Systems and methods for seizure prediction and detection

Abstract

To provide systems and methods for suitable seizure prediction and detection.SOLUTION: A method for seizure detection may include receiving a plurality of electroencephalography (EEG) signals over a plurality of channels for a subject, preprocessing the plurality of EEG signals by segmenting the plurality of EEG signals for each channel into a plurality of temporal data segments, extracting a plurality of features from each temporal data segment for each channel, and applying a machine learning algorithm to the plurality of features to perform a seizure binary classification for each temporal data segment for each channel. A control policy may be employed to determine a seizure burden on the aggregated seizure binary classifications. When the seizure burden is equal to or exceeds a threshold, a notification may be generated. The notification may be usable by a healthcare practitioner to assess whether the subject may be at risk of having a seizure.SELECTED DRAWING: Figure 1

Description

[Background technology]

[0001] This PCT application claims priority to U.S. Patent Application No. 16 / 578,032, filed September 20, 2019, the contents of which are incorporated herein by reference in their entirety.

[0002] Monitoring electroencephalography (EEG) signals is a key task for early diagnosis in seizures. Analyzing EEG signals plays an important role in monitoring the patient's brain activity, but experts are required to analyze all EEG recordings and detect seizure activity. This can be laborious and time-consuming, and timely and accurate diagnosis of seizure activity is essential to initiate therapy and reduce the risk of future seizures and seizure-related complications.

[0003] Currently, machine learning algorithms provide a means to classify EEG signals and minimize expert intervention. However, machine learning algorithms require good quality EEG signals to provide effective classification results. In many cases, the EEG signals provided to the machine learning algorithms need to be optimized to make them more effective at predicting seizures. Summary of the Invention [Means for solving the problem]

[0004] There is a need to optimize the quality of EEG signals provided to machine learning algorithms to make them more effective in preventing seizures. To optimize the quality of the EEG signals, the EEG signals can be used to build features. Each derived feature can focus on a distinct characteristic of the EEG signal that allows the machine learning algorithm to more easily discriminate between the measured EEG signals. Through a combination of using the features and then classifying them using a machine learning algorithm, a more effective means of predicting seizures is enabled. Furthermore, after classification, implementing a control policy and a seizure load calculation allows for an even better means of accurately determining seizures by removing false positive or inaccurate readings from the provided EEG signals.

[0005] Aspects of the present disclosure provide a method for seizure detection. The method may include receiving a plurality of electroencephalography (EEG) signals for a subject via a plurality of channels. The method may further include preprocessing the plurality of EEG signals by segmenting the plurality of EEG signals for each channel into a plurality of temporal data segments. In some cases, the method may extract a plurality of features from each temporal data segment for each channel. In some cases, the method may apply a machine learning algorithm to the plurality of features to perform a seizure binary classification for each temporal data segment for each channel.

[0006] In some embodiments, pre-processing the EEG signals may further include applying a filter to the EEG signals via the channels prior to segmenting the EEG signals. In some cases, the filter may include a band pass filter configured to filter the EEG signals between 1 Hz and 35 Hz.

[0007] In some embodiments, the seizure binary classification may include classifying each temporal data segment per channel as (1) seizure positive or (2) seizure negative. In some cases, each temporal data segment may be associated with a time epoch. In some cases, a cluster of seizure positive classifications may indicate potential electrographic seizures across the corresponding time epoch. In some cases, the method may further include sequentially comparing the classifications across the multiple time epochs on each channel and discarding a subset of the classifications if the subset includes fewer than three consecutive seizure positive classifications. In some cases, the method may further include classifying a particular time epoch as associated with a potential electrographic seizure if the temporal data segment for a subset of the multiple channels is classified as seizure positive. In some cases, the subset may include at least half of the multiple channels. In some cases, the subset may include at least half of the multiple channels.

[0008] In some embodiments, the method may further include aggregating the seizure binary classification for the plurality of temporal data segments for the plurality of channels over a moving time window. In some cases, the moving time window may range from about 1 minute to 10 minutes. In some cases, the moving time window may be about 5 minutes. In some cases, the plurality of channels may include at least three channels. In some cases, the plurality of channels may include eight channels. In some embodiments, each temporal data segment may have a duration ranging from about 1 second to 20 seconds. In some cases, the duration of each temporal data segment may be about 10 seconds.

[0009] In some embodiments, the plurality of features may include time and / or frequency domain features that are unique within the plurality of EEG signals. In some cases, the plurality of features may include at least 20 different time and / or frequency features. In some cases, the plurality of features may include a plurality of discrete values ​​associated with the time and / or frequency domain features.

[0010] In some embodiments, the machine learning algorithm may include a random forest, a boosted decision tree, a classification tree, a regression tree, a bagged tree, a neural network, or a rotation forest. In some cases, the machine learning algorithm may be applied individually to the features extracted for each channel, such that each channel has a separate iteration of the machine learning algorithm.

[0011] In some embodiments, the method may further include determining a seizure burden over a moving time window based on the aggregated seizure binary classification. In some cases, the seizure burden may include a percentage of time epochs that are classified as seizure positive. In some cases, determining the seizure burden may include averaging the seizure positive classifications over a moving time window.

[0012] In some embodiments, the method may further include generating one or more notifications when the seizure load equals or exceeds one or more thresholds. In some cases, the one or more notifications may be usable by a health care practitioner to assess whether the subject is at risk of having a seizure. In some cases, the one or more notifications may be generated in the form of a visual, audio, and / or text alert. In some cases, a first notification indicating frequent seizure activity may be generated when the seizure load equals or exceeds a first threshold of 10%. In some cases, a second notification indicating a large amount of seizure activity may be generated when the seizure load equals or exceeds a second threshold of 50%. In some cases, a third notification indicating persistent seizure activity may be generated when the seizure load equals or exceeds a third threshold of 90%.

[0013] In another aspect, the present disclosure provides a seizure detection system. The seizure detection system may include a pre-processing module configured to receive a plurality of electroencephalography (EEG) signals for a subject via a plurality of channels. The pre-processing module may also be configured to pre-process the plurality of EEG signals by segmenting the plurality of EEG signals for each channel into a plurality of temporal data segments. The seizure detection system may also include a processing module in communication with the pre-processing module. The processing module may be configured to receive the plurality of temporal data segments corresponding to the plurality of channels. The processing module may also be configured to extract a plurality of features from each temporal data segment for each channel. The processing module may also be configured to apply a machine learning algorithm to the plurality of features to perform a seizure binary classification for each temporal data segment for each channel.

[0014] In some embodiments, the seizure detection system may further comprise an output module in communication with the pre-processing module. The output module may be configured to aggregate the seizure binary classification for the multiple temporal data segments for the multiple channels over a moving time window. The output module may also be configured to determine a seizure load over the moving time window based on the aggregated seizure binary classification. The output module may also be configured to generate one or more notifications when the seizure load equals or exceeds one or more thresholds.

[0015] In another aspect, the present disclosure provides a system comprising one or more computer processors. The system may also comprise a memory comprising machine executable instructions that, upon execution by the one or more computer processors, implement a method for seizure detection. The method may include receiving a plurality of electroencephalography (EEG) signals for a subject via a plurality of channels. The method may also include pre-processing the plurality of EEG signals by segmenting the plurality of EEG signals for each channel into a plurality of temporal data segments. The method may also further include extracting a plurality of features from each temporal data segment for each channel. The method may also further include applying a machine learning algorithm to the plurality of features to perform a seizure binary classification for each temporal data segment for each channel.

[0016] In another aspect, the present disclosure provides a non-transitory computer-readable medium comprising machine-executable instructions that, upon execution by one or more computer processors, implement a method for seizure detection. The method may include receiving a plurality of electroencephalography (EEG) signals for a subject via a plurality of channels. The method may include pre-processing the plurality of EEG signals by segmenting the plurality of EEG signals for each channel into a plurality of temporal data segments. The method may include extracting a plurality of features from each temporal data segment for each channel. The method may also include applying a machine learning algorithm to the plurality of features to perform a seizure binary classification for each temporal data segment for each channel.

[0017] In another aspect, the present disclosure provides a method for seizure detection. The method may include receiving a plurality of electroencephalography (EEG) signals for a subject via a plurality of channels. The method may also include preprocessing the plurality of EEG signals by segmenting the plurality of EEG signals for each channel into a plurality of temporal data segments. The method may also include extracting a plurality of features from each temporal data segment for each channel. The method may also include applying a machine learning algorithm to the plurality of features to perform a seizure binary classification for each temporal data segment for each channel. The method may also further include determining a seizure burden over a moving time window based on the aggregated seizure binary classification. The method may also further include presenting the seizure burden to a user.

[0018] In another aspect, the present disclosure provides a method for seizure detection. The method may include receiving a plurality of electroencephalography (EEG) signals for a subject via a plurality of channels. The method may further include preprocessing the plurality of EEG signals by segmenting the plurality of EEG signals for each channel into a plurality of temporal data segments. The method may also include extracting a plurality of features from each temporal data segment for each channel, where each temporal data segment is associated with a time epoch. The method may further include applying a machine learning algorithm to the plurality of features to perform a seizure binary classification for each temporal data segment for each channel, thereby generating a plurality of classifications for the plurality of temporal data segments, where the seizure binary classification for each temporal data segment includes classifying each temporal data segment for each channel as (1) seizure positive or (2) seizure negative, where the plurality of classifications are compared sequentially across the plurality of time epochs for each channel, and a subset of the classifications is discarded if the subset includes fewer than three consecutive seizure positive classifications. The method may further include aggregating remaining non-discarded classifications for the plurality of temporal data segments for the plurality of channels over the moving time window. The method may also include determining a seizure burden over the moving time window based on the aggregated classifications, the seizure burden including a percentage of the temporal data segments classified as seizure positive, the seizure burden being a metric that provides a measure of a degree of severity or likelihood of a seizure. The method may also include generating one or more notifications when the seizure burden equals or exceeds one or more thresholds, the one or more notifications being indicative of different seizure activity and usable to assess whether the subject is at risk of having a seizure.

[0019] In some embodiments, a cluster of seizure positive classifications indicates a potential electrographic seizure over the corresponding time epoch. In some cases, each temporal data segment has a duration ranging from about 1 second to 20 seconds. In some cases, the duration of each temporal data segment is about 10 seconds. In some cases, the moving time window ranges from about 1 minute to 10 minutes.

[0020] In some embodiments, one or more notifications are generated in the form of visual, audio, and / or text alerts. In some cases, a first notification, indicative of frequent seizure activity, is generated when the seizure load equals or exceeds a first threshold of 10%. In some cases, a second notification, indicative of a large amount of seizure activity, is generated when the seizure load equals or exceeds a second threshold of 50%. In some cases, a third notification, indicative of persistent seizure activity, is generated when the seizure load equals or exceeds a third threshold of 90%.

[0021] In some embodiments, the plurality of channels includes at least three channels. In some cases, the plurality of channels includes at least eight channels. In some cases, the plurality of features includes time and / or frequency domain features unique within the plurality of EEG signals. In some cases, the plurality of features includes at least 20 distinct time and / or frequency features.

[0022] In some embodiments, the plurality of features comprises a plurality of discrete values ​​associated with the time and / or frequency domain features.

[0023] In some embodiments, the machine learning algorithm comprises a random forest, a boosted decision tree, a classification tree, a regression tree, a bagged tree, a neural network, or a rotation forest. In some cases, the machine learning algorithm is applied individually to the features extracted for each channel, such that each channel has a separate iteration of the machine learning algorithm.

[0024] In some embodiments, pre-processing the EEG signals further includes applying a filter to the EEG signals via the channels prior to segmenting the EEG signals, in some cases the filter includes a band pass filter configured to filter the EEG signals between 1 Hz and 35 Hz.

[0025] In some embodiments, the method may further include classifying a particular time epoch as associated with a potential electrographic seizure if the temporal data segment for a subset of the plurality of channels is classified as seizure positive. In some cases, the subset includes at least half of the plurality of channels. In some cases, the sequential time periods formed by the time windows are non-overlapping.

[0026] In some embodiments, the method may further include outputting the seizure burden as a graphical visual on a display. In some cases, the method may include displaying the one or more thresholds in the graphical visual. In some cases, the graphical visual includes a time series plot, a bar graph, or a chart. In some cases, the time series plot is configured to change color as the seizure burden crosses the one or more thresholds.

[0027] In some embodiments, the methods may also include using the graphical visual element to (i) assess a condition in a subject, (ii) determine a course of treatment, (iii) monitor the effectiveness of a course of treatment if a course of treatment is provided to the subject, or (iv) monitor the progression of a subject's condition over time. The present specification also provides, for example, the following items: (Item 1) 1. A method for seizure detection, comprising: receiving a plurality of electroencephalography (EEG) signals relating to a subject via a plurality of channels; pre-processing the EEG signals by segmenting the EEG signals per channel into a plurality of temporal data segments; Extracting a plurality of features from each temporal data segment for each channel; applying a machine learning algorithm to the plurality of features to perform a binary seizure classification for each temporal data segment for each channel; A method comprising: (Item 2) 2. The method of claim 1, wherein the seizure binary classification includes classifying each temporal data segment for each channel as (1) seizure positive or (2) seizure negative. (Item 3) 3. The method of claim 2, wherein each temporal data segment is associated with a time epoch. (Item 4) 4. The method of claim 3, wherein a cluster of seizure positive classifications indicates a potential electrographic seizure over the corresponding time epoch. (Item 5) 2. The method of claim 1, wherein each temporal data segment has a duration ranging from about 1 second to 20 seconds. (Item 6) 6. The method of claim 5, wherein each temporal data segment has a duration of about 10 seconds. (Item 7) 4. The method of claim 3, further comprising aggregating the seizure binary classification for the plurality of temporal data segments for the plurality of channels over a moving time window. (Item 8) 8. The method of claim 7, wherein the moving time window ranges from about 1 minute to 10 minutes. (Item 9) 9. The method of claim 8, wherein the moving time window is about 5 minutes. (Item 10) 8. The method of claim 7, further comprising determining seizure burden over the moving time window based on the aggregated seizure binary classification. (Item 11) 11. The method of claim 10, wherein the seizure burden comprises the percentage of the time epochs that are classified as seizure positive. (Item 12) 11. The method of claim 10, wherein determining the seizure burden comprises averaging the seizure positivity classifications over the moving time window. (Item 13) 11. The method of claim 10, further comprising generating one or more notifications when the seizure burden equals or exceeds one or more thresholds. (Item 14) 14. The method of claim 13, wherein the one or more notifications are usable by a health care practitioner to assess whether the subject is at risk for having a seizure. (Item 15) Item 14. The method of item 13, wherein the one or more notifications are generated in the form of a visual, audio, and / or text alert. (Item 16) 14. The method of claim 13, wherein a first notification indicative of frequent seizure activity is generated when the seizure burden equals or exceeds a first threshold of 10%. (Item 17) 14. The method of claim 13, wherein a second notification indicative of significant seizure activity is generated when the seizure load equals or exceeds a second threshold of 50%. (Item 18) 14. The method of claim 13, wherein a third notification indicative of sustained seizure activity is generated when the seizure burden equals or exceeds a third threshold of 90%. (Item 19) Item 2. The method of item 1, wherein the plurality of channels includes at least three channels. (Item 20) 20. The method of claim 19, wherein the plurality of channels comprises eight channels. (Item 21) 2. The method of claim 1, wherein the plurality of features comprises time and / or frequency domain features inherent in the plurality of EEG signals. (Item 22) 22. The method of claim 21, wherein the plurality of features comprises at least 20 different time and / or frequency features. (Item 23) 22. The method of claim 21, wherein the plurality of features comprises a plurality of discrete values ​​associated with the time and / or frequency domain features. (Item 24) 2. The method of claim 1, wherein the machine learning algorithm comprises a random forest, a boosted decision tree, a classification tree, a regression tree, a bagged tree, a neural network, or a rotation forest. (Item 25) 2. The method of claim 1, wherein the machine learning algorithm is applied individually to the plurality of features extracted for each channel, such that each channel has a separate iteration of the machine learning algorithm. (Item 26) Pre-processing the plurality of EEG signals further comprises: 2. The method of claim 1, further comprising applying a filter to the EEG signals via the channels prior to segmenting the EEG signals. (Item 27) 27. The method of claim 26, wherein the filter comprises a bandpass filter configured to filter the plurality of EEG signals from 1 Hz to 35 Hz. (Item 28) sequentially comparing said classifications across multiple time epochs on each channel; discarding the subset of classes if the subset contains fewer than three consecutive seizure positive classes; The method of claim 3, further comprising: (Item 29) 4. The method of claim 3, further comprising classifying a particular time epoch as associated with a potential electrographic seizure if the temporal data segment for the subset of the plurality of channels is classified as seizure positive. (Item 30) 30. The method of claim 29, wherein the subset includes at least half of the plurality of channels. (Item 31) 1. A method for seizure detection, comprising: (a) receiving a plurality of electroencephalography (EEG) signals relating to a subject via a plurality of channels; (b) pre-processing the EEG signals by segmenting the EEG signals for each channel into a plurality of temporal data segments; (c) extracting a plurality of features from each temporal data segment for each channel, each temporal data segment being associated with a time epoch; (d) applying a machine learning algorithm to the plurality of features to perform a binary seizure classification for each temporal data segment per channel, thereby generating a plurality of classifications for the plurality of temporal data segments, wherein the binary seizure classification for each temporal data segment includes classifying each temporal data segment per channel as (1) seizure positive or (2) seizure negative, the plurality of classifications being compared sequentially across a plurality of time epochs for each channel, and a subset of the classifications being discarded if the subset includes fewer than three consecutive seizure positive classifications; (e) aggregating the remaining non-discarded classifications for the plurality of temporal data segments for the plurality of channels over a moving time window; (f) determining a seizure burden over the moving time window based on the aggregated classification, the seizure burden comprising a percentage of the temporal data segments that are classified as seizure positive, the seizure burden being a metric that provides a measure of a degree of severity or likelihood of a seizure; and (g) generating one or more notifications when the seizure burden equals or exceeds one or more thresholds, the one or more notifications being indicative of distinct seizure activity and usable to assess whether the subject is at risk for having a seizure; A method comprising: (Item 32) 32. The method of claim 31, wherein a cluster of seizure positive classifications indicates a potential electrographic seizure over the corresponding time epoch. (Item 33) 32. The method of claim 31, wherein each temporal data segment has a duration ranging from about 1 second to 20 seconds. (Item 34) Item 34. The method of item 33, wherein each temporal data segment has a duration of about 10 seconds. (Item 35) 32. The method of claim 31, wherein the moving time window ranges from about 1 minute to 10 minutes. (Item 36) 32. The method of claim 31, wherein the one or more notifications are generated in the form of a visual, audio, and / or text alert. (Item 37) 32. The method of claim 31, wherein a first notification indicative of frequent seizure activity is generated when the seizure burden equals or exceeds a first threshold of 10%. (Item 38) 32. The method of claim 31, wherein a second notification indicative of significant seizure activity is generated when the seizure load equals or exceeds a second threshold of 50%. (Item 39) 32. The method of claim 31, wherein a third notification indicative of sustained seizure activity is generated when the seizure burden equals or exceeds a third threshold of 90%. (Item 40) 32. The method of claim 31, wherein the plurality of channels includes at least three channels. (Item 41) Item 41. The method of item 40, wherein the plurality of channels includes eight channels. (Item 42) 32. The method of claim 31, wherein the plurality of features comprises time and / or frequency domain features inherent in the plurality of EEG signals. (Item 43) Item 43. The method of item 42, wherein the plurality of features comprises at least 20 different time and / or frequency features. (Item 44) 43. The method of claim 42, wherein the plurality of features comprises a plurality of discrete values ​​associated with the time and / or frequency domain features. (Item 45) 32. The method of claim 31, wherein the machine learning algorithm comprises a random forest, a boosted decision tree, a classification tree, a regression tree, a bagged tree, a neural network, or a rotation forest. (Item 46) 32. The method of claim 31, wherein the machine learning algorithm is applied individually to the features extracted for each channel such that each channel has a separate iteration of the machine learning algorithm. (Item 47) Pre-processing the plurality of EEG signals further comprises: 32. The method of claim 31, comprising applying a filter to the plurality of EEG signals via the plurality of channels prior to segmenting the plurality of EEG signals. (Item 48) Item 48. The method of item 47, wherein the filter includes a bandpass filter configured to filter the plurality of EEG signals from 1 Hz to 35 Hz. (Item 49) 32. The method of claim 31, further comprising classifying a particular time epoch as associated with a potential electrographic seizure if the temporal data segment for the subset of the plurality of channels is classified as seizure positive. (Item 50) 50. The method of claim 49, wherein the subset includes at least half of the plurality of channels. (Item 51) 32. The method of claim 31, wherein the sequential periods formed by the time windows are non-overlapping. (Item 52) 32. The method of claim 31, further comprising outputting the seizure burden as a graphical visual element on a display. (Item 53) 53. The method of claim 52, further comprising displaying the one or more threshold values ​​within the graphical visualization. (Item 54) Item 53. The method of item 52, wherein the graphical visual element comprises a time series plot, a bar graph, or a chart. (Item 55) 55. The method of claim 54, wherein the time series plot is configured to change color as the seizure burden crosses the one or more thresholds. (Item 56) 53. The method of claim 52, further comprising using the graphical visual element to (i) assess a condition of the subject, (ii) determine a course of treatment, (iii) monitor the effectiveness of a course of treatment if the course of treatment is provided to the subject, or (iv) monitor the progression of the subject's condition over time.

[0028] Additional aspects and advantages of the present disclosure will become readily apparent to those skilled in the art from the following detailed description, in which only illustrative embodiments of the present disclosure are shown and described. As will be realized, the present disclosure is capable of other and different embodiments, and its several details are capable of modification in various obvious respects, all without departing from the present disclosure. Thus, the drawings and description are to be regarded as illustrative in nature, and not restrictive. (Incorporated by reference)

[0029] All publications, patents, and patent applications mentioned herein are incorporated by reference to the same extent as if each individual publication, patent, or patent application was specifically and individually indicated to be incorporated by reference. To the extent that the publications and patents or patent applications incorporated by reference conflict with the present disclosure contained within this specification, the present specification is intended to supersede and / or take precedence over any such conflicting material. [Brief description of the drawings]

[0030] The novel features of the invention are set forth with particularity in the appended claims. A better understanding of the features and advantages of the present invention will be obtained by reference to the following detailed description that sets forth illustrative embodiments, in which the principles of the invention are utilized, and the accompanying drawings (also referred to herein as "figure" and "FIG") in which:

[0031] [Figure 1] FIG. 1 illustrates an EEG device configured to provide an EEG signal to a seizure detection module according to an embodiment of the present disclosure.

[0032] [Diagram 2] FIG. 2 shows an illustration of an EEG signal workflow for seizure detection according to an embodiment of the present disclosure.

[0033] [Diagram 3] FIG. 3 shows a seizure burden graph according to an embodiment of the present disclosure.

[0034] [Figure 4] FIG. 4 shows an illustration of an EEG device with display visualization of seizure detection output according to an embodiment of the present disclosure.

[0035] [Diagram 5] FIG. 5 shows an illustration of the EEG device software with display visualization of seizure detection output according to an embodiment of the present disclosure.

[0036] [Figure 6] FIG. 6 illustrates a computer system that is programmed or otherwise configured to implement the methods provided herein, according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0037] Detailed Description While various embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Numerous variations, changes, and substitutions may occur to those skilled in the art without departing from the invention. It is understood that various alternatives to the embodiments of the present invention described herein may be employed.

[0038] Whenever the term "up to about" or "at least about" precedes a first number in a series of two or more numbers, the term "up to about" or "at least about" applies to each and every number in the series. For example, up to about 3, 2, or 1 is equivalent to up to about 3, up to about 2, or up to about 1. overview

[0039] Manual review of electroencephalography (EEG) brain signals can be a time-consuming and labor-intensive process. Valuable time can be lost in manually analyzing such EEG signals when a patient may be suffering from a seizure. Automated techniques that can analyze EEG signals can help provide a timely and accurate diagnosis of seizure activity and assist clinicians in initiating therapy and reducing the risk of future seizures and seizure-related complications.

[0040] The machine learning algorithm provides a means to automatically classify EEG signals and minimize expert intervention. However, the machine learning algorithm requires good quality EEG signals to provide effective classification results. In addition, the EEG signals provided to the machine learning algorithm may need to be provided as features that describe the characteristics of the EEG signals with respect to seizure activity. Furthermore, after classification of the features by the machine learning algorithm, a control strategy comprising a set of rules, together with a seizure burden calculation, allows the method and / or system to more accurately delineate whether a subject is suffering or potentially suffering from a seizure.

[0041] The seizure detection system and method described herein provides a user with an EEG detection device coupled to a seizure detection module capable of automatic and accurate seizure detection. In addition, the seizure detection module can notify the user of an impending / active seizure. The seizure detection module captures and pre-processes EEG signals from the EEG device. The EEG signals are then analyzed and valuable EEG features are extracted. The EEG features are classified using a machine learning algorithm module. The classification of features over a given time epoch is then governed by a control policy to calculate a seizure burden value. If the seizure burden value is equal to or exceeds one or more thresholds, the method and / or system generates one or more notifications that can be used by a health care practitioner to assess whether the subject may be at risk of having a seizure. An increase in the value of seizure burden can be used by a health care practitioner as an indication of increasing seizure severity. Seizure burden values ​​equal to or exceeding one or more thresholds may be used by health care practitioners as an indication of certain medical conditions, such as status epilepticus. Changes in seizure burden over time, i.e., the characteristic shape of the seizure burden graph, may be used by health care practitioners to determine a course of treatment for a subject or to assess the effectiveness of a course of treatment for a subject. Seizure Detection Module I. Signal Acquisition and Preprocessing II. Signal analysis III. Seizure Burden Calculation and Output IV. After seizure detection V. Computer Systems I. Signal Acquisition and Preprocessing (a) EEG signal / acquisition

[0042] For ease of explanation, the figures and corresponding description below are described below with reference to analysis of signals representative of brain activity (e.g., electroencephalography (EEG) signals) and / or cardiac activity (e.g., electrocardiogram (ECG) signals) of a biological subject. However, those skilled in the art will recognize that signals representative of other bodily functions (e.g., electromyography (EMG) signals, or electronystagmography (ENG) signals, pulse oximetry signals, capnography signals, and / or photoplethysmography signals) may be substituted or used in addition to (e.g., in conjunction with) one or more signals representative of brain activity and / or cardiac activity.

[0043] A system for measuring bioelectrical signals may generally include one or more electrodes electrically coupled to a controller and / or output device via corresponding conductive wires. In other variations, the electrodes may be wirelessly coupled to the controller and / or output device. The electrodes may be contained within an electrode carrier system that is secured around the patient's head. The electrode carrier system may be configured as a head band or incorporated into any number of other platforms or positioning mechanisms for maintaining the electrodes relative to the patient's body. The individual electrode assemblies may be spaced apart from one another such that when the head band is positioned on the patient's head, the electrode assemblies may be optimally aligned to receive EEG signals.

[0044] The controller and / or output device may generally comprise any number of devices for receiving electrical signals, such as electrophysiological monitoring devices, and may be used in combination with any number of brain imaging devices, e.g., fMRI, PET, NIRS, etc. In one particular variation, the electrode embodiments described herein may be used in combination with devices such as those configured to receive electrical signals from the electrodes and process them.

[0045] In some embodiments, the signal corresponding to brain electrical activity corresponds to an electrical signal obtained from a human brain and obtained from a single neuron or from multiple neurons. In some embodiments, the sensor includes one or more sensors (e.g., extracranial sensors) that are externally attached (e.g., taped, attached, glued) to the human scalp. For example, the extracranial sensor may include an electrode (e.g., electroencephalography (EEG) electrode) or multiple electrodes (e.g., electroencephalography (EEG) electrodes) that are externally attached to the scalp (e.g., glued to the skin via a conductive gel) or, more commonly, positioned at a discrete location outside the scalp. Alternatively, dry electrodes can also be used in some implementations (e.g., conductive sensors that are mechanically placed against the body of a living subject rather than embedded within the body of the living subject or held in place with a conductive gel). An example of a dry electrode is a headband with one or more metal sensors (e.g., electrodes) that are worn by the living subject during use. Signals obtained from extracranial sensors may sometimes be referred to as EEG signals or time-domain EEG signals. In some cases, the sensors may be accelerometers or inertial measurement units (IMUs) that may measure mechanical movement of the subject and / or device (e.g., produce one or more electrical signals that correspond to mechanical movement of the subject and / or device). The system may be configured to utilize one or more sensors to assist in seizure detection as described elsewhere herein.

[0046] In one aspect, the present disclosure provides a method for seizure detection. In some cases, the method may include receiving a plurality of signals (e.g., EEG signals, EKG signals, EMG signals, etc.) related to a subject via a plurality of channels. The method may include receiving a plurality of electroencephalography (EEG) signals related to a subject via a plurality of channels. The plurality of EEG signals may be provided to a seizure detection module. FIG. 1 shows an illustration of an EEG signal collection workflow by an EEG device module 110 for a seizure detection module 115. FIG. 2 shows a detailed illustration of a workflow by an EEG device module and a seizure detection module for seizure prediction. As shown in FIG. 2, the seizure detection module may include a pre-processing module, a signal analysis module, and a seizure load calculation and output module. The EEG device module 110 may have a plurality of channels 205 for obtaining EEG signals from a subject. In some cases, the plurality of channels may include 1 to 256 channels. In some cases, the plurality of channels may include 8 to 256 channels. In some cases, the plurality of channels may include more than 256 channels. In some cases, the multiple channels may include 8, 10, 16, 20, 32, 64, 128, or 256 channels. (b) Preprocessing of EEG signals

[0047] In some embodiments, the EEG device module may have one or more analog front ends configured to receive sensor EEG signals from the sensors. The EEG signals may be pre-processed as described elsewhere herein. In some embodiments, separate (e.g., independent) analog front ends may be provided to interface with each of the sets of sensors. In some embodiments, one or more analog front ends may be provided to interface with the set of EEG sensors.

[0048] In some embodiments, the method may include preprocessing the multiple signals by segmenting the multiple signals per channel into multiple temporal data segments. In some embodiments, the method may include preprocessing the multiple EEG signals by segmenting the multiple EEG signals per channel into multiple temporal data segments. FIG. 2 shows an illustration of the seizure detection module 120. The seizure detection module takes the EEG signals from the multiple channels from the EEG device module. The seizure detection module may preprocess the EEG signals from the multiple channels using a preprocessing module 210 configured to preprocess the EEG signals. As shown in FIG. 2, the preprocessing module may include a signal filtering module 215, a signal segmentation module 220, and a signal conditioning module 225.

[0049] In some embodiments, the filtering module 215 may be configured to filter the EEG signals from the incoming set of channels from the EEG device module, as described elsewhere herein. In some cases, the pre-processing may be, for example, segmenting the EEG signals, filtering the EEG signals based on frequency, conditioning the EEG signals, or as described elsewhere herein.

[0050] 2, the signal segmentation module 220 can be configured to segment the EEG signals. In some embodiments, the multiple EEG signals may be segmented into 1 to 100,000 data segments. In some cases, the number of EEG data segments may depend on the duration of the EEG recording. In some cases, the number of EEG data segments may be fixed regardless of the duration of the EEG recording.

[0051] In some embodiments, each temporal data segment may have a duration of about 1 second to 1 hour. In some cases, each temporal data segment may have a duration of about 1 second to 30 seconds. In some cases, each temporal data segment may have a duration of about 1 second to 10 seconds. In some cases, the duration of each temporal data segment may be fixed throughout the EEG recording. In some cases, the duration of each temporal data segment may be variable or adaptive during the EEG recording.

[0052] In some embodiments, pre-processing the plurality of EEG signals may include applying one or more filtering steps to the plurality of EEG signals via the plurality of channels. Pre-processing the plurality of EEG signals may include using at least one filter, two filters, three filters, four filters, five filters, six filters, seven filters, eight filters, nine filters, ten filters, fifteen filters, or more filters. Pre-processing the plurality of EEG signals may include using up to about fifteen filters, ten filters, nine filters, eight filters, seven filters, six filters, five filters, four filters, three filters, two filters, or less filters. Pre-processing the plurality of EEG signals may include using any number of 1-15 filters, 1-10 filters, 1-5 filters, or 1-3 filters.

[0053] In some embodiments, one or more filtering steps may be applied before, during, and / or after segmentation of the EEG signals. One or more of the filtering steps may include, for example, a digital filter, an analog filter, or a combination thereof. One or more of the filtering steps may include, for example, a band pass filter, a low pass filter, a high pass filter, a band reject filter, an all pass filter, a Kalman filter, an adaptive filter, or a notch filter, etc. In some cases, the low frequency cutoff of the filter may be 0.1 Hz to 5 Hz. In some cases, the high frequency cutoff of the filter may be 5 Hz to 200 Hz. In some cases, the notch filter frequency may match the local power line frequency. In some cases, the notch filter frequency may be 50 Hz or 60 Hz to match the local power line frequency.

[0054] In some embodiments, each temporal data segment may be associated with a time epoch. For each corresponding time epoch, a cluster of seizure positive classifications may indicate potential electrographic seizures. In some cases, a cluster of seizure positive classifications may include approximately 1-50 seizure positive classifications. In some cases, a cluster of seizure positive classifications may include 1-10 seizure positive classifications.

[0055] In some embodiments, the method may further include sequentially comparing the classifications across multiple time epochs on each channel. In some cases, before / after / during the sequential comparison of the classifications across multiple time epochs on each channel, the sequential classifications across multiple time epochs on each channel may be discarded. In some cases, a subset of the classifications may be discarded. In some cases, a subset of less than about 1-20 classifications may be discarded. In some cases, a subset of less than 3 classifications may be discarded.

[0056] In some embodiments, a subset of seizure positive classifications may be discarded, for example because they may be random readings, low reliability, inaccurate classifications, incorrect classifications, calibration, system errors, disconnected electrodes, artifact signals, system interference, or other signals, etc.

[0057] In some embodiments, a subset of the seizure positive classifications may be discarded, for example, to conserve memory space, improve processing speed, reduce energy usage, reduce system heat, reduce computational costs, conserve processing power, save processing time, increase reliability, or reduce random access memory usage.

[0058] In some embodiments, a higher number of consecutive seizure positive classifications may indicate a higher confidence level. The higher the confidence level of the seizure positive classification, the more accurate the decision to detect a seizure in a patient. In some cases, a higher confidence level of the seizure positive classification may indicate machine learning algorithm accuracy, the quality of the data (EEG signal), or the health status of the EEG detection system, etc. In some embodiments, a particular time epoch may be classified as associated with a potential electrographic seizure if a temporal data segment for a subset of the plurality of channels is classified as seizure positive. In some cases, the subset may be at least 5%, 10%, 20%, 30%, 40%, 50%, or more of the plurality of channels. In some cases, the subset may be up to about 50%, 40%, 30%, 20%, 10%, 5%, or less of the plurality of channels. (c) EEG signal conditioning

[0059] 2 illustrates a signal conditioning module 225 configured to condition an EEG signal. In some embodiments, the method may condition any EEG signal. Conditioning an EEG signal may include, for example, increasing and / or decreasing the amplitude of the EEG signal, adding or decreasing the noise level of the EEG signal, increasing and / or decreasing the time epochs of the EEG signal, increasing and / or decreasing the intensity of the EEG signal, increasing and / or decreasing the signal frequency of the EEG signal, increasing and / or decreasing the voltage of the EEG signal, changing the morphological structure of the EEG signal (e.g., the shape of the EEG signal), increasing and / or decreasing the periodicity of the EEG signal, increasing or decreasing the synchrony of the EEG waves, spectral subtraction, normalization, etc.

[0060] In some cases, the EEG signal may be reduced. In some cases, the EEG signal may be downsampled to a lower sampling frequency. For example, EEG data recorded at a sampling frequency of 500 Hz may be downsampled by a factor of two to 250 Hz.

[0061] In some cases, the EEG signal may undergo bit-width reduction. In some cases, the level of resolution at which the EEG signal is recorded may not be required by the method to achieve accurate seizure detection. In some cases, bit-width reduction may reduce the EEG signal to a lower number of bits per sample, for example from 32 bits per sample to 12 bits per sample, through standard quantization of the EEG signal. In some cases, bit-width reduction may be advantageous because it may be useful to reduce power consumption due to reduced processing load when the method is implemented in a portable system.

[0062] In some cases, spectral subtraction may be used to reduce the amount of additive noise in the EEG signal. In some cases, the noise may be caused by the external surroundings. In some cases, the noise may be caused by the measurement equipment. In some cases, the noise may be caused by the user. In some cases, the average frequency spectrum of the non-seizure EEG signal may be calculated to provide a base level estimate of the noise frequency spectrum over a period of time. In some cases, the EEG signal may be transformed into the frequency domain as the EEG signal is recorded. In some cases, the average noise spectrum may then be subtracted from the EEG frequency spectrum. In some cases, the resulting spectrum and phase information from the original noise signal may be combined. In some cases, the resulting spectrum may be transformed back to the time domain to produce a denoised signal.

[0063] In some embodiments, the EEG signals may be standardized by eliminating the effect of the montage used in collecting the EEG signals. In some cases, independent component analysis (ICA) or principal component analysis (PCA) methods may be used to provide montage rejection. In some cases, ICA or PCA methods may separate the EEG signals into a set of sources independent of the montage used to record them. In some cases, using standardized EEG data may remove errors introduced by the varying practices of clinicians.

[0064] In some cases, a non-negative matrix factorization (NMF) method may be applied to each channel as a form of artifact removal. In some cases, the spectrum of the signal may be decomposed into extracted bases to obtain weights. In some cases, the spectrum may be reconstructed using the bases and corresponding weights of the artifacts removed from the initial EEG signal. II. Signal analysis (a) Feature Extraction

[0065] In FIG. 2, the seizure detection module may include a signal analysis module. The signal analysis module may include a feature extraction module 245 and a machine learning classification module 250. The feature extraction module 245 may be configured to obtain pre-processed measured data (e.g., EEG signals) from the pre-processing module 210 and construct derived values ​​(e.g., features). In some embodiments, the feature extraction may start from an initial set of measured data (e.g., EEG signals, EEG signals at a given time epoch, etc.) and construct derived values ​​(e.g., features) that are intended to be informative and non-redundant. In some cases, the feature extraction module may include extracting multiple features from each temporal data segment for each channel individually. In some cases, the feature extraction module may include extracting multiple features from each temporal data segment for all channels together. In some cases, the feature extraction module may include extracting multiple features from each temporal data segment of one or more groups, each grouping consisting of one or more channels. 2, the extracted features can be relayed to a machine learning classification module 250, which can be configured to analyze and classify the extracted features as described elsewhere herein. In some cases, feature extraction can facilitate subsequent learning and generalization steps of a machine learning algorithm. In some cases, feature extraction can lead to better human interpretation. In some cases, feature extraction can be associated with dimensionality reduction.

[0066] In some cases, when the input data to a machine learning algorithm (e.g., EEG signals) is suspected to be too large to process and redundant (e.g., repetition of the same measurements or characteristics in both Hz and seconds), the data can be transformed into a reduced set of features.

[0067] In some cases, determining the subset of initial features may be referred to as feature selection. In some cases, the selected features may be expected to contain relevant information from the input data (e.g., EEG signals). In some cases, the selected features may be expected to contain relevant information from the input data such that a desired task can be performed by using this reduced representation instead of the complete initial data.

[0068] In some embodiments, feature extraction may involve reducing the number of resources required to explain a large set of data (e.g., EEG signals). In some cases, analysis involving a large number of variables may require large amounts of memory and computational power. In some cases, this may cause machine learning algorithms to overfit to training samples and generalize poorly to new samples. In some cases, feature extraction may construct combinations of variables to accurately explain the data with sufficient accuracy. In some cases, feature extraction may construct combinations of variables to accurately explain the data with sufficient accuracy while preventing overfitting.

[0069] In some embodiments, the results may be improved using a constructed set of application-dependent features. In some cases, the constructed set may be built by experts. In some cases, general dimensionality reduction techniques may be used. In some cases, the general dimensionality reduction techniques may be, for example, independent component analysis, isomap, kernel PCA, latent semantic analysis, partial least squares, principal component analysis, multi-factor dimensionality reduction, nonlinear dimensionality reduction, multilinear principal component analysis, multilinear subspace learning, semidefinite embedding, autoencoder, etc.

[0070] In some cases, a set of numerical features may be described by a feature vector, which in some cases may be an n-dimensional vector of numerical features that represents an object.

[0071] In some embodiments, a data analysis software package may provide feature extraction. In some cases, the data analysis software package may provide dimensionality reduction. In some cases, the data analysis software package may include a programming environment such as MATLAB, SciLab, NumPy, or R language. In some cases, a programming language script may be used to extract features from the EEG signal. In some cases, the programming language script may be, for example, MATLAB, python, java, Javascript, Ruby, C, C++, or Perl.

[0072] In some cases, the features may be unique within the EEG signals. Unique may be features of the EEG signals that may be measured, such as the amplitude of the EEG signal, the duration of the EEG signal, the variance of the EEG signal, the power of the EEG signal, local maxima / minima of the EEG signal, the pattern of the EEG signal, the regularity of the EEG signal, the spectral power distribution of the EEG signal, or the frequency of the EEG signal. In some cases, the features may be measurements of the power of the signal with a particular frequency. The frequency may be, for example, about 0 Hz to 100 Hz. In some cases, the power of the signal may be normalized to the total power. In some cases, the power of the signal may be a ratio of the power between one or more signals. In some cases, the feature may be a function performed on the signal to obtain a value. For example, the function may measure the root mean square (RMS) of the signal (e.g., the EEG signal) to obtain the RMS value of the signal. In some cases, the feature may compare one signal (e.g., the EEG signal) to one or more signals. In some cases, the features may compare one or more signals (e.g., EEG signals) to one or more signals. In some cases, the features may measure attributes of the signals (e.g., EEG signals). In some cases, the features may compare one or more attributes of the signals (e.g., EEG signals) to one or more attributes of the signals. The attributes may be, for example, inherent properties of the EEG signals. In some cases, the features of the EEG signals may be continuous and / or discrete time.

[0073] In some cases, the plurality of features may include at least 20 different time and / or frequency features. In some cases, the plurality of features may include up to 1,000 time and / or frequency features. In some cases, the plurality of features may include between about 10 features and 200 features. In some cases, the plurality of features may include between about 10 features and 100 features. In some cases, the plurality of features may include between about 10 features and 50 features.

[0074] In some cases, the plurality of features may include multiple discrete values ​​associated with time domain, frequency domain, time-frequency domain, information theory, and nonlinear dynamical systems theory features. In some cases, the plurality of features may include multiple discrete values ​​associated with time and / or frequency domain features. The plurality of features may include multiple continuous values ​​associated with time and / or frequency domain features.

[0075] In some cases, the signals may be converted to digital signals. In some cases, the signals may be converted to digital signals and then to analog signals.

[0076] In some cases, the features may be sampled from a portion of the EEG signal to reduce the processing time and power required.

[0077] In some embodiments, features may be associated with a certain weighting value. The weighting value may give a higher score to one feature for detecting seizures. A higher score may indicate that the feature may be more relevant in predicting seizure activity. The method may adjust the weighting value of any feature at any given time. The method may adjust the weighting value of any feature at any given time by increasing and / or decreasing. (b) Classification using machine learning

[0078] In some embodiments, the method may include applying a machine learning algorithm to the features to perform seizure classification for each temporal data segment for each channel individually. In some cases, the machine learning classification module may include performing seizure classification for each temporal data segment for all channels together. In some cases, the machine learning classification module may include performing seizure classification for each temporal data segment of one or more groups, each group consisting of one or more channels. Figure 2 shows a machine learning classification module 250 that may determine the collected / extracted features from the pre-processing step and classify the features. In some cases, the features may be extracted without a pre-processing step.

[0079] In some cases, machine learning algorithms may need to extract and draw relationships between features because traditional statistical techniques may not be sufficient. In some cases, machine learning algorithms may be used in conjunction with traditional statistical techniques. In some cases, traditional statistical techniques may provide pre-processed features to the machine learning algorithm.

[0080] In some embodiments, the plurality of features may be classified into any number of categories. A temporal segment may be classified, for example, as seizure positive, seizure negative, seizure-like, uncertain seizure activity, etc. In some cases, the plurality of features may be classified into 1-20 categories. Individual categories may also be divided into subcategories. For example, a temporal segment classified as seizure positive may be further subdivided into focal vs. generalized seizure events.

[0081] In some embodiments, the method may include applying a machine learning algorithm to the plurality of features to perform a binary seizure classification for each temporal data segment for each channel.

[0082] In some embodiments, one or more collected features may be discarded prior to or during the machine learning classification.

[0083] In some embodiments, a human may select and discard features prior to / during machine learning classification. In some cases, a computer may select and discard features. In some cases, features may be discarded based on a threshold.

[0084] In some embodiments, any number of features may be classified by the machine learning algorithm. The machine learning algorithm may classify at least 10 features. In some cases, the plurality of features may include between about 10 features and 200 features. In some cases, the plurality of features may include between about 10 features and 100 features. In some cases, the plurality of features may include between about 10 features and 50 features. In some embodiments, the machine learning algorithm may be, for example, an unsupervised learning algorithm, a supervised learning algorithm, or a combination thereof. The unsupervised learning algorithm may be, for example, clustering, hierarchical clustering, k-means, mixture models, DBSCAN, OPTICS algorithm, anomaly detection, local outlier factor methods, neural networks, autoencoders, deep belief networks, Hebbian learning, generative adversarial networks, self-organizing maps, expectation maximization algorithms (EM), method of moments, blind source separation techniques, principal component analysis, independent component analysis, non-negative matrix factorization, singular value decomposition, or a combination thereof. The supervised learning algorithm may be, for example, a support vector machine, a linear regression, a logistic regression, a linear discriminant analysis, a decision tree, a k-nearest neighbor algorithm, a neural network, similarity learning, or a combination thereof. In some embodiments, the machine learning algorithm may include a deep neural network (DNN). The deep neural network may include a convolutional neural network (CNN). The CNN may be, for example, U-Net, ImageNet, LeNet-5, AlexNet, ZFNet, GoogleNet, VGGNet, ResNet18, or ResNet, etc.Other neural networks may be, for example, deep feedforward neural networks, recurrent neural networks, LSTMs (long short-term memories), GRUs (gated recurrent units), autoencoders, variational autoencoders, adversarial autoencoders, denoising autoencoders, sparse autoencoders, Boltzmann machines, RBMs (restricted BMs), deep belief networks, generative adversarial networks (GANs), deep residual networks, capsule networks, or attention / transformer networks, etc.

[0085] In some embodiments, the machine learning algorithm may be, for example, a random forest, a boosted decision tree, a classification tree, a regression tree, a bagged tree, a neural network, or a rotation forest. The machine learning algorithm may be applied to multiple features extracted for each channel individually, such that each channel may have a separate iteration of the machine learning algorithm.

[0086] In some embodiments, the method may apply one or more machine learning algorithms. In some embodiments, the method may apply one or more machine learning algorithms per channel.

[0087] In FIG. 2, the machine learning classification module 250 may include any number of machine learning algorithms. In some embodiments, the random forest machine learning algorithm may be an ensemble of bagged decision trees. In some cases, the ensemble of bagged decision trees may classify each temporal data segment per channel as (1) seizure positive or (2) seizure negative. The ensemble may be at least about 1, 2, 3, 4, 5, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 120, 140, 160, 180, 200, 250, 500, 1,000, or more bagged decision trees. The ensemble may be up to about 1,000, 500, 250, 200, 180, 160, 140, 120, 100, 90, 80, 70, 60, 50, 40, 30, 20, 10, 5, 4, 3, 2, or less bagged decision trees. The ensemble may be about 1-1,000, 1-500, 1-200, 1-100, or 1-10 bagged decision trees.

[0088] In some embodiments, the method may include applying the machine learning classifier to any number of channels. The method may include applying the machine learning classifier to at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 50, 100, 500, 1,000, or more channels. The method may include applying the machine learning classifier to up to about 1,000, 500, 100, 50, 25, 20, 15, 10, 9, 8, 7, 6, 5, 4, 3, 2, or less channels. The method may include applying the machine learning classifier to about 1-1,000, 1-100, 1-25, or 1-5 channels.

[0089] In some embodiments, the method may include applying a machine learning classifier to a subset of channels. The subset of channels may be at least about 1%, 5%, 10%, 20%, 30%, 40%, 50%, or more of the total set of channels. The subset of channels may be up to about 50%, 40%, 30%, 20%, 10%, 5%, 1%, or less of the total set of channels. The subset of channels may be about 1%-50%, 1%-40%, 1%-30%, 1%-20%, 1%-10%, or 1%-5% of the total set of channels.

[0090] In some embodiments, the machine learning algorithm may have various parameters, such as a learning rate, a small batch size, a number of epochs to train, momentum, a learning weight decay rate, or neural network layers.

[0091] In some embodiments, the learning rate may be between about 0.00001 and 0.1.

[0092] In some embodiments, the small batch size may be about 16-128.

[0093] In some embodiments, the neural network may comprise neural network layers. The neural network may have at least about 2 to 1,000 or more neural network layers.

[0094] In some embodiments, the number of epochs for training may be at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, 90, 95, 100, 150, 200, 250, 500, 1,000, 10,000, or more.

[0095] In some embodiments, the momentum may be at least about 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, or more, In some embodiments, the momentum may be up to about 0.9, 0.8, 0.7, 0.6, 0.5, 0.4, 0.3, 0.2, 0.1, or less.

[0096] In some embodiments, the learning weighted decay rate may be at least about 0.00001, 0.0001, 0.001, 0.002, 0.003, 0.004, 0.005, 0.006, 0.007, 0.008, 0.009, 0.01, 0.02, 0.03, 0.04, 0.05, 0.06, 0.07, 0.08, 0.09, 0.1, or more. In some embodiments, the learning weighted attenuation may be up to about 0.1, 0.09, 0.08, 0.07, 0.06, 0.05, 0.04, 0.03, 0.02, 0.01, 0.009, 0.008, 0.007, 0.006, 0.005, 0.004, 0.003, 0.002, 0.001, 0.0001, 0.00001, or less.

[0097] In some embodiments, the machine learning algorithm may use a loss function, which may be, for example, a regression loss, a mean absolute error, a mean bias error, a hinge loss, an Adam optimizer, and / or a cross-entropy.

[0098] In some embodiments, the parameters of the machine learning algorithm may be adjusted with the aid of a human and / or a computer system.

[0099] In some embodiments, the machine learning algorithm may prioritize certain features. The machine learning algorithm may prioritize features that may be more relevant for detecting seizures. A feature may be more relevant for detecting seizures if it is classified more frequently than another feature. In some cases, features may be prioritized using a weighting system. In some cases, features may be prioritized on a probability statistic based on the frequency and / or number of occurrences of the feature. The machine learning algorithm may prioritize features with the aid of a human and / or a computer system.

[0100] In some embodiments, one or more of the features may be used in conjunction with machine learning or traditional statistical techniques to determine whether a segment is likely to contain an artifact. FIG. 2 shows an artifact rejection module 255, which identifies segments that contain artifacts. The identified artifacts may be the result of electrical interference, electrode instability or movement, subject movement, subject eye movement or blinking, subject chewing, subject muscle tension, subject electrocardiogram artifacts, etc. In some cases, a movement sensor or other sensor may be used as an additional input to the artifact rejection module. In some cases, the identified artifacts may be rejected from use in seizure classification. In some cases, the identified artifacts may be reduced, eliminated, or eliminated, and the remaining signal may still be processed for seizure classification.

[0101] In some cases, a machine learning algorithm may prioritize certain features to reduce computational cost, conserve processing power, conserve processing time, increase reliability, reduce random access memory usage, etc. III. Seizure Burden Calculation and Output (a) Control strategy and seizure burden

[0102] In some embodiments, the seizure binary classification may include classifying each temporal data segment per channel as (1) seizure positive or (2) seizure negative. The seizure binary classification may use a machine learning algorithm as described elsewhere herein. The method may include aggregating the seizure binary classification for the multiple temporal data segments for the multiple channels over a moving time window. The aggregated seizure classification may be according to a control strategy module 275 of the seizure burden calculation and output module 270 as shown in FIG. 2. FIG. 2 illustrates the seizure burden calculation and output module 270. As shown in FIG. 2, the seizure burden calculation and output module 270 may include a control strategy module 275, a seizure burden calculation module 280, a seizure burden plot module 285, and a seizure burden notification module 290. The control policy module 275 may be configured to implement a control policy, the seizure burden calculation module 280 may be configured to calculate a seizure burden value, the seizure burden plotting module 285 may be configured to plot the seizure burden value, and the seizure burden notification module 290 may be configured to provide a notification, each as described elsewhere herein.

[0103] In some cases, the moving window may have a duration of 1 minute to 1 hour. In some cases, the duration of the moving window may be dynamic or adjustable instead of fixed. In some cases, the duration of the moving window may be subject dependent.

[0104] In some embodiments, a cluster of seizure positive classifications on one or more channels may result in an overall determination of a patient's seizure over the corresponding time epoch according to the control strategy module 275.

[0105] The control strategy may be a set of rules that results in an overall determination of seizures for the patient over the corresponding time epoch. The control strategy may take a set of parameters as input and act on the set of parameters according to a set of rules, resulting in an overall determination of seizures for the patient over the corresponding time epoch. The set of rules may be as described anywhere herein. The set of rules may be adjusted at any time to act on more parameters or act on fewer parameters. The set of rules may be adjusted at any time to include more rules or remove rules. The set of rules may be at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 50, 100, 500, 1,000, or more rules. The set of rules may be up to about 1,000, 500, 100, 50, 25, 20, 15, 10, 9, 8, 7, 6, 5, 4, 3, 2, or less rules. The set of rules may be about 1-1,000, 1-500, 1-100, 1-25, 1-10, 1-5, or 1-3 rules.

[0106] In some embodiments, parameter inputs for the control strategy may include the number of classifications of channels as seizure positive, the number of classifications of channels as seizure negative, classification of channels as seizure positive, classification of channels as seizure negative, corresponding time epochs, number of channels, the machine learning algorithm used for classification, the moving window time length, the quality of the connection for each channel, information derived from the EKG signal, information derived from the EMG signal, information about the patient's demographics, information about the patient's current or previous condition, information about a treatment or medication regimen applied to the patient, information derived from a movement sensor (e.g., an accelerometer or an inertial measurement unit), etc.

[0107] In some embodiments, a control strategy may have inputs of any number of parameters. A control strategy may have inputs of at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 20, 25, 50, 100, 500, 1,000, or more parameters. A control strategy may have inputs of up to about 1,000, 500, 100, 50, 25, 20, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, or less parameters. A control strategy may have inputs of about 1-1,000, 1-500, 1-100, 1-50, 1-25, 1-15, 1-10, or 1-5 parameters.

[0108] In some embodiments, the set of rules may dictate that the control policy discards the classification of the channel. For example, if the control policy receives an input of a single seizure positive classification over the corresponding time epoch, the set of rules may discard the seizure positive classification over the corresponding time epoch. In some cases, the control policy may receive at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 50, 100, 500, or more seizure positive classifications, and the set of rules may discard each seizure positive classification over the corresponding time epoch. In some cases, the control policy may receive up to about 500, 100, 50, 10, 9, 8, 7, 6, 5, 4, 3, 2, or less seizure positive classifications, and the set of rules may discard each seizure positive classification over the corresponding time epoch. In some cases, the control strategy may receive approximately 1-500, 1-100, 1-50, 1-10, or 1-5 seizure positive classifications, and the set of rules may discard each seizure positive classification over the corresponding time epoch.

[0109] In some embodiments, the set of rules may dictate that the control policy output a seizure positive classification for a set of channels corresponding to a time epoch. For example, if the control policy receives a set of four or more channels that each register a seizure positive classification over a corresponding time epoch, the set of rules may output a seizure positive classification over the corresponding time epoch. In some cases, the control policy may receive a set of seizure positive classifications for at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 50, 100, 1,000, or more channels, and the set of rules may output a seizure positive classification for the set of seizure positive classifications over the corresponding time epoch. In some cases, the control policy may receive a set of seizure positive classifications for up to about 1,000, 100, 50, 25, 20, 15, 10, 9, 8, 7, 6, 5, 4, 3, 2, or less channels, and the set of rules may output a seizure positive classification for the set of seizure positive classifications over a corresponding time epoch. In some cases, the control policy may receive a set of about 1-1,000, 1-500, 1-100, 1-50, 1-25, 1-10, or 1-5 seizure positive classifications, and the set of rules may output a seizure positive classification for the set of seizure positive classifications over a corresponding time epoch.

[0110] In some embodiments, the method may include calculating the patient's seizure burden as a percentage of seizure positive classifications within a defined time period. As shown in FIG. 2, a seizure burden calculation module 280 may be configured to calculate the patient's seizure burden. In some cases, the time period used for the seizure burden calculation may be between 1 minute and 1 hour. In some cases, the time period used for the seizure burden calculation may be the entirety of the recording session. In some cases, the time period used for the seizure burden calculation may be dynamic or adjustable instead of being fixed.

[0111] In some embodiments, the seizure burden may form a continuous output measure by calculating seizure burden over a moving window of time, resulting in seizure burden values ​​over individual sequential periods. In some cases, the duration of the moving window may be from 1 minute to 1 hour. In some cases, the duration of the moving window may be dynamic or adjustable instead of fixed. In some cases, the sequential periods formed by the moving window may overlap. In some cases, the sequential periods formed by the moving window may be non-overlapping. In some cases, the moving window may move in time increments from 1 second to 1 hour. In some cases, the moving window may pause or skip periods such that the resulting seizure burden values ​​are not continuous or sequential in time.

[0112] FIG. 2 illustrates a seizure burden plot module 285 configured to plot a subject's seizure burden. As shown in FIG. 3, the seizure burden output may be displayed to a user as a time series plot 310, with each point representing seizure burden over a period of time. In some embodiments, the seizure burden output may display one or more thresholds (e.g., 10%, 50%, 90%, etc.) to the user on the time series plot. In some embodiments, the seizure burden output may be displayed to the user as a time series plot, a bar graph, a chart, or the like. In some embodiments, the time series plot may be depicted in a color to highlight the thresholds that have been exceeded. As shown in FIG. 3, when the time series plot of seizure burden exceeds a 50% threshold over a period of time, the time series plot may change from gray to orange. In some cases, when the time series plot exceeds 90%, the time series plot may change from orange to red. The time series plot may be any color, and the exceedance of a threshold may be illustrated in any color. In some cases, the seizure burden plot module may display various information, such as the duration measured, the date, or the time of initial acquisition. In some cases, the seizure burden plot may be usable by a health care practitioner to assess a subject's condition and determine a course of treatment. The seizure burden plot may also be usable by a health care practitioner to monitor the progression of a subject's condition over time or to monitor the effectiveness of a course of treatment.

[0113] FIG. 2 illustrates a seizure load notification module 290 configured to generate a notification. In some embodiments, the method may include generating one or more notifications when a seizure positive classification is made or when the seizure load value equals or exceeds one or more thresholds. As shown in FIG. 4, when the seizure load value equals or exceeds a threshold (e.g., a 90% threshold), the system may display a notification 410 to the subject (e.g., patient) or user (e.g., health care practitioner, doctor, nurse, etc.) that the system is detecting persistent seizure activity. The notification may also include any color. For example, the background of the screen that displays the notification may be red. The text of the notification may be any color, for example, white. The color of the screen background may correlate with the value of the seizure load calculation. For example, if the seizure load is equal to or exceeds a threshold, the selected color of the screen background may indicate that the seizure load is equal to or exceeds a threshold. The color of the text of the notification may correlate with the value of the seizure load calculation. For example, if the seizure load is equal to or exceeds a threshold, a selected color of the notification text may indicate that the seizure load is equal to or exceeds the threshold.

[0114] The system may also display various information to the subject or user in addition to the notification of the detected sustained seizure activity. The system may display a seizure burden plot 415, a percentage of the calculated seizure burden 420, the time period over which sustained seizure activity was detected (e.g., 7:40pm-7:50pm), etc. The one or more notifications may be usable by a health care practitioner to assess the subject's condition and determine a course of treatment. In some cases, the one or more notifications may be generated when the seizure burden value equals or exceeds one or more thresholds, as described elsewhere herein. In some cases, the one or more notifications may be generated in the form of visual, audio, and / or text alerts. The device may include a speaker 425 to provide audio notifications. In some cases, the one or more notifications may be delivered via networked communication technologies, such as the Internet, telephone, fax, pager, short message service, etc. In some cases, the form, content, or delivery mechanism of the one or more notifications generated may depend on the seizure burden value. In some cases, a user may be able to select the form, content, or delivery mechanism of one or more notifications that are generated.

[0115] FIG. 5 shows an illustration showing a seizure detection output 510 provided by the methods and systems described herein. The interface may provide an indication of EEG signal activity for multiple channels from the EEG device module 110. FIG. 5 shows examples of parameters that the user may adjust, such as, for example, time display, scale, high pass frequency, low pass frequency, or notch value. The interface may also provide a seizure burden plot as described elsewhere herein. The interface may also provide seizure burden results over different time periods. The interface may also depict the seizure determinations per time segment. The interface may also provide a mechanism for the user to approve or reject the algorithm-derived activation determination or seizure burden calculation. The interface may also provide a mechanism for the user to input its own determination of a seizure, including a segment or seizure burden. In some cases, the seizure burden may be adjusted as a result of user-entered information regarding the seizure episode or seizure burden. The displayed seizure burden and seizure burden notification may be based only on algorithm-derived activation determinations, only on user-entered seizure determinations, or on a combination of algorithm and user seizure determinations.

[0116] In some embodiments, the seizure load calculation module may calculate a seizure load value. The seizure load notification module may output a notification if the seizure load value exceeds a threshold. For example, a notification may be generated if the seizure load value exceeds a 10% threshold. In another example, a notification may be generated if the seizure load value exceeds a 50% threshold. In another example, a notification may be generated if the seizure load value exceeds a 90% threshold. In some cases, a first notification, indicative of frequent seizure activity, may be generated when the seizure load equals or exceeds a first threshold of 10%. In some cases, a second notification, indicative of a large amount of seizure activity, may be generated when the seizure load equals or exceeds a second threshold of 50%. In some cases, a third notification, indicative of persistent seizure activity, may be generated when the seizure load equals or exceeds a third threshold of 90%. In some cases, a notification may be generated to a specific person that the method is programmed to notify.

[0117] In some embodiments, the threshold for notification may be any percentage. The threshold for notification may be at least about 1%, 5%, 10%, 15%, 20%, 25%, 30%, 35%, 40%, 45%, 50%, 55%, 60%, 65%, 70%, 80%, 90%, 95%, 99%, or more. The threshold for notification may be up to about 99%, 95%, 90%, 85%, 80%, 75%, 70%, 65%, 60%, 55%, 50%, 45%, 40%, 35%, 30%, 25%, 20%, 15%, 10%, 5%, 1%, or less. The threshold for notification may be approximately 0% to 100%, 1% to 99%, 5% to 95%, 10% to 90%, 20% to 80%, 30% to 70%, or 40% to 60%. The threshold for notification may also be user adjustable.

[0118] In some embodiments, the seizure load notification module may have any number of thresholds. The seizure load notification module may have a threshold of at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 50, 100, 500, 1,000, 5,000, or more. The seizure load notification module may have a maximum of about 5,000, 1,000, 500, 100, 50, 25, 20, 15, 10, 9, 8, 7, 6, 5, 4, 3, 2, or less. The seizure load notification module may have a threshold of about 1-5,000, 1-1,000, 1-500, 1-100, 1-50, 1-25, 1-10, or 1-5. The amount of thresholds may be increased or decreased at any time.

[0119] In some embodiments, the seizure load notification module may provide any number of notifications. The seizure load notification module may provide at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 30, 35, 40, 50, 100, 500, 1,000, or more notifications. The seizure load notification module may provide up to about 1,000, 500, 100, 50, 40, 35, 30, 25, 20, 15, 10, 9, 8, 7, 6, 5, 4, 3, 2, or less notifications. The seizure load notification module may provide about 1-1,000, 1-500, 1-100, 1-50, 1-25, 1-10, or 1-5 notifications.

[0120] In some embodiments, the seizure load calculation and output module may apply criteria for outputting a notification in addition to the seizure load threshold. In some cases, the duration that the seizure load exceeds a threshold may be used to determine whether a notification is output. In some cases, the amount of time since a previous exceedance of the seizure load threshold may be used to determine whether a notification is output. In some cases, dynamic criteria may be applied along with a combination of time-based, seizure load-based, and other policies to determine whether a notification is output.

[0121] In some embodiments, the system may be coupled with other systems, which in some cases may be eye trackers, movement sensors (e.g., accelerometers or inertial measurement units), electromyograms (EMG), electrocardiograms (ECG or EKG), etc. IV. After seizure detection

[0122] In some embodiments, the method may include generating one or more notifications as described anywhere herein.

[0123] In some embodiments, the method may provide a response to the user to minimize or prevent a detected seizure. The method may provide a response to minimize or reduce the risk of developing a seizure. In some cases, a therapeutic drug may be delivered to the subject to prevent and / or mitigate a predicted seizure. In some cases, a neuromodulation signal pattern may be applied to the subject to prevent and / or mitigate a predicted seizure. In some cases, the method may adjust the amount of neuromodulation or therapeutic drug delivered to the subject. V. Computer Systems

[0124] The present disclosure provides computer systems that are programmed to implement the methods of the present disclosure, including controlling the seizure detection system, controlling hardware components, receiving and processing data, interfacing with a user, etc. The seizure detection system and its various components may include computer hardware (and associated firmware) that may be electrically connected to a computer control system. A control unit may include such a computer system.

[0125] 6 illustrates a computer system 601 that is programmed or otherwise configured to operate and / or control the EEG device module and the seizure detection module. The computer system 601 can coordinate various aspects of the seizure detection system of the present disclosure, such as, for example, processing the EEG signal, segmenting the EEG signal, extracting features, processing the features using machine learning algorithms, implementing control policies and seizure burden, calculating seizure burden values, plotting seizure burden, providing notifications, etc. The computer system 601 can be a user's electronic device or a computer system located remotely to the electronic device. The electronic device can be a mobile electronic device.

[0126] The computer system 601 includes a central processing unit (CPU, also referred to herein as "processor" and "computer processor") 605, which can be a single-core or multi-core processor or multiple processors for parallel processing. The computer system 601 also includes a memory or memory location 610 (e.g., random access memory, read-only memory, flash memory), an electronic storage unit 615 (e.g., hard disk), a communication interface 620 (e.g., network adapter) for communicating with one or more other systems, and peripheral devices 625, such as cache, other memory, data storage devices, and / or electronic display adapters. The memory 610, the storage unit 615, the interface 620, and the peripheral devices 625 communicate with the CPU 605 through a communication bus (solid lines) such as a motherboard. The storage unit 615 can be a data storage unit (or data repository) for storing data. The computer system 601 can be operatively coupled to a computer network ("network") 630 with the aid of the communication interface 620. Network 630 can be the Internet, an Internet and / or an extranet, or an intranet and / or an extranet in communication with the Internet. Network 630 is, in some cases, a telecommunications and / or data network. Network 630 can include one or more computer servers, which can enable distributed computing such as cloud computing. Network 630 can, in some cases, implement a peer-to-peer network with the help of computer system 601, which can enable devices coupled to computer system 601 to behave as clients or servers.

[0127] CPU 605 can execute sequences of machine-readable instructions, which may be embodied in a program or software. The instructions may be stored in a memory location, such as memory 610. The instructions can be directed to CPU 605, which can then be programmed or otherwise configured to implement the methods of the present disclosure. Examples of operations performed by CPU 605 can include fetching, decoding, executing, and writeback.

[0128] The CPU 605 can be part of a circuit, such as an integrated circuit. One or more other components of the system 601 can be included in the circuit. In some cases, the circuit is an application specific integrated circuit (ASIC).

[0129] The storage unit 615 can store files such as drivers, libraries, and saved programs. The storage unit 615 can store user data, such as user preferences and user programs. The computer system 601 can include one or more additional data storage units that are, in some cases, external to the computer system 601, such as located on a remote server that communicates with the computer system 601 through an intranet or the Internet.

[0130] The computer system 601 can communicate with one or more remote computer systems through the network 630. For example, the computer system 601 can communicate with a user's remote computer system (e.g., a seizure detection system manager, a seizure detection system user, a seizure detection data acquirer, a seizure detection system scribe, etc.). Examples of remote computer systems include a server, a personal computer (e.g., a portable PC), a slate or tablet PC (e.g., an Apple® iPad®, a Samsung® Galaxy Tab), a phone, a smartphone (e.g., an Apple® iPhone®, an Android®-enabled device, a BlackBerry®), or a personal digital assistant. A user can access the computer system 601 through the network 630. In some cases, the EEG device module and the seizure detection module will be located in the same computer system. In some cases, the EEG device module is located in one computer system, which is networked to a remote computer system that contains the seizure detection module. After performing seizure detection, the remote computer system can then transmit the seizure detection results to the computer system containing the EEG device module and to other remote computer systems that can be used for displaying the seizure detection results. In some cases, different parts of the seizure detection module may be located in different computer systems that are networked together.

[0131] Methods as described herein may be implemented using machine (e.g., computer processor) executable code stored on electronic storage locations of computer system 601, such as, for example, on memory 610 or electronic storage unit 615. The machine executable or machine readable code may be provided in the form of software. During use, the code may be executed by processor 605. In some cases, the code may be read from storage unit 615 and stored on memory 610 for easy access by processor 605. In some circumstances, electronic storage unit 615 may be omitted and machine executable instructions are stored on memory 610.

[0132] The code can be pre-compiled and configured for use with a machine having a processor adapted to execute the code, or can be compiled during run-time. The code can be provided in a programming language that can be selected to allow the code to be executed in a pre-compiled or as-compiled manner.

[0133] Aspects of the systems and methods provided herein, such as the computer system 601, can be embodied in programming. Various aspects of the technology may be considered as "products" or "articles of manufacture" in the form of machine (or processor) executable code and / or associated data, typically carried on or embodied in some type of machine-readable medium. The machine executable code may be stored on an electronic storage unit such as memory (e.g., read-only memory, random access memory, flash memory) or hard disk. A "storage" type medium may include any tangible memory of a computer, processor, or the like, or its associated modules, such as various semiconductor memories, tape drives, disk drives, and the like, which may provide non-transitory storage for software programming from time to time. All or part of the software may be communicated from time to time over the Internet or various other telecommunications networks. Such communication may, for example, enable loading of the software from one computer or processor to another, for example, from a management server or host computer to a computer platform of an application server. Thus, other types of media that may carry software elements include optical, electrical, and electromagnetic waves, such as those used across physical interfaces between local devices, through wired and optical terrestrial networks, and via various wireless links. Physical elements that carry such waves, such as wired or wireless links, optical links, or the like, may also be considered software-bearing media. As used herein, unless limited to non-transitory tangible "storage" media, terms such as computer or machine "readable medium" refer to any medium that participates in providing instructions to a processor for execution.

[0134] Thus, a machine-readable medium such as a computer executable code may take many forms, including, but not limited to, a tangible storage medium, a carrier wave medium, or a physical transmission medium. Non-volatile storage media include optical or magnetic disks, such as any of the storage devices in any computer or equivalent, such as may be used to implement the databases, etc., shown in the figures. Volatile storage media include dynamic memory, such as the main memory of such a computer platform. Tangible transmission media include coaxial cables, copper wire, and optical fibers, including the wires that comprise a bus within a computer system. Carrier wave transmission media may take the form of electric or electromagnetic signals or acoustic or light waves, such as those generated during radio frequency (RF) and infrared (IR) data communications. Common forms of computer readable media thus include, for example, a floppy disk, a flexible disk, a hard disk, a magnetic tape, any other magnetic medium, a CD-ROM, a DVD or DVD-ROM, any other optical medium, punch cards paper tape, any other physical storage medium with a pattern of holes, RAM, ROM, PROM and EPROM, Flash EPROM, any other memory chip or cartridge, a carrier wave carrying data or instructions, a cable or link carrying such a carrier wave, or any other medium from which a computer may read programming code and / or data. Many of these forms of computer readable media may be involved in carrying one or more sequences of one or more instructions to a processor for execution.

[0135] The computer system 601 may include or communicate with an electronic display 635 with a user interface (UI) 640, for example, programmed to control the seizure detection system and functionality, to provide a login screen for administrator access to the software, and / or to provide operational status health of the seizure detection system. Examples of UIs include, but are not limited to, graphical user interfaces (GUIs) and web-based user interfaces.

[0136] The methods and systems of the present disclosure can be implemented using one or more algorithms. The algorithms can be implemented using software, responsive to execution by the central processing unit 605. The algorithms can be, for example, software components described elsewhere herein, and may modulate seizure detection system parameters (e.g., EEG signal processing, machine learning algorithms, control strategies, seizure load, notifications, etc.).

[0137] While preferred embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. It is not intended that the present invention be limited by the specific examples provided herein. Although the present invention has been described with reference to the foregoing specification, the description and illustration of the embodiments herein are not intended to be construed in a limiting sense. Numerous variations, changes, and substitutions will now occur to those skilled in the art without departing from the present invention. Furthermore, it is to be understood that all aspects of the present invention are not limited to the specific depictions, configurations, or relative proportions set forth herein, which depend upon a variety of conditions and variables. It is to be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention. It is therefore contemplated that the present invention also covers any such alternatives, modifications, variations, or equivalents. The following claims define the scope of the present invention, and it is intended that methods and structures within the scope of these claims and their equivalents be covered thereby.

Claims

[Claim 1] The invention described in this specification.

Images