that, when executed, cause the one or more processors to: obtain additional first model training data; retrain the first model using the additional first model training data; and store the retrained first model on a memory to generate subsequent LTBSC data using the retrained first model.

[0511] 237. The non-transitory computer-readable medium of any one of aspects 220 to 236, wherein: the one or more machine learning models include a second model trained using second model training data to generate second model output data indicating a seizure of the patient in the LTEEG of the LTBSC data, based upon receiving the LTBSC data; to retrain the second model using the LTBSC data further comprises instructions that, when executed, cause the one or more processors to store the retrained second model in the memory to generate subsequent second model output data using the retrained second model; and to provide the LTBSC data as the input to the second model further comprises instructions that, when executed, cause the one or more processors to, in response to generating the second model output data, provide the second model output data to the user device.

[0512] 238. The non-transitory computer-readable medium of aspect 237, wherein the second model training data comprises historical long-term brain state change (LTBSC) data of historical patients.

[0513] 239. The non-transitory computer-readable medium of aspect 238, wherein the second model is trained to determine, for the historical patients, associations between at least one historical LTBSC and an historical seizure.

[0514] 240. The non-transitory computer-readable medium of any one of aspects 237 to 239, wherein the second model output data includes a classification of the seizure and / or a confidence metric associated with the indication of the seizure.

[0515] 241 . The non-transitory computer-readable medium of any one of aspects 237 to 240, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0516] 242. The non-transitory computer-readable medium of any one of aspects 237 to 241 , further comprising instructions that, when executed, cause the one or more processors to: obtain additional second model training data; retrain the second model using the additional second model training data; and store the retrained second model in the memory to generate subsequent second model output data using the retrained second model.

[0517] 243. The non-transitory computer-readable medium of any one of aspects 220 to 236, wherein: the one or more machine learning models is third model trained using third model training data to generate third model output data, third model output data including a revision of an indication of a seizure in the LTBSC data, based upon receiving the LTBSC data, wherein: the third model output data includes a revision of an indication of a seizure in the LTEEG data of the LTBSC data; the indication of the seizure is not based upon the at least one LTBSC of the LTBSC data; and the revision to the indication of the seizure is based upon the at least one LTBSC of the LTBSC data; to retrain the third model using the LTBSC data further comprises instructions that, when executed, cause the one or more processors to store the retrained third model in the memory to generate subsequent third model output data using the retrained third model; and to provide the LTBSC data as the input to the third model further comprises instructions that, when executed, cause the one or more processors to, in response to generating the third model output data, provide the third model output data to the user device.

[0518] 244. The non-transitory computer-readable medium of aspect 243, wherein the third model training data comprises historical LTBSC data of historical patients.

[0519] 245. The non-transitory computer-readable medium of aspects 243 or 244, wherein the third model is trained to determine, for the historical patients, associations between an historical seizure and at least one historical LTBSC.

[0520] 246. The non-transitory computer-readable medium of any one of aspects 243 to 245, wherein the revision includes a reclassification of the seizure and / or a change in a confidence metric associated with the indication of the seizure.

[0521] 247. The non-transitory computer-readable medium of any one of aspects 243 to 246, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0522] 248. The non-transitory computer-readable medium of any one of aspects 243 to 247, further comprising instructions that, when executed, cause the one or more processors to: obtain additional third model training data; retrain the third model using the additional third model training data; and store the retrained third model in the memory to generate subsequent third model output data using the retrained third model.

[0523] 249. The non-transitory computer-readable medium of any one of aspects 220 to 236, wherein: the one or more machine learning models include a fourth model trained using fourth model training data to generate fourth model output data indicating a prediction of a seizure of the patient based upon receiving the LTBSC data; to retrain the fourth model using the LTBSC data further comprises instructions that, when executed, cause the one or more processors to store the retrained fourth model in the memory to generate subsequent fourth model output data using the retrained fourth model; and to provide the LTBSC data as the input to the fourth model further comprises instructions that, when executed, cause the one or more processors to, in response to generating the fourth model output data, provide the fourth model output data to the user device.

[0524] 250. The non-transitory computer-readable medium of aspect 249, wherein the fourth model training data comprises historical LTBSC data of historical patients.

[0525] 251 . The non-transitory computer-readable medium of aspect 250, wherein the fourth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and timing of an historical seizure.

[0526] 252. The non-transitory computer-readable medium of any one of aspects 249 to 251 , wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0527] 253. The non-transitory computer-readable medium of any one of aspects 249 to 252, wherein the fourth model output data includes at least one date of the prediction of the seizure, and / or a confidence metric associated with the prediction.

[0528] 254. The non-transitory computer-readable medium of any one of aspects 249 to 253, further comprising instructions that, when executed, cause the one or more processors to: obtain additional fourth model training data; retrain the fourth model using the additional fourth model training data; and store the retrained fourth model in the memory to generate subsequent fourth model output data using the retrained fourth model.

[0529] 255. The non-transitory computer-readable medium of any one of aspects 220 to 236, wherein: the one or more machine learning models include a fifth model trained using fifth model training data to generate fifth model output data indicating at least one date to implement a medication titration of the patient based upon receiving the LTBSC data; to retrain the fifth model using the LTBSC data further comprises instructions that, when executed, cause the one or more processors to store the retrained fifth model in the memory to generate subsequent fifth model output data using the retrained fifth model; and to provide the LTBSC data as the input to the fifth model further comprises instructions that, when executed, cause the one or more processors to, in response to generating the fifth model output data, provide the fifth model output data to the user device.

[0530] 256. The non-transitory computer-readable medium of aspect 255, wherein the fifth model training data comprises historical LTBSC data of historical patients.

[0531] 257. The non-transitory computer-readable medium of aspects 255 or 256, wherein the fifth model is trained to determine, for the historical patients, associations between at least one historical LTBSC, timing of an historical medication titration, and associated historical side effects from the historical medication titration.

[0532] 258. The non-transitory computer-readable medium of any one of aspects 255 to 257, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0533] 259. The non-transitory computer-readable medium of any one of aspects 255 to 258, further comprising instructions that, when executed, cause the one or more processors to: obtain additional fifth model training data; retrain the fifth model using the additional fifth model training data; and store the retrained fifth model in the memory to generate subsequent fifth model output data using the retrained fifth model.

[0534] 260. The non-transitory computer-readable medium of any one of aspects 220 to 236, wherein: the one or more machine learning models include a sixth model trained using sixth model training data to generate sixth model output data indicating an association between at least one environmental characteristic of the patient and the at least one LTBSC of the patient, based upon receiving environmental data indicating the at least one environmental characteristic of the patientand the LTBSC data; to retraining the sixth model using the LTBSC data further comprises instructions that, when executed, cause the one or more processors to store the retrained sixth model in the memory to generate subsequent sixth model output data using the retrained sixth model; and to provide the LTBSC data as the input to the sixth model further comprises instructions that, when executed, cause the one or more processors to: obtain the environmental data associated with the LTBSC data; provide the environmental data and the LTBSC data to the sixth model to generate the sixth model output data; and in response to generating the sixth model output data, provide the sixth model output data to the user device.

[0535] 261 . The non-transitory computer-readable medium of aspect 260, wherein the sixth model training data comprises historical LTBSC data of historical patients and historical environmental data of the historical patients.

[0536] 262. The non-transitory computer-readable medium of aspect 261 , wherein the sixth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and at least one historical environmental characteristic.

[0537] 263. The non-transitory computer-readable medium of any one of aspects 260 to 262, wherein the at least one environmental characteristic is associated with a medication.

[0538] 264. The non-transitory computer-readable medium of aspect 263, wherein the association with the medication includes a new medication taken by the patient, and / or a new titration of the medication taken by the patient.

[0539] 265. The non-transitory computer-readable medium of any one of aspects 260 to 264, wherein: the environmental data is obtained in response to a LTBSC of the patient; and the sixth model output data indicates a cause of the LTBSC.

[0540] 266. The non-transitory computer-readable medium of any one of aspects 260 to 265, wherein the sixth model output data indicates a prediction of a future LTBSC in response to the at least one environmental characteristic, and / or a treatment pathway to a target brain state.

[0541] 267. The non-transitory computer-readable medium of any one of aspects 260 to 266, wherein the environmental data is generated by an implant device of the patient.

[0542] 268. The non-transitory computer-readable medium of any one of aspects 260 to 267, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

[0543] 269. The non-transitory computer-readable medium of any one of aspects 260 to 268, further comprising instructions that, when executed, cause the one or more processors to: obtain additional sixth model training data; retrain the sixth model using the additional sixth model trainingdata; and store the retrained sixth model in the memory to generate subsequent sixth model output data using the retrained sixth model.

Claims

CLAIMS1 . A computer-implemented method for determining a long-term brain state change (LTBSC) of a patient, the method comprising: obtaining, by one or more processors, long-term electroencephalogram (LTEEG) data comprising at least one LTEEG signal associated with a brain of the patient and gathered via an implant device of the patient; providing, by the one or more processors, the LTEEG data to a first model trained using first model training data to generate LTBSC data indicating at least one LTBSC of the patient; in response to generating the LTBSC data, one or more of:(i) retraining, by the one or more processors, one or more machine learning models using the LTBSC data;(ii) providing, by the one or more processors, the LTBSC data as an input to the one or more machine learning models; or(iii) generating, by the one or more processors, a medical insight of the patient based upon the LTBSC data; and providing, by the one or more processors, the medical insight to a user device.

2. The computer-implemented method of claim 1 , wherein the LTEEG data includes at least one substantially continuous LTEEG signal of the patient gathered over more than a week of time.

3. The computer-implemented method of claims 1 or 2, wherein: the first model training data comprises historical LTEEG data of historical patients and historical LTBSC data of historical patients corresponding to the historical LTEEG data.

4. The computer-implemented method of claim 3, wherein: the first model is trained to determine, for the historical patients, associations between an historical LTBSC and at least one historical LTEEG signal corresponding to the historical LTBSC.

5. The computer-implemented method of any one of claims 1 to 4, wherein the LTBSC data includes at least a portion of the LTEEG data corresponding to the at least one LTBSC.

6. The computer-implemented method of any one of claims 1 to 5, wherein the LTBSC data includes a classification and / or confidence metric associated with the at least one LTBSC.

7. The computer-implemented method of any one of claims 1 to 6, further comprising: obtaining, by the one or more processors, long-term biometric (LTB) data indicating long-term biometrics of the patient corresponding to the at least one LTBSC; and providing, by the one or more processors, the LTB data to the first model, wherein generating the LTBSC data is further based upon the LTB data.

8. The computer-implemented method of claim 7, wherein the long-term biometrics indicate one or more of a voice biomarker, eye attentiveness, or a facial change.

9. The computer-implemented method of claims 7 or 8, wherein: the first model training data further comprises historical LTB data of historical patients corresponding to the historical LTBSC data of the historical patients; and the first model is further trained to determine, for the historical patients, associations between the historical LTBSC and historical long-term biometrics corresponding to the historical LTBSC.

10. The computer-implemented method of any one of claims 7 to 9, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.11 . The computer-implemented method of any one of claims 1 to 10, wherein generating the LTBSC data includes one or more of: analyzing, by the one or more processors, at least one frequency band of the at least one LTEEG signal; identifying, by the one or more processors, at least one LTEEG segment in the at least one LTEEG signal based upon one or more of a root means square, a standard deviation, or a power spectrum density, of the at least one LTEEG signal;determining, by the one or more processors, one or more feature values of the at least one LTEEG segment by applying a principal components analysis algorithm or a support vector machine to the LTEEG data; determining, by the one or more processors, at least one similar characteristic of the LTEEG data and / or the one or more feature values, across long-duration time points; classifying, by the one or more processors, LTEEG segments having similar characteristics; or determining, by the one or more processors, an LTBSC boundary based upon detecting an edge of a phase transition of the at least one LTEEG signal.

12. The computer-implemented method of claim 1 1 , wherein the one or more feature values are associated with one or more characteristics of an electroencephalogram (EEG) signal.

13. The computer-implemented method of claims 1 1 or 12, wherein the at least one frequency band is selected from the group consisting of 1 to 2 hertz (Hz), 0.5 to 4 Hz, 4 to 8 Hz, 8 to 13 Hz, 13 to 30 Hz, 30 to 100 Hz, 80 to 250 Hz, and 250 to 500 Hz.

14. The computer-implemented method of any one of claims 1 to 13, wherein the at least one LTBSC is associated with one or more of epilepsy, a psychiatric disorder, a stroke, Alzheimer's disease, a sleep disorder, depression, medication efficacy, or a medication side effect.

15. The computer-implemented method of any one of claims 1 to 14, wherein retraining the one or more one machine learning models using the LTBSC data includes fine-tuning the one or more machine learning models.

16. The computer-implemented method of any one of claims 1 to 15, wherein the medical insight is one or more of: an indication of the at least one LTBSC; a risk of sudden unexplained death by epilepsy; or an association between the at least one LTBSC and the long-term biometrics.

17. The computer-implemented method of any one of claims 1 to 16, further comprising: obtaining, by the one or more processors, additional first model training data; retraining, by the one or more processors, the first model using the additional first model training data; and storing, by the one or more processors, the retrained first model on a memory to generate subsequent LTBSC data using the retrained first model.

18. The computer-implemented method of any one of claims 1 to 17, wherein: the one or more machine learning models include a second model trained using second model training data to generate second model output data indicating a seizure of the patient in the LTEEG of the LTBSC data, based upon receiving the LTBSC data; retraining the second model using the LTBSC data further comprises storing, by the one or more processors, the retrained second model in the memory to generate subsequent second model output data using the retrained second model; and providing the LTBSC data as the input to the second model further comprises, in response to generating the second model output data, providing, by the one or more processors, the second model output data to the user device.

19. The computer-implemented method of claim 18, wherein the second model training data comprises historical LTBSC data of historical patients.

20. The computer-implemented method of claim 19, wherein the second model is trained to determine, for the historical patients, associations between at least one historical LTBSC and an historical seizure.21 . The computer-implemented method of any one of claims 18 to 20, wherein the second model output data includes a classification of the seizure and / or a confidence metric associated with the indication of the seizure.

22. The computer-implemented method of any one of claims 18 to 21 , wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

23. The computer-implemented method of any one of claims 18 to 22, further comprising: obtaining, by the one or more processors, additional second model training data; retraining, by the one or more processors, the second model using the additional second model training data; and storing, by the one or more processors, the retrained second model in the memory to generate subsequent second model output data using the retrained second model.

24. The computer-implemented method of any one of claims 1 to 17, wherein: the one or more machine learning models include a third model trained using third model training data to generate third model output data, based upon receiving the LTBSC data, wherein: the third model output data includes a revision of an indication of a seizure in the LTEEG data of the LTBSC data; the indication of the seizure is not based upon the at least one LTBSC of the LTBSC data; and the revision to the indication of the seizure is based upon the at least one LTBSC of the LTBSC data; retraining the third model using the LTBSC data further comprises storing, by the one or more processors, the retrained third model in the memory to generate subsequent third model output data using the retrained third model; and providing the LTBSC data as the input to the third model further comprises, in response to generating the third model output data, providing, by the one or more processors, the third model output data to the user device.

25. The computer-implemented method of claim 24, wherein the third model training data comprises historical LTBSC data of historical patients.

26. The computer-implemented method of claims 25, wherein the third model is trained to determine, for the historical patients, associations between an historical seizure and at least one historical LTBSC.

27. The computer-implemented method of any one of claims 24 to 26, wherein the revision includes a reclassification of the seizure and / or a change in a confidence metric associated with the indication of the seizure.

28. The computer-implemented method of any one of claims 24 to 27, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

29. The computer-implemented method of any one of claims 24 to 28, further comprising: obtaining, by the one or more processors, additional third model training data; retraining, by the one or more processors, the third model using the additional third model training data; and storing, by the one or more processors, the retrained third model in the memory to generate subsequent third model output data using the retrained third model.

30. The computer-implemented method of any one of claims 1 to 17, wherein: the one or more machine learning models include a fourth model trained using fourth model training data to generate fourth model output data indicating a prediction of a seizure of the patient, based upon receiving the LTBSC data; retraining the fourth model using the LTBSC data further comprises storing, by the one or more processors, the retrained fourth model in the memory to generate subsequent fourth model output data using the retrained fourth model; and providing the LTBSC data as the input to the fourth model further comprises, in response to generating the fourth model output data, providing, by the one or more processors, the fourth model output data to the user device.31 . The computer-implemented method of claim 30, wherein the fourth model training data comprises historical LTBSC data of historical patients.

32. The computer-implemented method of claim 31 , wherein the fourth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and timing of an historical seizure.

33. The computer-implemented method of any one of claims 30 to 32, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

34. The computer-implemented method of any one of claims 30 to 33, wherein the fourth model output data includes at least one date of the prediction of the seizure, and / or a confidence metric associated with the prediction.

35. The computer-implemented method of any one of claims 30 to 34, further comprising: obtaining, by the one or more processors, additional fourth model training data; retraining, by the one or more processors, the fourth model using the additional fourth model training data; and storing, by the one or more processors, the retrained fourth model in the memory to generate subsequent fourth model output data using the retrained fourth model.

36. The computer-implemented method of any one of claims 1 to 17, wherein: the one or more machine learning models include a fifth model trained using fifth model training data to generate fifth model output data indicating at least one date to implement a medication titration of the patient, based upon receiving the LTBSC data; retraining the fifth model using the LTBSC data further comprises storing, by the one or more processors, the retrained fifth model in the memory to generate subsequent fifth model output data using the retrained fifth model; and providing the LTBSC data as the input to the fifth model further comprises, in response to generating the fifth model output data, providing, by the one or more processors, the fifth model output data to the user device.

37. The computer-implemented method of claim 36, wherein the fifth model training data comprises historical LTBSC data of historical patients.

38. The computer-implemented method of claim 37, wherein the fifth model is trained to determine, for the historical patients, associations between at least one historical LTBSC, timing of an historical medication titration, and associated historical side effects from the historical medication titration.

39. The computer-implemented method of any one of claims 36 to 38, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

40. The computer-implemented method of any one of claims 36 to 39, further comprising: obtaining, by the one or more processors, additional fifth model training data; retraining, by the one or more processors, the fifth model using the additional fifth model training data; and storing, by the one or more processors, the retrained fifth model in the memory to generate subsequent fifth model output data using the retrained fifth model.41 . The computer-implemented method of any one of claims 1 to 17, wherein: the one or more machine learning models include a sixth model trained using sixth model training data to generate sixth model output data indicating an association between at least one environmental characteristic of the patient and the at least one LTBSC of the patient, based upon receiving environmental data indicating the at least one environmental characteristic of the patient and the LTBSC data; retraining the sixth model using the LTBSC data further comprises storing, by the one or more processors, the retrained sixth model in the memory to generate subsequent sixth model output data using the retrained sixth model; and providing the LTBSC data as the input to the sixth model further comprises: obtaining, by the one or more processors, the environmental data associated with the LTBSC data; providing, by the one or more processors, the environmental data and the LTBSC data to the sixth model to generate the sixth model output data; and in response to generating the sixth model output data, providing, by the one or more processors, the sixth model output data to the user device.

42. The computer-implemented method of claim 41 , wherein the sixth model training data comprises historical LTBSC data of historical patients and historical environmental data of the historical patients.

43. The computer-implemented method of claim 42, wherein the sixth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and at least one historical environmental characteristic.

44. The computer-implemented method of any one of claims 41 to 43, wherein the at least one environmental characteristic is associated with a medication.

45. The computer-implemented method of claim 44, wherein the association with the medication includes a new medication taken by the patient, and / or a new titration of the medication taken by the patient.

46. The computer-implemented method of any one of claims 41 to 45, wherein: the environmental data is obtained in response to a LTBSC of the patient; and the sixth model output data indicates a cause of the LTBSC.

47. The computer-implemented method of any one of claims 41 to 46, wherein the sixth model output data indicates a prediction of a future LTBSC in response to the at least one environmental characteristic, and / or a treatment pathway to a target brain state.

48. The computer-implemented method of any one of claims 41 to 47, wherein the environmental data is generated by an implant device of the patient.

49. The computer-implemented method of any one of claims 41 to 48, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

50. The computer-implemented method of any one of claims 41 to 48, further comprising: obtaining, by the one or more processors, additional sixth model training data; retraining, by the one or more processors, the sixth model using the additional sixth model training data; and storing, by the one or more processors, the retrained sixth model in the memory to generate subsequent sixth model output data using the retrained sixth model.51 . A system for determining a long-term brain state change (LTBSC) of a patient, the system comprising: an implant device comprising a sensor array and a processor device, the sensor array comprising a plurality of electrodes and a local processing device comprising local device communication circuitry, the sensor array being communicatively coupled to the processor device via the local processing device communication circuitry, the sensor array being configured to gather long-term electroencephalogram (LTEEG) data comprising at least one LTEEG signal associated with a brain of the patient, and provide the LTEEG data to the processing device; and the processor device comprising a microprocessor, a memory, and communication circuitry, the processing device being communicatively coupled to the sensor array via the communication circuitry; a first model, stored in the memory and configured to be executed by the microprocessor, the first model trained using first model training data and operable to: receive the LTEEG data; determine one or more feature values of the LTEEG data; and based on the one or more feature values, generate LTBSC data indicating at least one LTBSC of the patient; and in response to generating the LTBSC data, the processor device is configured to one or more of:(i) retrain one or more machine learning models using the LTBSC data;(ii) provide the LTBSC data as an input to the one or more machine learning models; or(iii) generate a medical insight of the patient based upon the LTBSC data; and provide the medical insight to a user device.

52. The system of claim 51 , wherein the processor device is a wearable processor device.

53. The system of claim 51 , wherein the processor device is embedded in the sensor array or disposed adjacent to the sensor array.

54. The system of claim 51 , wherein the processor device communicates wirelessly with a server.

55. The system of claim 51 , wherein the processor device communicates wirelessly with the user device.

56. The system of claim 55, wherein the user device is configured to communicate data from the processor device to one or more servers via an Internet.

57. The system of claim 51 , wherein the processor device is the user device.

58. The system of claim 51 , wherein the processor device is a server.

59. The system of claim 51 , wherein the processor device is remote from the sensor array and is communicatively coupled to the sensor array via an Internet.

60. The system of any one of claims 51 to 59, wherein the plurality of electrodes provide one or more electrical signals indicating a presence or absence of a biomarker.61 . The system of claim 60, wherein the biomarker includes epileptiform activity biomarker data determined from the LTEEG data received from one or more sensor arrays.

62. The system of any one of claims 50 to 61 , wherein the sensor array comprises a wireless transceiver.

63. The system of any one of claims 51 to 62, wherein the LTEEG data includes at least one substantially continuous LTEEG signal of the patient gathered over more than a week of time.

64. The system of any one of claims 51 to 63, wherein: the first model training data comprises historical LTEEG data and historical LTBSC data of historical patients corresponding to the historical LTEEG data.

65. The system of claim 64, wherein:the first model is trained to determine, for the historical patients, associations between an historical LTBSC and at least one historical LTEEG signal corresponding to the historical LTBSC.

66. The system of any one of claims 51 to 65, wherein the LTBSC data includes at least a portion of the LTEEG data corresponding to the at least one LTBSC.

67. The system of any one of claims 51 to 66, wherein the LTBSC data includes a classification and / or confidence metric associated with the at least one LTBSC.

68. The system of any one of claims 51 to 67, wherein the one or more feature values of the LTEEG data include one or more characteristics of an electroencephalogram (EEG) signal.

69. The system of any one of claims 51 to 68, wherein: the processor device is further configured to obtain long-term biometric (LTB) data indicating long-term biometrics of the patient corresponding to the at least one LTBSC; and the first model, when executed by the microprocessor, is further operable to: receive the LTB data; and determine one or more feature values of the LTB data, wherein generating the LTBSC data is further based upon the LTB data.

70. The system of claim 69, wherein the long-term biometrics indicate one or more of a voice biomarker, eye attentiveness, or a facial change.71 . The system of claims 69 or 70, wherein the one or more feature values of the LTB data include biometrics of the face and / or eyes.12.. The system of any one of claims 69 to 71 , wherein: the first model training data further comprises historical LTB data of historical patients corresponding to the historical LTBSC data of the historical patients; and the first model is further trained to determine, for the historical patients, associations between the historical LTBSC and historical long-term biometrics corresponding to the historical LTBSC.

73. The system of any one of claims 69 to 72, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

74. The system of any one of claims 51 to 73, wherein to generate the LTBSC, the first model, when executed by the microprocessor, is further operable to: analyze at least one frequency band of the at least one LTEEG signal; identify at least one LTEEG segment in the at least one LTEEG signal based upon one or more of a root means square, a standard deviation, or a power spectrum density, of the at least one LTEEG signal; determine one or more feature values of the at least one LTEEG segment by applying a principal components analysis algorithm or a support vector machine to the LTEEG data; determine at least one similar characteristic of the LTEEG data and / or the one or more feature values, across long-duration time points; classify LTEEG segments having similar characteristics; or determine an LTBSC boundary based upon detecting an edge of a phase transition of the at least one LTEEG signal.

75. The system of claim 74, wherein the one or more feature values of the LTEEG segment include one or more characteristics of an electroencephalogram (EEG) signal.

76. The system of claims 74 or 75, wherein the at least one frequency band is selected from the group consisting of 1 to 2 hertz (Hz), 0.5 to 4 Hz, 4 to 8 Hz, 8 to 13 Hz, 13 to 30 Hz, 30 to 100 Hz, 80 to 250 Hz, and 250 to 500 Hz.

77. The system of any one of claims 51 to 76, wherein the at least one LTBSC is associated with one or more of epilepsy, a psychiatric disorder, a stroke, Alzheimer's disease, a sleep disorder, depression, medication efficacy, or a medication side effect.

78. The system of any one of claims 51 to 77, wherein retraining the one or more one machine learning models using the LTBSC data includes fine-tuning the one or more machine learning models.

79. The system of any one of claims 51 to 78, wherein the medical insight is one or more of: an indication of the at least one LTBSC; a risk of sudden unexplained death by epilepsy; or an association between the at least one LTBSC and the long-term biometrics.

80. The system of any one of claims 51 to 79, wherein the processor device is further configured to: receive additional first model training data; retrain the first model using the additional first model training data; and store the retrained first model in the memory to generate subsequent LTBSC data using the retrained first model.81 . The system of any one of claims 51 to 80, wherein: the one or more machine leaning models include a second model, stored in the memory and configured to be executed by the microprocessor, the second model trained using second model training data and operable to generate second model output data indicating a seizure of the patient in the LTEEG of the LTBSC data, based upon receiving the LTBSC data; to retrain the second model using the LTBSC data, the processor device is further configured to store the retrained second model in the memory to generate subsequent second model output data using the retrained second model; and to provide the LTBSC data as the input to the second model, the processor device is further configured to, in response to the second model generating the second model output data, provide the second model output data to the user device.

82. The system of claim 81 , wherein the second model training is trained using second model training data comprising historical LTBSC data of historical patients.

83. The system of claim 82, wherein the second model is trained to determine, for the historical patients, associations between at least one historical LTBSC and an historical seizure.

84. The system of any one of claims 81 to 83, wherein the second model output data includes a classification of the seizure and / or a confidence metric associated with the indication of the seizure.

85. The system of any one of claims 81 to 84, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

86. The system of any one of claims 81 to 85, wherein the processor is further configured to: obtain additional second model training data; retrain the second model using the additional second model training data; and store the retrained second model in the memory to generate subsequent second model output data using the retrained second model.

87. The system of any one of claims 51 to 80, wherein: the one or more machine leaning models include third model, stored in the memory and configured to be executed by the microprocessor, the third model trained using third model training data and operable to generate third model output data, based upon receiving the LTBSC data, wherein: the third model output data includes a revision of an indication of a seizure in the LTEEG data of the LTBSC data; the indication of the seizure is not based upon the at least one LTBSC of the LTBSC data; and the revision to the indication of the seizure is based upon the at least one LTBSC of the LTBSC data; to retrain the third model using the LTBSC data, the processor device is further configured to store the retrained third model in the memory to generate subsequent third model output data using the retrained third model; and to provide the LTBSC data as the input to the third model, the processor device is further configured to, in response to generating the third model output data, provide the third model output data to the user device.

88. The system of claim 87, wherein the third model is trained using third model training data comprising historical LTBSC data of historical patients.

89. The system of claim 88, wherein the third model is trained to determine, for the historical patients, associations between an historical seizure and at least one historical LTBSC.

90. The system of any one of claims 87 to 89, wherein the revision includes a reclassification of the seizure and / or a change in a confidence metric associated with the indication of the seizure.91 . The system of any one of claims 87 to 90, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

92. The system of any one of claims 87 to 91 , wherein the processor device is further configured to: receive additional third model training data; retrain the third model using the additional third model training data; and store the retrained third model in the memory to generate subsequent third model output data using the retrained third model.

93. The system any one of claims 51 to 80, wherein: the one or more machine leaning models include a fourth model, stored in the memory and configured to be executed by the microprocessor, the fourth model trained using fourth model training data and operable to generate fourth model output data indicating a prediction of a seizure of the patient, based upon receiving the LTBSC data; to retraining the fourth model using the LTBSC data, the processor device is further configured to store the retrained fourth model in the memory to generate subsequent fourth model output data using the retrained fourth model; and to provide the LTBSC data as the input to the fourth model, the processor device is further configured to, in response to the fourth model generating the fourth model output data, provide the fourth model output data to the user device.

94. The system of claim 93, wherein the fourth model is trained using fourth model training data comprising historical LTBSC data of historical patients.

95. The system of claim 94, wherein the fourth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and timing of an historical seizure.

96. The system of any one of claims 93 to 95, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

97. The system of any one of claims 93 to 96, wherein the fourth model output data includes at least one date of the prediction of the seizure, and / or a confidence metric associated with the prediction.

98. The system of any one of claims 93 to 97, wherein the processor device is further configured to: receive additional fourth model training data; retrain the fourth model using the additional fourth model training data; and store the retrained fourth model in the memory to generate subsequent fourth model output data using the retrained fourth model.

99. The system any one of claims 51 to 80, wherein the one or more machine leaning models include a fifth model, stored in the memory and configured to be executed by the microprocessor, the fifth model trained using fifth model training data and operable to generate fifth model output data indicating at least one date to implement a medication titration of the patient based upon receiving the LTBSC data; to retrain the fifth model using the LTBSC data, the processor device is further configured to store the retrained fifth model in the memory to generate subsequent fifth model output data using the retrained fifth model; and to provide the LTBSC data as the input to the fifth model, the processor device is further configured to, in response to the fifth model generating the fifth model output data, provide the fifth model output data to the user device.

100. The system of claim 99, wherein the fifth model training data comprises historical LTBSC data of historical patients.101 . The system of claim 100, wherein the fifth model is trained to determine, for the historical patients, associations between at least one historical LTBSC, timing of an historical medication titration, and associated historical side effects from the historical medication titration.

102. The system of any one of claims 99 to 101 , wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

103. The system of any one of claims 99 to 102, wherein the processor device is further configured to: receive additional fifth model training data; retrain the fifth model using the additional fifth model training data; and store the retrained fifth model in the memory to generate subsequent fifth model output data using the retrained fifth model.

104. The system of any one of claims 51 to 80, wherein: the one or more machine leaning models include a sixth model, stored in the memory and configured to be executed by the microprocessor, the sixth model trained using sixth model training data and operable to generate sixth model output data indicating an association between at least one environmental characteristic of the patient and the at least one LTBSC of the patient, based upon receiving environmental data indicating the at least one environmental characteristic of the patient and the LTBSC data; to retrain the sixth model using the LTBSC data, the processor device is further configured to store the retrained sixth model in the memory to generate subsequent sixth model output data using the retrained sixth model; and to provide the LTBSC data as the input to the sixth model, the processor device is further configured: obtain the environmental data associated with the LTBSC data; provide the environmental data and the LTBSC data to the sixth model to generate the sixth model output data; andin response to generating the sixth model output data by the sixth model, provide the sixth model output data to the user device.

105. The system of claim 104, wherein the sixth model training data comprises historical LTBSC data of historical patients and historical environmental data of the historical patients.

106. The system of claim 105, wherein the sixth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and at least one historical environmental characteristic.

107. The system of any one of claims 104 to 106, wherein the at least one environmental characteristic is associated with a medication.

108. The system of any one of claim 57, wherein the association with the medication includes a new medication taken by the patient, and / or a new titration of the medication taken by the patient.

109. The system of any one of claims 104 to 108, wherein: the environmental data is obtained in response to a LTBSC of the patient; and the sixth model output data indicates a cause of the LTBSC.

110. The system of any one of claims 104 to 109, wherein the sixth model output data indicates a prediction of a future LTBSC in response to the at least one environmental characteristic, and / or a treatment pathway to a target brain state.11 1. The system of any one of claims 104 to 110, wherein the environmental data is generated by the implant device of the patient.

112. The system of any one of claims 104 to 11 1 , wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

113. The system of any one of claims 104 to 112, wherein the processor device is further configured to:receive additional sixth model training data; retrain the sixth model using the additional sixth model training data; and store the retrained sixth model in the memory to generate subsequent sixth model output data using the retrained sixth model.

114. A first model for determining a long-term brain state change (LTBSC) of a patient, the first model comprising: the first model, stored on one or more memories and configured to be executed by one or more processors, the first model trained using first model training data and operable to: receive long-term electroencephalogram (LTEEG) data comprising at least one LTEEG signal associated with a brain of the patient and gathered via an implant device of the patient; determine one or more feature values of the LTEEG data; and based on the one or more feature values, generate LTBSC data indicating at least one LTBSC of the patient.

115. The first model of claim 1 14, wherein the first model is a machine learning model.

116. The first model of claim 1 14, wherein the first model is based upon a static algorithm.

117. The first model of any one of claims 1 14 to 1 16, wherein the LTEEG data includes at least one substantially continuous LTEEG signal of the patient gathered over more than a week of time.

118. The first model of any one of claims 1 14 to 1 17, wherein the first model training data comprises historical LTEEG data of historical patients and historical LTBSC data of historical patients corresponding to the historical LTEEG data.

119. The first model of claim 1 18, wherein the first model is trained to determine, for the historical patients, associations between an historical LTBSC and at least one historical LTEEG signal corresponding to the historical LTBSC.

120. The first model of any one of claims 1 14 to 1 19, wherein the LTBSC data includes at least a portion of the LTEEG data corresponding to the at least one LTBSC.

121. The first model of any one of claims 1 14 to 120, wherein the LTBSC data includes a classification and / or confidence metric associated with the at least one LTBSC.

122. The first model of any one of claims 1 14 to 121 , the first model further configured, when executed by the one or more processors, to be operable to: obtain long-term biometric (LTB) data indicating long-term biometrics of the patient corresponding to the at least one LTBSC; and determine one or more feature values of the LTB data, wherein generating the LTBSC data is further based upon the LTB data.

123. The first model of claim 122, wherein the long-term biometrics indicate one or more of a voice biomarker, eye attentiveness, or a facial change.

124. The first model of claims 122 or 123, wherein: the first model training data further comprises historical LTB data of historical patients corresponding to the historical LTBSC data of the historical patients; and the first model is further trained to determine, for the historical patients, associations between the historical LTBSC and historical long-term biometrics corresponding to the historical LTBSC.

125. The first model of any one of claims 122 to 124, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

126. The first model of any one of claims 1 14 to 125, wherein the one or more feature values are associated with one or more characteristics of an electroencephalogram (EEG) signal.

127. The first model of any one of claims 1 14 to 126, wherein the at least one LTBSC is associated with one or more of epilepsy, a psychiatric disorder, a stroke,Alzheimer's disease, a sleep disorder, depression, medication efficacy, or a medication side effect.

128. A second model trained using second model training data to generate second model output data indicating a seizure of a patient in long-term brain state change (LTBSC) data, the second model stored on one or more memories and configured to be executed by one or more processors, the second model operable to: receive the LTBSC data indicating at least one LTBSC of the patient; determine one or more feature values of the LTBSC data; and based on the one or more feature values, generate the second model output data.

129. The second model of claim 128, wherein the second model is a machine learning model.

130. The second model of claim 128, wherein the second model is based upon a static algorithm.131 . The second model of any one of claims 128 to 130, wherein the second model training data comprises historical LTBSC data of historical patients.

132. The second model of claim 131 , wherein the second model is trained to determine, for the historical patients, associations between at least one historical LTBSC and an historical seizure.

133. The second model of any one of claims 128 to 132, wherein the second model output data includes a classification of the seizure and / or a confidence metric associated with the indication of the seizure.

134. The second model of any one of claims 128 to 133, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

135. A fourth model trained using fourth model training data to generate fourth model output data indicating a prediction of a seizure of a patient, the fourth model, storedon one or more memories and configured to be executed by one or more processors, the fourth model operable to: receive long-term brain state change (LTBSC) data indicating at least one LTBSC of the patient; determine one or more feature values of the LTBSC data; and based on the one or more feature values, generate the fourth model output data.

136. The fourth model of claim 135, wherein the fourth model is a machine learning model.

137. The fourth model of claim 135, wherein the fourth model is based upon a static algorithm.

138. The fourth model of any one of claims 135 to 137, wherein the fourth model training data comprises historical LTBSC data of historical patients.

139. The fourth model of claim 138, wherein the fourth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and timing of an historical seizure.

140. The fourth model of any one of claims 135 to 139, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

141. The fourth model of any one of claims 135 to 140, wherein the fourth model output data includes a date of the prediction of the seizure, and / or a confidence metric associated with the prediction.

142. A third model trained using third model training data to generate third model output data, the third model, stored on one or more memories and configured to be executed by one or more processors, the third model operable to: receive long-term brain state change (LTBSC) data indicating at least one LTBSC of a patient; determine one or more feature values of the LTBSC data; andbased on the one or more feature values, generate the third model output data, wherein: the third model output data includes a revision of an indication of a seizure in LTEEG data of the LTBSC data; the indication of the seizure is not based upon the at least one LTBSC of the LTBSC data; and the revision to the indication of the seizure is based upon the at least one LTBSC of the LTBSC data.

143. The third model of claim 142, wherein the third model is a machine learning model.

144. The third model of claim 142, wherein the third model is based upon a static algorithm.

145. The third model of any one of claims 142 to 144, wherein the third model training data comprises historical LTBSC data of historical patients.

146. The third model of claim 142, wherein the third model is trained to determine, for the historical patients, associations between an historical seizure and at least one historical LTBSC.

147. The third model of any one of claims 142 to 146, wherein the revision includes a reclassification of the seizure and / or a change in a confidence metric associated with the indication of the seizure.

148. The third model of any one of claims 142 to 147, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

149. A fifth model trained using fifth model training data to generate fifth model output data indicating at least one date to implement a medication titration of a patient, the fifth model, stored on one or more memories and configured to be executed by one or more processors, the fifth model operable to:receive long-term brain state change (LTBSC) data indicating at least one LTBSC of the patient; determine one or more feature values of the LTBSC data; and based on the one or more feature values, generate the fifth model output data.

150. The fifth model of claim 149, wherein the fifth model is a machine learning model.151 . The fifth model of claim 149, wherein the fifth model is based upon a static algorithm.

152. The fifth model of any one of claims 149 to 151 , wherein the fifth model training data comprises historical LTBSC data of historical patients.

153. The fifth model of claim 152, wherein the fifth model is trained to determine, for the historical patients, associations between at least one historical LTBSC, timing of an historical medication titration, and associated historical side effects from the historical medication titration.

154. The fifth model of any one of claims 149 to 153, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

155. A sixth model trained using sixth model training data to generate sixth model output data indicating an association between at least one environmental characteristic of a patient and the at least one long-term brain state change (LTBSC) of the patient, the sixth model, stored on one or more memories and configured to be executed by one or more processors, the sixth model operable to: obtain LTBSC data indicating at least one LTBSC of the patient, and environmental data indicating the at least one environmental characteristic of the patient; determine one or more feature values of the LTBSC data and the environmental data; and based on the one or more feature values, generate the sixth model output data.

156. The sixth model of claim 155, wherein the sixth model is a machine learning model.

157. The sixth model of claim 155, wherein the sixth model is based upon a static algorithm.

158. The sixth model of any one of claims 155 to 157, wherein the sixth model training data comprises historical LTBSC data of historical patients and historical environmental data of the historical patients.

159. The sixth model of claim 158, wherein the sixth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and at least one historical environmental characteristic.

160. The sixth model of any one of claims 155 to 159, wherein the at least one environmental characteristic is associated with a medication.161 . The sixth model of claim 160, wherein the association with the medication includes a new medication taken by a patient, and / or a new titration of the medication taken by the patient.

162. The sixth model of any one of claims 155 to 161 , wherein: the environmental data is obtained in response to a LTBSC of the patient; and the sixth model output data indicates a cause of the LTBSC.

163. The sixth model of any one of claims 155 to 162, wherein the sixth model output data indicates a prediction of a future LTBSC in response to the at least one environmental characteristic, and / or a treatment pathway to a target brain state.

164. The sixth model of any one of claims 155 to 163, wherein the environmental data is generated by an implant device of the patient.

165. The sixth model of any one of claims 155 to 164, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

166. A computer-implemented method for training a first model to determine a long-term brain state change (LTBSC) of a patient, the method comprising: obtaining, by one or more processors, a first training dataset; receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; training, by the one or more processors, a first iteration of the first model using the first training dataset using the feature values defined for the first training dataset to generate primary LTBSC data having associated first error rates; obtaining, by the one or more processors, a second training dataset; receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and training, by the one or more processors, a second iteration of the first model using the second training dataset using the feature values defined for the second training dataset to generate secondary LTBSC data having associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

167. The computer-implemented method of claim 166, wherein the one or more attributes comprise one or more characteristics of an electroencephalogram (EEG) signal.

168. The computer-implemented method of claims 166 or 167, wherein a training dataset comprises historical long-term EEG (LTEEG) data of historical patients and historical LTBSC data of historical patients corresponding to the historical LTEEG data.

169. The computer-implemented method of any one of claim 168, wherein the first model is trained to determine, for the historical patients, associations between an historical LTBSC and at least one historical LTEEG signal corresponding to the historical LTBSC.

170. The computer-implemented method of claim 168 or 169, wherein the LTBSC data includes at least a portion of the LTEEG data corresponding to the at least one LTBSC.

171. The computer-implemented method of any one of claims 166 to 170, wherein:the training dataset further comprises historical LTB data of historical patients corresponding to the historical LTBSC data of the historical patients; and the first model is further trained to determine, for the historical patients, associations between the historical LTBSC and historical long-term biometrics corresponding to the historical LTBSC.112.. The computer-implemented method of claim 171 , wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

173. The computer-implemented method of any one of claims 166 to 172, comprising: obtaining, by the one or more processors, LTEEG data associated with one or more patients; and based upon the LTEEG data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

174. The computer-implemented method of any one of claims 166 to 173, further comprising: receiving one or more labels for the first training dataset, wherein training the first iteration of the first model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the first LTBSC data; and receiving one or more labels for the second training dataset, wherein training the second iteration of the first model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the second LTBSC data.

175. A computer-implemented method for training a second model to generate second model output data indicating a seizure of the patient, the method comprising: obtaining, by one or more processors, a first training dataset; receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; training, by the one or more processors, a first iteration of the second model using the first training dataset using the feature values defined for the first training dataset to generate primary second model output data having associated first error rates; obtaining, by the one or more processors, a second training dataset;receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and training, by the one or more processors, a second iteration of the second model using the second training dataset using the feature values defined for the second training dataset to generate secondary second model output data having associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

176. The computer-implemented method of claim 175, wherein the one or more attributes comprise one or more characteristics of an electroencephalogram (EEG) signal.

177. The computer-implemented method of claims 175 or 176, wherein a training dataset comprises historical LTBSC data of historical patients.

178. The computer-implemented method of claim 177, wherein the second model is trained to determine, for the historical patients, associations between at least one historical LTBSC and an historical seizure.

179. The computer-implemented method of any one of claims 175 to 178, wherein the second model output data includes at least a portion of LTBSC data corresponding to the seizure.

180. The computer-implemented method of any one of claims 175 to 179, wherein: the training dataset further comprises historical long-term biometric (LTB) data corresponding to the historical LTBSC data of the historical patients; and the second model is further trained to determine, for the historical patients, associations between the historical LTBs, and the corresponding historical LTBSCs and / or historical seizures.181 . The computer-implemented method of claim 180, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

182. The computer-implemented method of any one of claims 175 to 181 , comprising:obtaining, by the one or more processors, LTBSC data associated with one or more patients; and based upon the LTBSC data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

183. The computer-implemented method of any one of claims 175 to 182, further comprising: receiving one or more labels for the first training dataset, wherein training the first iteration of the second model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the primary second model output data; and receiving one or more labels for the second training dataset, wherein training the second iteration of the second model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the secondary second model output data.

184. A computer-implemented method of any one of claims 175 to 183, wherein: the first training dataset does not include long-term based brain state change(LTBSC) data, and the second training dataset includes the LTBSC data, the first and second training datasets being used to fine-tune the second model.

185. A computer-implemented method for training a fourth model to generate fourth model output data indicating a prediction of a seizure of the patient, the method comprising: obtaining, by one or more processors, a first training dataset; receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; training, by the one or more processors, a first iteration of the fourth model using the first training dataset using the feature values defined for the first training dataset to generate primary fourth model output data having associated first error rates; obtaining, by the one or more processors, a second training dataset; receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; andtraining, by the one or more processors, a second iteration of the fourth model using the second training dataset using the feature values defined for the second training dataset to generate secondary fourth model output data having associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

186. The computer-implemented method of claim 185, wherein the one or more attributes comprise one or more characteristics of an electroencephalogram (EEG) signal.

187. The computer-implemented method of claims 185 or 186, wherein a training dataset comprises historical LTBSC data of historical patients.

188. The computer-implemented method of claim 187, wherein the fourth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and timing of an historical seizure.

189. The computer-implemented method of any one of claims 185 to 188, wherein the fourth model output data includes a date of the prediction of the seizure, and / or a confidence metric associated with the prediction.

190. The computer-implemented method of any one of claims 185 to 189, wherein: the training dataset further comprises historical long-term biometric (LTB) data corresponding to the historical LTBSC data of the historical patients; and the fourth model is further trained to determine, for the historical patients, associations between the historical LTBs, and the corresponding historical LTBSCs and / or historical predictions of seizures.191 . The computer-implemented method of claim 190, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

192. The computer-implemented method of any one of claims 185 to 191 , comprising: obtaining, by the one or more processors, LTBSC data associated with one or more patients; andbased upon the LTBSC data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

193. The computer-implemented method of any one of claims 185 to 192, further comprising: receiving one or more labels for the first training dataset, wherein training the first iteration of the fourth model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the primary fourth model output data; and receiving one or more labels for the second training dataset, wherein training the second iteration of the fourth model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the secondary fourth model output data.

194. A computer-implemented method of any one of claims 185 to 193, wherein: the first training dataset does not include long-term based brain state change(LTBSC) data, and the second training dataset includes the LTBSC data, the first and second training datasets being used to fine-tune the fourth model.

195. A computer-implemented method for training a third model to generate third model output data, the third model output data including a revision of an indication of a seizure in the long-term EEG data of the LTBSC data, the method comprising: obtaining, by one or more processors, a first training dataset; receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; training, by the one or more processors, a first iteration of the third model using the first training dataset using the feature values defined for the first training dataset to generate primary third model output data having associated first error rates; obtaining, by the one or more processors, a second training dataset; receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and training, by the one or more processors, a second iteration of the third model using the second training dataset using the feature values defined for the second training dataset to generate secondary third model output data having associated second error rates,wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

196. The computer-implemented method of claim 195, wherein the one or more attributes comprise one or more characteristics of an electroencephalogram (EEG) signal.

197. The computer-implemented method of claims 195 or 196, wherein a training dataset comprises historical LTBSC data of historical patients.

198. The computer-implemented method of claim 197, wherein the third model is trained to determine, for the historical patients, associations between an historical seizure and at least one historical LTBSC.

199. The computer-implemented method of any one of claims 195 to 198, wherein the revision includes a reclassification of the seizure and / or a change in a confidence metric associated with the indication of the seizure.

200. The computer-implemented method of any one of claims 195 to 199, wherein: the training dataset further comprises historical long-term biometric (LTB) data corresponding to the historical LTBSC data of the historical patients; and the third model is further trained to determine, for the historical patients, associations between the historical LTBs, and the corresponding historical LTBSCs and / or historical indications of seizures.201 . The computer-implemented method of claim 200, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

202. The computer-implemented method of any one of claims 195 to 201 , comprising: obtaining, by the one or more processors, the LTBSC data indicating seizures of patients; and based upon the LTBSC data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

203. The computer-implemented method of any one of claims 195 to 202, further comprising: receiving one or more labels for the first training dataset, wherein training the first iteration of the third model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the primary third model output data; and receiving one or more labels for the second training dataset, wherein training the second iteration of the third model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the secondary third model output data.

204. A computer-implemented method for training a fifth model to generate fifth model output data indicating at least one date to implement a medication titration of the patient, the method comprising: obtaining, by one or more processors, a first training dataset; receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; training, by the one or more processors, a first iteration of the fifth model using the first training dataset using the feature values defined for the first training dataset to generate primary fifth model output data having associated first error rates; obtaining, by the one or more processors, a second training dataset; receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and training, by the one or more processors, a second iteration of the fifth model using the second training dataset using the feature values defined for the second training dataset to generate secondary fifth model output data having associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

205. The computer-implemented method of claim 204, wherein the one or more attributes comprise one or more characteristics of an electroencephalogram (EEG) signal.

206. The computer-implemented method of claims 204 or 205, wherein a training dataset comprises historical LTBSC data of historical patients.

207. The computer-implemented method of claim 206, wherein the fifth model is trained to determine, for the historical patients, associations between at least one historical LTBSC, timing of an historical medication titration, and associated historical side effects from the historical medication titration.

208. The computer-implemented method of any one of claims 204 to 207, wherein: the training dataset further comprises historical long-term biometric (LTB) data corresponding to the historical LTBSC data of the historical patients; and the fifth model is further trained to determine, for the historical patients, associations between the historical LTBs, and one or more of the corresponding historical LTBSCs, historical times in which medication titrations were implemented, and associated historical side effects.

209. The computer-implemented method of claim 208, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

210. The computer-implemented method of any one of claims 204 to 209, comprising: obtaining, by the one or more processors, the LTBSC data; and based upon the LTBSC data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.21 1 . The computer-implemented method of any one of claims 204 to 210, further comprising: receiving one or more labels for the first training dataset, wherein training the first iteration of the fifth model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the primary fifth model output data; and receiving one or more labels for the second training dataset, wherein training the second iteration of the fifth model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the secondary fifth model output data.

212. A computer-implemented method for training a sixth model to generate sixth model output data indicating an association between at least one environmentalcharacteristic of the patient and the at least one LTBSC of the patient, the method comprising: obtaining, by one or more processors, a first training dataset; receiving as feature values, by the one or more processors, a selection of one or more attributes of the first training dataset; training, by the one or more processors, a first iteration of the sixth model using the first training dataset using the feature values defined for the first training dataset to generate primary sixth model output data having associated first error rates; obtaining, by the one or more processors, a second training dataset; receiving as feature values, by the one or more processors, the selection of one or more attributes of the second training dataset; and training, by the one or more processors, a second iteration of the sixth model using the second training dataset using the feature values defined for the second training dataset to generate secondary sixth model output data having associated second error rates, wherein the second error rates have a reduced overall error rate compared to an overall error rate of the first error rates.

213. The computer-implemented method of claim 212, wherein the one or more attributes comprise one or more characteristics of an electroencephalogram (EEG) signal.

214. The computer-implemented method of claims 212 or 213, wherein a training dataset comprises historical LTBSC data of historical patients and historical environmental data of the historical patients.

215. The computer-implemented method of claim 214, wherein the sixth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and at least one historical environmental characteristic.

216. The computer-implemented method of any one of claims 212 to 215, wherein: the training dataset further comprises historical long-term biometric (LTB) data corresponding to the historical LTBSC data of the historical patients; and the sixth model is further trained to determine, for the historical patients, associations between the historical LTBs, and the corresponding historical LTBSCs and / or historical environmental characteristics.

217. The computer-implemented method of claim 216, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

218. The computer-implemented method of any one of claims 212 to 217, comprising: obtaining, by the one or more processors, the LTBSC data; and based upon the LTBSC data, generating, by the one or more processors, one or more of the first training dataset or the second training dataset.

219. The computer-implemented method of any one of claims 212 to 218, further comprising: receiving one or more labels for the first training dataset, wherein training the first iteration of the sixth model using the first training dataset further comprises using the one or more labels for the first training dataset to generate the primary sixth model output data; and receiving one or more labels for the second training dataset, wherein training the second iteration of the sixth model using the second training dataset further comprises using the one or more labels for the second training dataset to generate the secondary sixth model output data.

220. A non-transitory computer-readable medium storing processor-executable instructions that, when executed by one or more processors, cause the one or more processors to at least: obtain long-term electroencephalogram (LTEEG) data comprising at least one LTEEG signal associated with a brain of the patient and gathered via an implant device of the patient; provide the LTEEG data to a first model trained using first model training data to generate long-term brain state change (LTBSC) data indicating at least one LTBSC of the patient; in response to generating the LTBSC data, one or more of:(i) retrain one or more machine learning models using the LTBSC data;(i) provide the LTBSC data as an input to the one or more machine learning models; or(iii) generate a medical insight of the patient; andprovide the medical insight to a user device.221 . The non-transitory computer-readable medium of claim 220, wherein the LTEEG data includes at least one substantially continuous LTEEG signal of the patient gathered over more than a week of time.

222. The non-transitory computer-readable medium of claims 220 or 221 , wherein: the first model training data comprises historical LTEEG data of historical patients and historical LTBSC data of historical patients corresponding to the historical LTEEG data.

232. The non-transitory computer-readable medium of claim 222, wherein: the first model is trained to determine, for the historical patients, associations between an historical LTBSC and at least one historical LTEEG signal corresponding to the historical LTBSC.

224. The non-transitory computer-readable medium of any one of claims 220 to223, wherein the LTBSC data includes at least a portion of the LTEEG data corresponding to the at least one LTBSC.

225. The non-transitory computer-readable medium of any one of claims 220 to224, wherein the LTBSC data includes a classification and / or confidence metric associated with the at least one LTBSC.

226. The non-transitory computer-readable medium of any one of claims 220 to225, further comprising instructions that, when executed, cause the one or more processors to: obtain long-term biometric (LTB) data indicating long-term biometrics of the patient corresponding to the at least one LTBSC; and provide the LTB data to the first model, wherein generating the LTBSC data is further based upon the LTB data.

227. The non-transitory computer-readable medium of claim 226, wherein the long-term biometrics indicate one or more of a voice biomarker, eye attentiveness, or a facial change.

228. The non-transitory computer-readable medium of claims 226 or 227, wherein: the first model training data further comprises historical LTB data of historical patients corresponding to the historical LTBSC data of the historical patients; and the first model is further trained to determine, for the historical patients, associations between the historical LTBSC and historical long-term biometrics corresponding to the historical LTBSC.

229. The non-transitory computer-readable medium of any one of claims 226 to228, wherein the LTBSC data includes at least a portion of the LTB data corresponding to the LTBSC data.

230. The non-transitory computer-readable medium of any one of claims 220 to229, wherein generating the LTBSC data includes one or more of: analyze at least one frequency band of the at least one LTEEG signal; identify at least one LTEEG segment in the at least one LTEEG signal based upon one or more of a root means square, a standard deviation, or a power spectrum density, of the at least one LTEEG signal; determine one or more feature values of the at least one LTEEG segment by applying a principal components analysis algorithm or a support vector machine to the LTEEG data; determine at least one similar characteristic of the LTEEG data and / or the one or more feature values, across long-duration time points; classify LTEEG segments having similar characteristics; or determine an LTBSC boundary based upon detecting an edge of a phase transition of the at least one LTEEG signal.231 . The non-transitory computer-readable medium of claim 230, wherein the one or more feature values are associated with one or more characteristics of an electroencephalogram (EEG) signal.

232. The non-transitory computer-readable medium of claims 230 or 231 , wherein the at least one frequency band is selected from the group consisting of 1 to 2 hertz (Hz),0.5 to 4 Hz, 4 to 8 Hz, 8 to 13 Hz, 13 to 30 Hz, 30 to 100 Hz, 80 to 250 Hz, and 250 to 500Hz.

233. The non-transitory computer-readable medium of any one of claims 220 to232, wherein the at least one LTBSC is associated with one or more of epilepsy, a psychiatric disorder, a stroke, Alzheimer's disease, a sleep disorder, depression, medication efficacy, or a medication side effect.

234. The non-transitory computer-readable medium of any one of claims 220 to233, wherein retraining the one or more one machine learning models using the LTBSC data includes fine-tuning the one or more machine learning models.

235. The non-transitory computer-readable medium of any one of claims 220 to234, wherein the medical insight is one or more of: an indication of the at least one LTBSC; a risk of sudden unexplained death by epilepsy; or an association between the at least one LTBSC and the long-term biometrics.

236. The non-transitory computer-readable medium of any one of claims 220 to235, further comprising instructions that, when executed, cause the one or more processors to: obtain additional first model training data; retrain the first model using the additional first model training data; and store the retrained first model on a memory to generate subsequent LTBSC data using the retrained first model.

237. The non-transitory computer-readable medium of any one of claims 220 to236, wherein: the one or more machine learning models include a second model trained using second model training data to generate second model output data indicating a seizure of the patient in the LTEEG of the LTBSC data, based upon receiving the LTBSC data; to retrain the second model using the LTBSC data further comprises instructions that, when executed, cause the one or more processors to store the retrained second modelin the memory to generate subsequent second model output data using the retrained second model; and to provide the LTBSC data as the input to the second model further comprises instructions that, when executed, cause the one or more processors to, in response to generating the second model output data, provide the second model output data to the user device.

238. The non-transitory computer-readable medium of claim 237, wherein the second model training data comprises historical long-term brain state change (LTBSC) data of historical patients; and239. The non-transitory computer-readable medium of claim 238, wherein the second model is trained to determine, for the historical patients, associations between at least one historical LTBSC and an historical seizure.

240. The non-transitory computer-readable medium of any one of claims 237 to239, wherein the second model output data includes a classification of the seizure and / or a confidence metric associated with the indication of the seizure.241 . The non-transitory computer-readable medium of any one of claims 237 to240, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

242. The non-transitory computer-readable medium of any one of claims 237 to241 , further comprising instructions that, when executed, cause the one or more processors to: obtain additional second model training data; retrain the second model using the additional second model training data; and store the retrained second model in the memory to generate subsequent second model output data using the retrained second model.

243. The non-transitory computer-readable medium of any one of claims 220 to 236, wherein: the one or more machine learning models is third model trained using third model training data to generate third model output data, third model output data including arevision of an indication of a seizure in the LTBSC data, based upon receiving the LTBSC data, wherein: the third model output data includes a revision of an indication of a seizure in the LTEEG data of the LTBSC data; the indication of the seizure is not based upon the at least one LTBSC of the LTBSC data; and the revision to the indication of the seizure is based upon the at least one LTBSC of the LTBSC data; to retrain the third model using the LTBSC data further comprises instructions that, when executed, cause the one or more processors to store the retrained third model in the memory to generate subsequent third model output data using the retrained third model; and to provide the LTBSC data as the input to the third model further comprises instructions that, when executed, cause the one or more processors to, in response to generating the third model output data, provide the third model output data to the user device.

244. The non-transitory computer-readable medium of claim 243, wherein the third model training data comprises historical LTBSC data of historical patients.

245. The non-transitory computer-readable medium of claims 243 or 244, wherein the third model is trained to determine, for the historical patients, associations between an historical seizure and at least one historical LTBSC.

246. The non-transitory computer-readable medium of any one of claims 243 to245, wherein the revision includes a reclassification of the seizure and / or a change in a confidence metric associated with the indication of the seizure.

247. The non-transitory computer-readable medium of any one of claims 243 to246, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

248. The non-transitory computer-readable medium of any one of claims 243 to247, further comprising instructions that, when executed, cause the one or more processors to:obtain additional third model training data; retrain the third model using the additional third model training data; and store the retrained third model in the memory to generate subsequent third model output data using the retrained third model.

249. The non-transitory computer-readable medium of any one of claims 220 to 236, wherein: the one or more machine learning models include a fourth model trained using fourth model training data to generate fourth model output data indicating a prediction of a seizure of the patient based upon receiving the LTBSC data; to retrain the fourth model using the LTBSC data further comprises instructions that, when executed, cause the one or more processors to store the retrained fourth model in the memory to generate subsequent fourth model output data using the retrained fourth model; and to provide the LTBSC data as the input to the fourth model further comprises instructions that, when executed, cause the one or more processors to, in response to generating the fourth model output data, provide the fourth model output data to the user device.

250. The non-transitory computer-readable medium of claim 249, wherein the fourth model training data comprises historical LTBSC data of historical patients.251 . The non-transitory computer-readable medium of claim 250, wherein the fourth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and timing of an historical seizure.

252. The non-transitory computer-readable medium of any one of claims 249 to251 , wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

253. The non-transitory computer-readable medium of any one of claims 249 to252, wherein the fourth model output data includes at least one date of the prediction of the seizure, and / or a confidence metric associated with the prediction.

254. The non-transitory computer-readable medium of any one of claims 249 to 253, further comprising instructions that, when executed, cause the one or more processors to: obtain additional fourth model training data; retrain the fourth model using the additional fourth model training data; and store the retrained fourth model in the memory to generate subsequent fourth model output data using the retrained fourth model.

255. The non-transitory computer-readable medium of any one of claims 220 to 236, wherein: the one or more machine learning models include a fifth model trained using fifth model training data to generate fifth model output data indicating at least one date to implement a medication titration of the patient based upon receiving the LTBSC data; to retrain the fifth model using the LTBSC data further comprises instructions that, when executed, cause the one or more processors to store the retrained fifth model in the memory to generate subsequent fifth model output data using the retrained fifth model; and to provide the LTBSC data as the input to the fifth model further comprises instructions that, when executed, cause the one or more processors to, in response to generating the fifth model output data, provide the fifth model output data to the user device.

256. The non-transitory computer-readable medium of claim 255, wherein the fifth model training data comprises historical LTBSC data of historical patients.

257. The non-transitory computer-readable medium of claims 255 or 256, wherein the fifth model is trained to determine, for the historical patients, associations between at least one historical LTBSC, timing of an historical medication titration, and associated historical side effects from the historical medication titration.

258. The non-transitory computer-readable medium of any one of claims 255 to257, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

259. The non-transitory computer-readable medium of any one of claims 255 to258, further comprising instructions that, when executed, cause the one or more processors to:obtain additional fifth model training data; retrain the fifth model using the additional fifth model training data; and store the retrained fifth model in the memory to generate subsequent fifth model output data using the retrained fifth model.

260. The non-transitory computer-readable medium of any one of claims 220 to 236, wherein: the one or more machine learning models include a sixth model trained using sixth model training data to generate sixth model output data indicating an association between at least one environmental characteristic of the patient and the at least one LTBSC of the patient, based upon receiving environmental data indicating the at least one environmental characteristic of the patient and the LTBSC data; to retraining the sixth model using the LTBSC data further comprises instructions that, when executed, cause the one or more processors to store the retrained sixth model in the memory to generate subsequent sixth model output data using the retrained sixth model; and to provide the LTBSC data as the input to the sixth model further comprises instructions that, when executed, cause the one or more processors to: obtain the environmental data associated with the LTBSC data; provide the environmental data and the LTBSC data to the sixth model to generate the sixth model output data; and in response to generating the sixth model output data, provide the sixth model output data to the user device.261 . The non-transitory computer-readable medium of claim 260, wherein the sixth model training data comprises historical LTBSC data of historical patients and historical environmental data of the historical patients.

262. The non-transitory computer-readable medium of claim 261 , wherein the sixth model is trained to determine, for the historical patients, associations between at least one historical LTBSC and at least one historical environmental characteristic.

263. The non-transitory computer-readable medium of any one of claims 260 to262, wherein the at least one environmental characteristic is associated with a medication.

264. The non-transitory computer-readable medium of claim 263, wherein the association with the medication includes a new medication taken by the patient, and / or a new titration of the medication taken by the patient.

265. The non-transitory computer-readable medium of any one of claims 260 to264, wherein: the environmental data is obtained in response to a LTBSC of the patient; and the sixth model output data indicates a cause of the LTBSC.

266. The non-transitory computer-readable medium of any one of claims 260 to265, wherein the sixth model output data indicates a prediction of a future LTBSC in response to the at least one environmental characteristic, and / or a treatment pathway to a target brain state.

267. The non-transitory computer-readable medium of any one of claims 260 to266, wherein the environmental data is generated by an implant device of the patient.

268. The non-transitory computer-readable medium of any one of claims 260 to267, wherein the LTBSC data includes LTB data corresponding to the at least one LTBSC.

269. The non-transitory computer-readable medium of any one of claims 260 to268, further comprising instructions that, when executed, cause the one or more processors to: obtain additional sixth model training data; retrain the sixth model using the additional sixth model training data; and store the retrained sixth model in the memory to generate subsequent sixth model output data using the retrained sixth model.