Methods, systems, and apparatuses for distinguishing electrical biosignals

Abstract

Methods and systems for distinguishing one or more types of electrical biosignal data from one or more other or different types of electrical biosignal data from within sensed electrical biosignals that are optionally sensed from a subject using one or more wearable scalp sensors. Methods may include receiving or generating electrical biosignals sensed from a subject, computing a plurality of features, decomposing the plurality of features, and distinguishing one or more types of electrical biosignal data in the sensed electrical biosignals from one or more different types of electrical biosignal data in the sensed electrical biosignals.

Description

METHODS, SYSTEMS, AND APPARATUSES FOR DISTINGUISHING ELECTRICAL BIOSIGNALSCLAIM OF PRIORITY

[0001] This application claims the benefit of U.S. Provisional Application No. 63 / 584,084, titled “METHODS, SYSTEMS, AND APPARATUSES FOR ANALYZING SINGLECHANNEL ELECTROENCEPHALOGRAM CONFIGURATIONS,” filed September 20, 2023, which is herein incorporated by reference in its entirety for all purposes.INCORPORATION BY REFERENCE

[0002] All publications and patent applications mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication or patent application was specifically and individually indicated to be incorporated by reference. This application incorporates by reference the following publications: U.S. Patent Publication No(s). US 2024-0293066 Al; US 2024-0293048 Al; US 2024-0293083 Al; and U.S. Patent No. 11,020,035.BACKGROUND

[0003] Electroencephalography (EEG) is a powerful technique for recording brain electrical activity. Conventionally, EEG recordings employ configurations with multiple electrodes to capture widespread brain dynamics. EEG measures electrical activity on the scalp, but the signal captured by an electrode is not solely representative of brain activity. The signal is a composite of numerous electrical activities originating from various sources, such as brain activity, muscle activity, vasculature activity, eye movements, skin conductance and sweating, and external interference. With respect to brain activity, neurons, the primary brain cells responsible for communication within the brain, produce electrical signals. When large groups of neurons fire synchronously, their combined electrical activity can be detected on the scalp. This is the primary signal of interest in EEG, including in examples herein that are used in the monitoring, detection and / or prediction of glucose states, or other metabolic health related information. With respect to muscle activity, muscles also produce electrical activity when they contract, known as electromyography (EMG). When muscles of the face, scalp, or neck (like those involved in blinking or clenching the jaw) contract, their electrical signals can be detected by EEG electrodes. This activity is often much larger in amplitude than brain activity. With respect to the vasculature, this artifact can arise due to tiny movements of the scalp electrodes caused by the pulsation of the underlying arteries, whichcoincide with the cardiac cycle. The pulsations can cause small shifts in the relative position of the electrode, leading to potential changes that can be detected in the EEG recording. With respect to eye movements, electrooculography (EOG) captures the electrical activity resulting from eye movements. When the eyes move or blink, a large electrical potential is generated, which can be detected by nearby EEG electrodes. With respect to skin conductance and sweating, the skin's electrical properties can change due to sweating or other factors, leading to small shifts in detected voltage. Additionally, electrical devices and power lines can introduce external interference noise into an EEG recording, and proper shielding and grounding of the equipment can minimize this.

[0004] The electrical activity generated by the brain is very small compared to the overall voltage measured by an electrode. Brain waves, for instance, are typically in the range of microvolts. In contrast, muscle activity can produce signals that are much larger, often drowning out the delicate brain signals if not properly addressed.

[0005] Depending on the particular application, and given the nature of these mixed signals, it may be crucial to process and analyze scalp data correctly to ensure that the observed activity genuinely represents brain dynamics and not other sources of electrical activity. Techniques such as filtering, artifact rejection, and Independent Component Analysis (ICA) are often employed to separate and remove non-brain-related signals from EEG data. As described in more detail herein, there may be situations wherein traditional techniques such as ICA are less than ideal or inadequate. For example only, some applications (e.g., singlechannel sensors) may acquire less patient brain data than is acquired with traditional EEG. While single channel configurations have been attempted for seizure detection, the signals of interest in seizure detection are generally so large and distinct there is no need to account for noise (unwanted electrical disturbances) that is sensed with the desired brain activity signals.

[0006] There is a need for systems and methods of use that can effectively distinguish, extract or identify brain activity data from other non-brain activity data from sensed electrical biosignals. There is also a more general need for systems and methods of use that can effectively distinguish, extract or identify one or more types of activity data from one or more different types of activity data.

[0007] The disclosure below in this Background section is understood to be related to uses and applications related to blood glucose state determination and / or prediction, and is thus understood to be an exemplary use of the inventive aspects herein related to identifying, distinguishing or extracting brain electrical activity from other sensed electrical biosignals that include non-brain electrical activity, which may optionally obtained from single-channel wearable sensors, but which may be sensed with wearable sensors not limited to singlechannel configurations as well. The inventive aspects herein may be used to detect other physiological states or conditions, such as brain states and / or in other settings such as, for example only, seizure activity, a sleep state of a subject, stroke status of a subject, or an emergency room (ER) setting to detect any of the brain states herein.

[0008] Blood glucose is one of the most important blood parameters to measure or understand generally, as abnormal levels can cause severe complications. For example, blood glucose levels above or below standard levels can cause significant medical issues including medical emergencies.

[0009] Hypoglycemia, which can require medical attention, is a state in which the blood sugar (glucose) level is lower than the standard range, often considered as a blood sugar of 70 milligrams per deciliter (mg / dL), or 3.9 millimoles per liter (mmol / L) or lower. While hypoglycemia may be related to diabetes management, it can occur and be associated with non-diabetes medical conditions and / or environments (e.g., ICU patients, fatigue). Reversing hypoglycemia may include high sugar food or drink (e.g., juice) and / or medicine in an attempt to raise the blood glucose level. Hypoglycemia, if left untreated, may lead to loss of consciousness and / or seizures.

[0010] Hyperglycemia is a condition when blood glucose levels are higher than the standard range, often considered above 180 milligrams per deciliter (mg / dL). Hyperglycemia, which is commonly linked to diabetes, occurs when the body has too little insulin (the hormone that transports glucose into cells), or if the body cannot respond to or use insulin properly. Hyperglycemia can, however, be associated with non-diabetes medical states or conditions and / or environments (e.g., ICU patients). If left untreated for long periods of time, hyperglycemia can cause damage to nerves, blood vessels, tissues and organs.

[0011] Existing technology can test current blood glucose levels in near real-time. For example continuous glucose monitors (“CGM”) include a small sensor inserted under the skin, such as the abdomen or arm. The sensor measures the interstitial glucose (“ISG”) level, which is believed to be indicative of blood glucose levels. The sensor may test ISG every few minutes, such as every five minutes. A transmitter can wirelessly send the information to a monitor, which may be part of an insulin pump, or which may be a separate device, such as a smartphone or tablet. CGMs are essentially detecting existing or current interstitial fluid glucose levels, and since ISG levels follow blood glucose levels by several minutes, the blood glucose levels provided by CGMs may be indicative of past blood glucose levels. Insulin pumps can automatically initiate insulin injection if glucose levels get too high. A threshold glucose level may be set (stored) in the CGM such that insulin may be delivered once the glucose level reaches the preset threshold. Individualized thresholds may be set for patients,and the thresholds may need to be reset over time (currently at office visits), such as if changes in the patient’s life necessitate resetting the threshold (e.g., going through puberty). Currently, the best insulin pumps for diabetes management achieve a peripheral glucose target range of 70 to 180 mg / dL <70% of the time, which is not ideal and still quite broad of a glycemic range.

[0012] Non-diabetic states, conditions and environments exist where it is desirable or necessary to know blood glucose levels. For example only, ICU patients, who may or may not be diabetic, typically have their blood glucose levels frequently checked, typically using blood from a finger prick, which can be tested on a small strip with a meter that provides the glucose levels.

[0013] Additionally, subjects who are pre-diabetic may want to be able to monitor (and optionally takes steps to manage) their glucose levels in an attempt to avoid becoming diabetic.

[0014] Additionally still, a subject who may be considered healthy may simply want more information about their body, their health, and their overall well-being, including an understanding of their glucose levels.

[0015] Alternative approaches are needed that can provide information about a glucose state of a subject without relying solely on a glucose sensor, and optionally without requiring a glucose sensor at all.SUMMARY OF THE DISCLOSURE

[0016] An aspect of the disclosure of is related to computational methods for distinguishing one or more types of electrical biosignal data from one or more different or other types of electrical biosignal data from within sensed electrical biosignals, where the sensed electrical biosignals are optionally sensed from a subject using one or more wearable scalp sensors.

[0017] In this aspect, the method optionally includes receiving or generating electrical biosignals (“biosignals”) sensed from a subject with one or more wearable scalp sensors, wherein the sensed electrical biosignals may include electrical biosignals from a plurality of different physiological sources;

[0018] In this aspect, the method optionally includes distinguishing brain activity data from non-brain activity data from within sensed electrical biosignals

[0019] In this aspect, the method optionally is stored on one or more devices, and optionally executable by one or more processors.

[0020] In this aspect, the method optionally includes, based on sensed electrical biosignals, computing a plurality of features, optionally for each of a plurality of epochs of time during which the signals were sensed.

[0021] In this aspect, the method optionally includes decomposing a plurality of computed features, optionally for each of a plurality of epochs of time, into constituent components.

[0022] In this aspect, the method optionally includes, based on a decomposing step, distinguishing one or more types of electrical biosignal data in the sensed electrical biosignals from one or more different types of electrical biosignal data in the sensed electrical biosignals.

[0023] In this aspect, the method optionally includes, based on a decomposing step, distinguishing brain activity data in the sensed electrical biosignals from non-brain activity data in the sensed electrical biosignals.

[0024] In this aspect, a distinguishing step optionally comprises distinguishing brain activity data from one or more of muscle activity data, vasculature activity data, eye movement data, or skin conductance and / or sweating data.

[0025] In this aspect, a distinguishing step optionally comprises computationally clustering brain activity data from non-brain activity data.

[0026] In this aspect, computing a plurality of features optionally comprises computing one or more statistical moments for each of one or more computed characteristics of biosignals in each of a plurality of epochs of time, optionally wherein one or more statistical moments comprise one or more of a mean, a variance, a skewness, or a kurtosis. One or more characteristics of biosignals optionally comprises one or more of power, amplitude, entropy, Hjorth mobility, or Hjorth complexity for each epoch, and wherein power optionally comprises power of one or more of delta, theta, alpha, beta, or gamma bands.

[0027] In this aspect, computing a plurality of features optionally comprises computing one or more characteristics of the biosignals in each of a plurality of epochs of time. One or more characteristics optionally comprise one or more of power, amplitude, entropy, Hjorth mobility, or Hjorth complexity for one or more epochs. Power optionally comprises power of one or more of delta, theta, alpha, beta, or gamma bands.

[0028] In this aspect, decomposing a plurality of features for each of a plurality of epochs of time into constituent components optionally comprises applying an empirical mode decomposition (EMD) algorithm over a plurality of variables and creating a plurality of intrinsic mode functions.

[0029] In this aspect, distinguishing brain activity data from non-brain activity data optionally comprises inputting constituent components into a clustering algorithm, wherein a clustering algorithm is optionally a UMAP algorithm.

[0030] In this aspect, the method optionally further comprises outputting the distinguished brain activity data.

[0031] In this aspect, the method optionally further comprises visually presenting on a display brain activity data that was distinguished from non-brain activity data.

[0032] In this aspect, brain activity data may be visually presented separately clustered from non-brain activity data.

[0033] In this aspect, the method optionally further comprises inputting distinguished brain activity data or information indicative thereof into a prediction model trained to predict a glucose state of the subject based on the distinguished brain activity data, wherein at least some of the non-brain activity data is not input into the prediction model; predicting the glucose state based on the distinguished brain activity data; and outputting to a subject device information indicative of the predicted glucose state.

[0034] In this method, the method optionally distinguishes one or more types of electrical biosignal data from one or more different types of electrical biosignal data, optionally comprising inputting a constituent components into a clustering algorithm such as a UMAP algorithm.

[0035] One aspect of the disclosure is a computational method stored on a device, executable by one or more processors, for extracting brain activity data from electrical biosignals sensed from a subject using one or more wearable scalp sensors.

[0036] In this aspect, the method optionally includes receiving or generating electrical biosignals (“biosignals”) sensed from a subject with one or more wearable scalp sensors.

[0037] In this aspect, the method optionally includes based on the biosignals, computing a plurality of features for each of a plurality of epochs of time.

[0038] In this aspect, the method optionally includes decomposing a plurality of features for each of a plurality of epochs of time into constituent components.

[0039] In this aspect, the method optionally includes, based on a decomposing step, extracting brain activity data.

[0040] In this aspect, the method optionally includes outputting the extracted brain activity data.

[0041] This aspect may include any other suitably combinable feature described herein, including the Summary section, Claims, Description, and figures.

[0042] One aspect of the disclosure is a system.

[0043] In this aspect, the system may include one or more processors, and a memory coupled to the one or more processors, the memory storing computer-program instructions, that, when executed by the one or more processors, perform a computer-implemented method.

[0044] In this aspect, the computer-implemented method may include distinguishing brain activity data in sensed electrical biosignals from non-brain activity data in the sensed electrical biosignals.

[0045] In this aspect, the computer-implemented method may include distinguishing one or more types of electrical biosignal data in sensed electrical biosignals from one or more different types of electrical biosignal data in sensed electrical biosignals.

[0046] In this aspect, the computer-implemented method may include extracting brain activity data.

[0047] In this aspect, the computer-implemented method may be adapted to perform any one or more suitably combinable steps herein.

[0048] In any of the aspects in the Summary, methods may optionally further comprise any suitable method or step related to improving overall health described or shown in this application, including, without limitation, providing personalized information related to improving metabolic health and / or predicting glucose states.BRIEF DESCRIPTION OF THE FIGURES

[0049] FIG. 1 represents an exemplary method of predicting glucose states.

[0050] FIG. 2 represents an exemplary system adapted to predict glucose states.

[0051] FIG. 3 represents an exemplary system adapted to predict glucose states.

[0052] FIG. 4 illustrates exemplary behind the ear locations for one or more wearable sensor.

[0053] FIG. 5 is an exemplary method of training a glucose state prediction model.

[0054] FIG. 6 is an exemplary method of training a glucose state prediction model.

[0055] FIG. 7 illustrates an exemplary system and / or method for training and / or prediction.

[0056] FIGS. 8, 9, 10, 11, 12, 13 A, 13B, 14, 15, 16A, 16B, 17A, 17B, 18A, 18B, 19A, 19B and 20 illustrate data, plots and other information from Example 1 herein.

[0057] FIGS. 21A and 21B illustrate exemplary systems herein, online optional online portion or care-team device.

[0058] FIG. 22A illustrates a common EEG setup and general head locations and letternumber designation for a plurality of electrodes.

[0059] FIG. 22B illustrates the location of additional electrodes A2, M2 (and similarly Al, and Ml), all of which are in behind the ear location, in Example 2 herein.

[0060] FIG. 23 A illustrates a table from Example 2 herein show significant correlations between electrode location, frequency band, and whether it was positive or negative correlation.

[0061] FIG. 23B shows the high negative correlation between glucose and EEG behind the ear and high positive correlation between glucose and EEG occipitally.

[0062] FIGS. 24 A and 24B illustrate data from Example 2 in which correlations between brain signals and glucose levels were examined at a plurality of time ranges before the glucose values.

[0063] FIGS. 25 A and 25B show data from an Ml electrode in the 9-12 Hz frequency at a time range of - 11 minutes to -6 mm.

[0064] FIGS. 26 A and 26B show data from an M2 electrode in the 9-12 Hz frequency at a time range of - 11 minutes to -6 mm.

[0065] FIGS. 27A and 27B show data from an 01 electrode in the 1-12 Hz frequency at a time range of - 11 minutes to -6 mm.

[0066] FIG. 28A and 28B show data from an 02 electrode in the 5-8 Hz frequency at a time range of - 11 minutes to -6 mm.

[0067] FIG. 29 illustrates an example of a visual representation of glucose state information.

[0068] FIG. 30 illustrates an example of a visual representation of glucose state information on an exemplary device.

[0069] FIG. 31 illustrates an example of a visual representation of glucose state information, including exemplary recommendation.

[0070] FIG. 32 shows an exemplary comparison between glucose levels with a traditional CGM and with some optional examples herein which are adapted with management steps to main glucose within a certain range.

[0071] FIG. 33 is an exemplary method of training a glucose state prediction model to predict a prandial event.

[0072] FIGS. 34, 35, 36, 37, 38, 39, and 40 show data from Example 3 herein.

[0073] FIG. 41 illustrates an exemplary wearable sensor.

[0074] FIGS. 42 and 43 illustrates an exemplary wearable sensor.

[0075] FIGS. 44 and 45 illustrate an exemplary wearable sensor locations, with FIG. 45 showing a behind the ear location.

[0076] FIG. 46 illustrates exemplary steps during optional manufacturing methods of some wearable sensors herein.

[0077] FIGS. 47A, 47B, 47C and 47D illustrate an exemplary wearable sensor.

[0078] FIG. 48 is a merely exemplary block diagram of an exemplary data processing systems, any part or all of which may be used with any embodiments of any of the inventions or embodiments set forth herein.

[0079] FIG. 49 illustrates an exemplary method of future glucose state estimation using a CGM.

[0080] FIG. 50 illustrates an exemplary method of managing glucose states based on values sensed from a CGM.

[0081] FIG. 51 illustrates an exemplary method of creating an association between ISG data and EEG data.

[0082] FIG. 52 illustrates an exemplary method of calibrating a glucose monitor using EEG data.

[0083] FIG. 53 illustrates an exemplary method of calibrating a blood glucose prediction method.

[0084] FIG. 54 illustrates an exemplary method of using ISG data and EEG data to perform at least one calibration process.

[0085] FIG. 55 represents a method of distinguishing brain activity data from non-brain activity data sensed with one or more wearable scalp sensors.

[0086] FIG. 56 represents a method of distinguishing brain activity data from non-brain activity data.

[0087] FIG. 57 represents a method of distinguishing brain activity data from non-brain activity data and / or extract brain activity data.

[0088] FIG. 58 represents a method of distinguishing brain activity data in sensed electrical biosignals from non-brain activity data in sensed electrical biosignals.

[0089] FIG. 59 represents a method of distinguishing one or more types of electrical biosignal data from within sensed electrical biosignals from one or more different types of electrical biosignal data from within the sensed electrical biosignals.

[0090] FIG. 60 represents a method of distinguishing brain activity data in sensed electrical biosignals from non-brain activity data in sensed electrical biosignals.

[0091] FIG. 61 represents a method of distinguishing one or more types of electrical biosignal data from within sensed electrical biosignals from one or more different types of electrical biosignal data from within the sensed electrical biosignals.

[0092] FIG. 62 is an example of a method that includes decomposing a plurality of features into constituent components, using an EMD approach.

[0093] FIGS. 63 A, 63B, 64A, and 64B are examples of distinguishing, extracting or identifying data using an exemplary clustering algorithm (e.g., UMAP), and also exemplify visually presenting data distinguished from other types of data.DETAILED DESCRIPTION

[0094] One inventive aspect of this disclosure includes methods of, and systems adapted for extracting brain activity data from electrical biosignals sensed with one or more wearable sensors, and which may optionally be, but not necessarily, single-channel wearable scalp sensors, examples of which are described herein. Extracting brain activity data, as that phrase or similar phrases are used herein, may also be referred to as distinguishing brain activity data from other non-brain activity data from within a larger set of sensed electrical biosignal data (e.g., from one or more wearable scalp sensors).

[0095] In merely exemplary applications, at a time subsequent to extracting or distinguishing brain electrical activity, the extracted data may be used in any of the methods herein (or devices if those methods are computer executable methods stored in a non-transitory memory) to detect real-time (or near-real-time) and / or forecast / predict future glucose states (and / or other health information). It is thus understood that any and all of disclosure herein related to extracting brain electrical activity from sensed electrical biosignals that are sensed with wearable scalp sensors may be incorporated with any of the other disclosure herein, and vice versa.

[0096] It is also understood that while the disclosure herein may refer to exemplary uses with single channel EEG configurations (single channel sensors), these are exemplary applications, and the innovative aspects herein may be used with electrical biosignals sensed with one or more other types of sensors.

[0097] Additionally, as set forth herein, extracted data (brain or non-brain activity data) may be used in a variety of applications, such as detecting or predicting information related to seizure activity of a subject, a sleep state of a subject, stroke status or other indicator of stroke or lack thereof of a subject, including in emergency room (ER) settings to detect any of the brain states herein.

[0098] An exemplary and non-limiting aspect of this disclosure is related to using distinguished data to predict a predicted glucose state of a subject, wherein the predicted glucose state is based at least partially on one or more brain activity signals or data sensed from the subject. Aspects of the disclosure herein provide approaches that facilitate minimally invasive or non-invasive devices and systems that can predict glucose states without requiring an implantable device and that do not rely solely on devices that provideinformation with a lag time as do glucose monitors that are adapted to measure interstitial glucose.

[0099] As reported in the Nature article “Spectro-spatial features in distributed human intracranial activity proactively code peripheral metabolic activity,” that is incorporated by reference herein, there was an observed association between intracranially sensed brain activity signals and peripheral glucose activity. Aspects of this disclosure provide an approach to predicting predicted states of a subject based at least partially on brain activity signals sensed minimally invasively, and as shown in examples herein, sensed non-invasively with one or more wearable sensors placed on the scalp of a subject. As discussed above, there may be a variety of reasons for a subject wanting to be informed of their glucose state(s), from managing diabetes to managing overall health and well-being, and everything in between.

[0100] As used herein, a “glucose state” may refer to a variety of aspects of the subject’s condition. Glucose state may optionally refer to predicted and / or calculated blood glucose values (e.g., 140 mg / dL (or 7.8 mmol / L)), a general characterization of blood glucose values such as a designation between “high,” “normal” or “low,” a characterization whether the subject is “in range” or “out of range,” and / or may be related to blood glucose values and / or interstitial fluid glucose values.

[0101] Benefits of some of the approaches herein include being able to predict glucose states without accessing intracranial locations of the subject, and not relying solely on glucose sensors (and optionally do not rely on glucose sensors at all). Aspects of the disclosure include wearable devices that can be worn in an ambulatory manner and are optionally sized and configured to be worn discretely on the scalp of a subject. Additional exemplary benefits of the approaches herein are that the predicted glucose states may optionally be predicted before the glucose state or event occurs and / or the predicted glucose states may be more accurate and reliable than existing glucose monitors. The approaches herein may provide a large number of people with beneficial health information that can enhance their overall health care and / or well-being, including providing access to their glucose state information that people previously did not have access to. While portions of this disclosure may describe predictions for diabetic or prediabetic subjects, the benefits may be realized by the entire population. For example, as is set forth herein, forecasting (future predictions) and / or predicting real-time glucose levels using sensed EEG data may have benefits for a wide variety of individuals, including use by healthy individuals who want more control over their personalized nutrition and health. In fact, some attention has been recently given to using CGM outside of diabetes, and the benefits of forecasting and real-time glucose monitoringusing EEG (as described herein) can similarly have benefits outside of diabetes in a wide variety of applications. The following exemplary list provides examples of applications of the innovations herein: athletics; overall health and wellness; gestational diabetes; obesity; metabolic syndromes; personalized glycemic profile (how their body clears and otherwise adapts to glucose); disorders such as stress or anxiety; stress of an infection (sepsis); personalized nutrition; personalized post-meal glycemic profiles.

[0102] Using EEG data to predict glucose levels may be more accurate than existing monitoring approaches, such as sensing ISG as is done with existing CGMs. For example, predicting blood glucose states in the future can arm the subject with a predictive outlook on their future blood glucose levels, whether primarily for informational purposes or providing them with actionable information to allow them to improve their health. Additionally for example only, and as described in more details below, methods herein that use EEG data to predict a glucose state (e.g. one or more predicted blood glucose values) during a sleep-state (asleep) may be more accurate information than existing CGMs, including predicting glucose states before they occur, which may be beneficial since some subject’s may be more likely to become hypoglycemic while sleeping.

[0103] FIG. 1 represents an exemplary and general method of predicting a glucose state of a subject 10. Method 10 includes, at step 11, sensing brain activity signals or data from a subject. Method 10 includes, at step 12, predicting a predicted glucose state of the subject based at least partially on the sensed brain activity signals of the subject. Method 10 optionally includes, at optional step 13, communicating information indicative of the predicted glucose state (e.g., providing information on a display related to glucose state; providing a warning that ranges are trending out of range; communication information to a care-team). Method 10 optionally includes, at optional step 14, managing one or more aspects of the subject’s glucose state (e.g., automatically initiating a management step such as delivering insulin; or providing a recommendation to the subject such as recommendation on a time to exercise and / or eat, optionally via a display of a personal subject device).

[0104] Sensing at step 11 may comprise, and may preferably comprise, non-invasively sensing brain activity signals. Non-invasive sensing using wearable sensing devices herein can provide a large number of people access to personal health information and the benefits thereof compared with more intrusive approaches (e.g., solely intracranial and / or implantable).

[0105] Sensing at step 11 may comprise, and may preferably comprise, non-invasively sensing brain activity signals at one or more specific locations on the head of a subject, and optionally the scalp. The one or more specific locations are, at least in most beneficial usesherein, fewer than all the locations that are typically used during a typical electroencephalogram (EEG) where electrodes are worn at many locations about the head. One of the benefits of approaches herein is that brain activity signals can be sensed and utilized to predict glucose states in an ambulatory manner that is not available with typical full-head EEG recordings in clinical settings, providing myriad individuals access to health information heretofore inaccessible and even unknown. As used herein, “EEG” refers generally to recordings of brain activity electrical signals sensed from a subject at one or more particular locations, but in other uses relative to traditional full head EEG, it may refer to clinical recording of brain activity signals in which signals are obtained with many electrodes placed about the subject’s head in a clinical setting.

[0106] As described below, Examples 1-3 herein provide data that blood glucose levels may be predicted using brain data recorded from wearable sensing devices that may be placed at one or more behind-the-ear locations (e.g. M1 / M2 / A1 / A2 as shown herein) and / or at one or more occipital locations (e.g., 01 / 02 using 10-20 EEG set nomenclature).

[0107] FIG. 2 illustrates an exemplary and general system 20 adapted for predicting a predicted glucose state of subject 21. System 20 includes at least one wearable sensing device 22 (wearable sensor) that is configured to sense brain activity signals from a subject when worn on the head of the subject. System 20 also includes a computer executable method with instructions that cause the performance of predicting a predicted glucose state. The computer executed method may be stored in the wearable sensing device, in a different device (e.g., a personal device) or location (cloud).

[0108] FIG. 3 illustrates an exemplary system 30 that is adapted to predict a glucose state of subject 31. System 30 includes at least one wearable sensing device 32 (wearable sensor) that is configured to sense brain activity signals from subject 31 when worn on the scalp of subject 31. System 30 also includes a device 33 in communication with sensing device 32, wherein the device 33 is adapted to receive information from sensor 32 and output instructions to initiate a communication indicative of the predicted glucose state based on and in response to the predicted glucose state. Device 33 may be a device accessible by subject 31 (e.g., a smartphone or smartwatch) and / or device 33 may be accessible by a care-team for subject 31, optionally one not accessible by subject 31.

[0109] Prediction models herein may reside or be stored on one or more wearable sensing devices and / or other device in wireless communication with the wearable sensing device. Depending on processing power, it may be beneficial to store the prediction models on the personal device or care-team device.

[0110] An important aspect of some of the examples herein is related to wearable sensors that can be worn on the head of a subject and in a location that can reliably and non-invasively detect brain activity signals that can be used to predict predicted glucose states of the subject. FIG. 4 illustrates an exemplary behind-the-ear location 40, which is not meant to refer to a strictly defined region on the subject’s head. Location 40 refers to a region on the head that includes surfaces behind (posterior) to the ear as well as surface in the Cephalad direction relative to the ear. Location 40 may also extend to some extend inferior (or Caudal) to the ear. Behind the ear locations as used herein include the locations labeled M1 / M2 / A1 / A2. [OHl] Behind-the-ear location 40, including locations proximate the mastoid process, may be beneficial for non-invasively recording brain activity signals because there is a region of skin on which the sensing device may be positioned and maintained during sensing (and without requiring a full-head EEG set-up). FIG. 4 illustrates merely illustrative outer profiles of wearable sensors 42a, 42b and 42c positioned in a behind the ear location. A wearable device need not be placed entirely within a predefined region to be considered to be “behind- the-ear,” such as exemplary location 42c. All locations 42a-42c are considered behind-the-ear locations as described herein.

[0112] One aspect of this disclosure is related to methods of training machine learning algorithms or models to create a trained glucose state prediction model. Training predictive glucose state models herein may include one or more of supervised or unsupervised training techniques.

[0113] Methods of training models herein may include providing sensed brain activity signals (raw or processed to some extent) and a history of glucose values, whereby the model may select one or more features of the sensed brain activity signals most correlated with one or more aspects of the glucose values. For example, models may select one or more frequencies, frequency ranges, and / or frequency bands most correlated with glucose values. Additionally, for example, models may select one or more most correlated time periods between brain signal sensing and sensed glucose values. FIG. 5 illustrates an exemplary method 50 of training a predictive model. Method 50 includes, at step 51, providing sensed brain activity signals from a subject. Method 50 includes at step 52, providing glucose values sensed from the subject, and which have a temporal relationship with the brain activity signals sensed from the subject. Glucose values may be sensed or measured from one or more of a CGM; an implantable sensor; optical and non-invasive sensors; or from finger prick measurements. Method 50 includes, at step 53, selecting one or more features of the brain activity signals more correlated with glucose values than other brain signals features (e.g., one or more frequencies, frequency ranges, and / or medically established frequency bands).Once trained, at step 54, the model is adapted to receive or generate as input one or more features of subsequently sensed brain activity signals (e.g., brain signals within one or more frequency bands), and predict a predicted glucose state based at least partially on the brain activity signals.

[0114] Methods of training models herein may include providing a model with one or more features of sensed brain activity signals (e.g., one or more frequencies, frequency ranges, and / or medically established frequency bands) as inputs with glucose values as targets. Once trained, the model can predict a predicted glucose state based at least partially on subsequently received brain activity signals that include the one or more features. FIG. 6 provides an illustrative method 60 of training predictive models herein. Method 60 includes, at step 61, providing one or more features of sensed brain activity signals from a subject. Method 60 includes, at step 62, providing glucose values sensed from the subject that have a temporal relationship with the brain activity signals sensed from the subject (e.g., sensed from one or more of a CGM; an implantable sensor; optically and non-invasively; or from finger prick measurements). Method 60 includes, at step 63, the learning algorithm finding or determining a relationship (functions / parameters) between the one or more features of the sensed brain activity signals and one or more glucose values. Method 60, at step 64, includes, once trained, the prediction model is adapted to receive as input one or more features of subsequently received brain activity signals, and predict a predicted glucose state.

[0115] The glucose state prediction models or methods herein may comprise computer executable instructions executable by one or more processors, and may be stored in any device, such as a wearable sensor, a subject device, a care-team computing device, and / or a cloud device that is adapted to be in communication with any of the devices herein.

[0116] Features of brain activity signals that may be used in training and / or during the prediction step may include a variety of features of brain activity signals. For example, features may include powerband activity in certain frequencies, such as one or more frequencies between 0 - 50Hz. For example only, features may include powerband activity in one or more of the delta band (.5 - 4 Hz), theta (4 -7 Hz), alpha (8-13 Hz), beta (13-30 Hz), gamma (e.g., 30-60 Hz).

[0117] A mere example of a machine learning method can be adapted to identify a subset of the most important features from non-invasively sensed brain activity signals from one or more single channel locations, and then subsequently fit a linear regression model using the reduced set of features, implemented examples of which are described below.

[0118] FIG. 7 illustrates a merely exemplary system and method of training models and using the trained models to predict predicted glucose states, any aspect of which may beincorporated into any training and / or prediction method herein. In exemplary FIG. 7, brain activity signals and glucose values sensed from a CGM are shown to be used in the training process, and once trained, the model is adapted to predict glucose states based on subsequently recorded brain activity signal s / data.

[0119] Any of methods / algorithms herein may optionally be trained on one or more of normal / healthy individuals, hyperglycemia (e.g., diabetes; hyperglycemia ICU, sepsis, traumatic brain energy; diabetic ketoacidosis) or hypoglycemia states. Any of methods / algorithms herein may optionally be trained on data from the general population, some subset of the population, or personally trained based solely on data from an individual.

[0120] Training methods herein may include training on real-time glucose state values (i.e., predicting real-time blood glucose values), or training on future glucose states (predicting a glucose state at one or more times that are in the future relative to when the non-invasively sensed brain activity signals were sensed by a wearable sensor.

[0121] Preferably, trained prediction models herein have a relatively high accuracy for predicting blood glucose levels (optionally future predictions), although it is envisioned that it may not be critical that the accuracy be at or above a particular threshold, or consistently at a particular threshold, as having some degree of accuracy can still provide benefits, such as if, for example only, the predictions are used with other types of glucose monitors and sensors. Training methods herein can be adapted to select one or more features of brain signal data most correlated with glucose values.

[0122] As mentioned above, brain activity signals may be trained on glucose state data regardless of the manner in which the glucose state data is derived. For example only, brain activity signals can be trained on blood glucose values / levels sensed from existing CGMs, from measured finger prick blood glucose values (measured with a meter), blood glucose values from systems that are configured to optically detect glucose values (e.g., a smartwatch), or implantable sensors that sense glucose at a location more invasive than CGMs (e.g., from within a blood vessel). Trained models herein are adapted to predict a glucose state (e.g. values), and the “state” may be based on the manner in which the values were initially detected or sensed. It is thus understood that the glucose state predictions as described herein may be predictive based on the blood glucose values and the manner in which they were obtained or sensed. Regardless of the manner in which the glucose values are sensed, prediction models based on that data can still provide a predicted state, which provides the subject the predicted values before they occur.

[0123] While there may be instances where general features may be used to predict a glucose state for at least parts of a population, there may be a need or preference to determine one ormore features of sensed brain data for a particular subject (patient specific) that provides more accurate blood glucose prediction (e.g., due to a higher correlation between the one or more features and glucose levels). The phrase features as used herein in this context can refer to a variety of patient parameters, such as, without any limitation, location of one or more wearable sensors (e.g., using one or more of: one or more behind the ear location or one or more of 01 / 02 locations); one or more frequency bands (e.g. alpha, beta, gamma, theta); one or more frequency ranges (e.g., 0-50 Hz, or any subrange within 0-50Hz, and not necessarily the discrete “bands”), whether the range spans multiple bands or not; one or more particular and discrete frequencies; features related to brain signal sensing (e.g., AUC, Power, Phase synchrony or coherence between sensors); predictive temporal lag with higher accuracybetter correlation (examples of which are described below), etc.

[0124] In some aspects, developing a prediction forecasting method may include determining one or more features of extracranially and non-invasively-sensed brain activity signals that are more correlated with future glucose levels, which facilitates greater accuracy in predicting future glucose states, optionally in the absence of requiring blood glucose sensing (such as with a CGM, invasive, or non-invasive monitor). In these applications, while sensed glucose data may initially be used to develop or train predictive methods (e.g. supervised and / or unsupervised), the use of trained models (e.g., in a consumer product) may be adapted to rely on one or more features of non-invasively sensed brain activity data (from one or more wearable sensors) that can be used to predict a glucose state without necessarily requiring blood glucose measurements. In these aspects, peripheral markers / measures may optionally be used that can increase the predictability of a glucose state (blood glucose values) with brain signals, such as any of the other patient biosignals herein (e.g., heart rate, heart rate variability, ECG, skin conductance, etc.).

[0125] The example that follows is an illustrative, non-limiting, example of a process or method of training a model or method that includes determining or selecting one or more features in non-invasively sensed brain activity signals from a wearable sensor adapted for single channel sensing (or at least operable in bipolar sensing even if optionally also operable for monopolar sensing), as is shown in FIG. 5. Once trained, the model is adapted to predict future blood glucose levels based on subsequently received single channel EEG data. Alternatively, one or more features can be established as inputs and used to train a predictive model with glucose data as targets (FIG. 6).

[0126] Example 1

[0127] In this non-limiting example, data from 1 subject with type 1 diabetes and 3 nondiabetic subjects was sensed over the course of several days to about 45 days. The data thatwas sensed included brain activity signals or data sensed while wearing a wearable non- invasive single channel brain activity sensor, such as a sensor described in or similar to the single channel sensors in U.S. Patent No. 11,020,035, worn unilaterally at a location behind the ear on the scalp (skin), examples of which are shown in FIG. 4 (optionally one of an M1 / M2 / A1 / A2 location). Additionally, interstitial glucose levels were also contemporaneously sensed using a commercially available CGM traditionally worn and used. While a variety of software tools may be used (e.g. Python), the example herein utilized MATLAB® software.

[0128] While the following example includes several steps, it is understood that this is meant to be illustrative and enabling, but that it is an exemplary method and in other methods not all of these steps are necessarily performed or in this order. After sensing raw brain activity signals, the raw brain activity signal data may undergo data conversion, such as converting it into a particular toolbox, depending on the software used, for processing (e.g., convert the brain activity signal data into fieldtrip format for processing). The raw converted data is shown in the plot in FIG. 8.

[0129] The converted data was filtered with a low pass filter (50Hz) and high pass filter (5 Hz), but other filters may be used (e.g., high pass 1 Hz, low pass 70 Hz). After filtering, the data was plotted, as shown in FIG. 9. In this example, 5-50 Hz signals were processed, but in other examples other ranges may be processed such as .5-50 Hz. Signals were epoched into 5-minute segments, and single time points were rejected if they fell outside of 3.5 median average deviations from the median signal amplitude. Further, entire epochs were rejected if their power x frequency curve for that time bin did not follow the physiologically-defined 1 / f curve.

[0130] A power analysis was then performed for the range 5-50 Hz, in 1 Hz intervals, and the power fast fourier transform (FFT) was plotted, as shown in FIG. 10. An optional artifact rejection step was performed, which included binning the data into 5-minute periods. Spikes were removed that exceeded a certain median average deviation (MAD; 3.5 MAD), and the data without spikes was then reshaped and plotted, as shown in FIG. 11. A power analysis was then performed again after the spikes were removed. The FFT was plotted, as shown in FIG. 12.

[0131] Training predictive methods that are trained to predict a subject’s blood glucose state (real-time blood glucose or future blood glucose) may comprise receiving as inputs one or more aspects of processed brain activity signals (e.g., from a wearable sensor adapted to record a single channel (e.g. with a single pair of sensing electrodes), but wherein the wearable sensor may also be adapted to record in monopolar mode with a third referentialelectrode, examples of which are provided herein), and sensed glucose data (e.g., glucose values). In this particular example, this was performed individually for each subject, and is illustrative of the optional personalized training approach that may be used (or which may be necessary for accuracy) to train predictive models herein.

[0132] Next, the glucose monitor data was overlapped with or aligned in time with the brain activity data. In this example, one day of data was overlapped / aligned at a time, but other time periods can be used. The process further included matching the sampling rate of the brain activity signals and the glucose data. The process further included correlating all glucose data and brain activity data, running cross correlations (and plotting, an example of which is shown in FIG. 13 A), averaging individual frequencies into bands, and running cross correlations per band (and plotting, an example of which is shown in FIG. 13B). The process further included running correlations for sleep and awake time periods, optionally extracting indices that fall during standard sleep and wake time periods. The process further included, for each of sleep and awake time periods, extracting brain activity signals and glucose data, correlating frequency data, and averaging data into bands and correlating the bands. The correlations can be plotted or otherwise analyzed.

[0133] The process further included developing lag indices of, for example, 10-minute intervals, but other intervals can be used, such as, for example without limitation, from a 1 mm lag to a 20-minute lag, including any specific time interval therein (e.g., 5 min lag, 8 min lag, 15 min lag). For each lag, for each frequency band, linear mixed-effects (LME) model analysis was run with predicted blood glucose and fixed variable EEG. Residuals between the model and actual values of blood glucose were plotted, which are shown in FIGS. 14 and 15 for the asleep and awake correlations.

[0134] Least absolute shrinkage and selection operator (LASSO) was performed, the results plotted and saved (plots shown in FIGS. 16A, 16B, 17A, and 17B, for awake and sleep, respectively). Predicted blood glucose values were extracted from the model training on single-channel EEG and CGM values, from a single lag. FIG. 16A shows all power estimates in 1 :50 Hz intervals, while FIG. 16B is an average of those into four frequency bands (alpha, theta, beta, and gamma).

[0135] FIG. 20 plots % accuracy of predicting blood glucose levels vs EEG temporal prediction (hours) from the two subjects in this example on the trained model. As can be appreciated, the accuracy of predicting glucose values on the time scale up to 12.7 hours was 90% or greater from zero to 12.7 hours in advance, demonstrating significant accuracy. From this plot, and as part of a process of developing / creating / training a glucose state predicting algorithm / model, an exemplary approach may include selecting or determining that one ormore particular times or one or more particular time ranges with higher correlations is preferred as more accurate. For example only, an approach may include selecting or determining that during time period 2102, as labeled, the accuracy is at least 94%, and that time range of about 8.3 hours to 12.7 in advance, or one or more epochs within that range, is a preferred time range to input sensed brain data into the model and predict one or more glucose values for that particular subject. Temporal time lags may change for each subject over the course of the subject’s daily routine, such as changing from day to day. Optional systems and methods herein may thus benefit from periodically sensing not just brain activity data but also estimate blood glucose values with a glucose monitor, and adjust the temporal lag as may be desired or needed. Time range 2112 is a mere example of a temporal lag range for subject 2 that might be selected or determined due to the relatively high % accuracy of prediction. In some applications, a specific time point may be selected, such as time point 2104 or 2114, as a temporal lag for the relative subjects, due to a relatively high % accuracy of prediction. The temporal time lags in this context, if they are discussed in the context of a personalized lag with relatively high accuracy for that particular subject, are referred to herein as a personal time lag or other similar term.

[0136] Additionally, any of the Apps herein may be adapted to provide the functionality to allow the subject or care-team to select a time or time range in the future for which they would like predictions. For example only, any of the Apps herein (“Apps” described below) may be configured to visually present on a display or screen of the device a selectable user input to choose or select a time for which they would like predictions. For example only, a subject may prefer predictions 2-15 minutes in advance (for more near-term predictions), or, for example, 1 hour in advance, or 30 minutes in advance, for example. Any of the Apps herein can be configured to operate and predict with the selected or chosen time period in advance, and make glucose state predictions that are predicting that subject’s glucose levels that time in the future (relative to when the brain activity was sensed).

[0137] Training methods herein may utilize one or more features of the sensed brain activity signals (including one or more patient specific features that may lead to greater accuracy for the subject) on glucose states (however sensed / measured), and once trained, the model is adapted to receive subsequently sensed brain activity signals and predict glucose states (e.g. values), wherein the trained model may be specifically trained on the one or more features of the sensed brain activity signals. That is, the trained models herein (which may be stored on any of the subject devices herein, such as part of any of the Apps herein) may beneficially not need to receive or utilize all of the sensed brain activity signals. For example, only, trained models or other algorithms may utilize or receive a subset of the single-channel sensed brainactivity signals that have a more significant correlation with glucose values and more accurately predict glucose values, and which may be a more significant correlation between power of one or more bands and glucose values, described above.

[0138] Example 1 as described above illustrates that at least a portion of single-channel EEG, from a behind-the-ear location on the skin of a subject, is predictive of blood glucose values, and thus can be used to predict glucose values based on single channel EEG. It also showed that there may be, at least following the processes above in Example 1, some patient-to- patient variability in the % accuracy of prediction, such as when single channel signals may be most predictive. In fact, it may optionally be desirable or even necessary to train prediction methods on data from the intended subject for a personalized prediction App, which is referred to herein as personal or personalized prediction. That is, each subject may have a personalized App that is trained to process EEG data in a way that will be the most predictive for that individual, which may also be considered to be an individual calibration step or process prior to predictive use of the system. Additionally, a subject may benefit from periodically recalibrating their App to ensure it continues to process brain signals in a way that is most predictive for that subject. Personalized temporal lag, as described herein, is a mere example of a patient feature in or related to single channel, non-invasively sensed, EEG signals that can be used in the prediction methods herein.

[0139] Features of brain activity data upon which the models may be trained may be personalized to an individual feature, or they may be features that can be utilized across larger populations of people, or even all subjects using a predictive method.

[0140] An additional example of a feature of brain activity data is a subset of all frequency bands between 0 Hz and 50 Hz in the prediction method. That is, the prediction methods herein may optionally predict glucose state using only a subset of frequency bands between 0-50 Hz. The subset may include one or more well-described frequency bands (one or more of delta, theta, alpha, beta, and gamma (or a portion of gamma from 32 - 50 Hz), one or more frequency ranges generally, even if the range spans more than one well-described band, or one or more discrete frequencies. A subset, as that term is used herein, may have a more significant correlation with glucose values and / or the processing of which can lead to greater accuracy of prediction. For example, Example 1 performed above indicated that higher frequencies within the 0-50hz range, such as within the beta and / or gamma bands, may be more correlated to future blood glucose levels from 18 minutes to up to 13 hours in advance. Processing only a subset, such as one or two bands (or one or two frequency ranges, or certain frequencies) may beneficially require less power during processing, which may be performed in a wearable sensor or in an App stored on a subject device, for example. Similarto a personal temporal lag (where predictions are for a particular time in the future), it may optionally be beneficial or necessary to determine for each subject which frequency bands or frequency ranges are correlated with higher accuracy for glucose state prediction, or which have a more significant correlation with glucose states, either real-time or future predictions. It is understood that there may be more than one feature of brain signal data that may be personalized to a particular subject, such as a particular temporal lag (which may be reset over time), or a particular frequency range of EEG data to be processed. In some embodiments, correlation weight factors may be applied to one or more frequencies or frequency ranges. It is of course understood that features, while they may be personalized to a subject, may also apply to larger populations, and may in fact be used by all subjects using predictive methods herein if the accuracy is sufficient due to a sufficient correlation.

[0141] As can be seen in FIGS. 17B and 19B, the extracranially sensed EEG from a single channel was highly correlated with current or real-time blood glucose values sensed using a CGM during sleep time, as indicated by the zero “0” lag. It is of note that following the process above, the extracranially-sensed single channel EEG data was generally not as predictive of future glucose values during sleep time as during awake time. Optionally, methods and systems herein may use a future prediction approach that relies more heavily on brain activity signals sensed during the awake time than during asleep time, which again may provide a benefit of processing less information, thus requiring less power, while obtaining more accurate predictions. Relying more on brain activity signals sensed during the awake time than during the asleep time may optionally comprise applying a weighted factor to awake time signals that is greater than a weighted factor for asleep time signals, for example only. Relying more on brain activity signals sensed during the awake time than during the asleep time may optionally comprise not relying at all on sleep time signals for future glucose state predictions. Any of the methods and systems herein may include this feature of relying more heavily on awake time EEG data for predicting future glucose levels.

[0142] Based on the observed high correlation discovered between the single channel extracranially-sensed brain activity data from a behind-the-ear and on-skin location during sleep and blood glucose levels sensed using a CGM, any of the methods and systems herein may further comprise using single channel extracranially-sensed brain activity data during sleep to one or more of predict real-time glucose levels or near-term (e.g., 1-15 minutes in advance) future glucose levels, or even to calibrate a glucose monitor worn by the subject. In fact, in some uses, single channel brain activity signals sensed optionally from a behind the ear location during sleep may be used solely for calibrating or resetting a glucose monitor, and optionally not for blood glucose prediction. Additionally, brain activity data sensedduring sleep and used to predict glucose levels in the near-term future may allow near-term future predictions to be made during sleep time that are more accurate and provide more timely and actionable information than currently available CGMs. This may be highly beneficial to diabetic patients who are at risk of night-time episodes of hypoglycemia, particularly for children. It is generally reported that existing (CGMs) are relatively not as accurate during sleep time, perhaps because their readings follow or lag the actual, real-time levels (if the monitor's readings peak or fall too late compared to actual current blood glucose levels). This makes it more difficult for patients to timely recognize the potential hypoglycemic events that are not uncommon in diabetes, because they are asleep. Some subjects preempt this and set their monitor sensitivities during sleep time within a narrower range than the ranges during their wake time, which can cause more frequent alerts and wake them up more during the sleep time, the disturbances from which are known to be suboptimal from a general health perspective. Devices and systems that extracranially sense single channel brain activity signals during sleep can use the sleep time brain data (with or without the sensed glucose levels) to more accurately predict real-time glucose levels, and can even predict the levels in the near-term future to better avoid adverse glucose events.

[0143] One aspect of the disclosure is a method of predicting blood glucose levels of a subject while the subject is asleep, comprising: extracranially sensing single channel brain activity signals of a subject while the subject is asleep from a device optionally worn at a behind-the-ear location on the skin; optionally sensing the subject’s interstitial glucose (ISG) or blood glucose with any type of glucose monitor while the subject is asleep (e.g., from an optical monitor in a watch); and determining blood glucose levels while the subject is asleep using the extracranially-sensed brain activity signals and optionally the sensed glucose monitor data (e.g., ISG data). Predicting blood glucose levels using at least the extracranially- sensed brain activity signals may achieve a higher accuracy in detecting real-time blood glucose levels than using existing glucose monitors alone. In some instances, it may not even be necessary to use a blood glucose monitor, and rely on sensed brain activity signals instead (although in these cases, other non-glucose monitor biosignals may also optionally be used, such as heart rate, heart rate variability, etc.). Due to the observed high correlation between brain data and sleep-time levels, the glucose levels can even be predicted in the near-term future at sleep, providing technological improvement that could benefit many people.

[0144] FIG. 21A is an additional merely exemplary glucose prediction system (GPS), optionally continuously sensing (CGPS) that can be adapted and configured for training models and / or predicting blood states. System 2100 includes optional portal device 2108 accessible by a care team member, such as a physician. The GPS may be adapted to predictblood glucose values by itself (e.g., without input from other devices like CGM 2104), but optionally it may be adapted to also predict using or based on one or more inputs from one or more devices (e.g., a CGM 2104 as shown, or other devices that are Bluetooth enabled), which may optionally communicate with the GPS via Bluetooth, as shown. System 2100 includes a wearable sensor 2102 (optionally a second on the other side of the head behind the other ear) and subject device 2106, features of which are described elsewhere herein. It is understood that the optional CGM 2104 shown in FIG. 21 may be replaced with a different type of glucose monitor or sensor, such as an invasive monitor (e.g., implanted in an arm or abdomen) or a non-invasive monitor such as an optical monitor (e.g., integrated into a watch). FIG. 21B illustrates the system 2100 in FIG. 21 A, but without the optional glucose monitor. All reference numbers and descriptions from FIG. 21A apply to FIG. 21B.

[0145] Example 2

[0146] A second herein example used a similar process to Example 1, but additionally utilized more EEG data from a typical EEG setup. In the second example, both glucose data and EEG data were contemporaneously sensed / monitored from a subject for at least 2 days. The subject was wearing an existing continuous glucose monitor (CGM) sensing glucose data, while undergoing a common EEG recording with a setup shown in FIG. 22A, and with the addition of Al (left) and A2 (right) electrodes in exemplary behind the ear locations, as shown in FIG. 22B. Ml and M2 electrodes are also shown in behind the ear locations.

[0147] The preprocessing of data included: 1) only selected “awake” period, and epoched to 10 sec window; 2. aligned the timepoints in CGM and EEG signals; 3. selected 27 EEG channels and common average re-referenced; 4. band-passed [1,99] and notch-filtered

[5565] ; 5. eye blink artifact removed by ICA; and 6. bad trial rejected based on MAD,

[0148] The steps below illustrate exemplary power extraction and CGM matching: 1) scalp EEG power is extracted using fast fourier transform in FieldTrip toolbox for each 5-s epochs; 2) The trials of power from 1 min before and after to CGM timepoint were selected to align the CGM value; 3) The trials and each frequency band were averaged; 4) linear regression model was used to obtain the correlation between frequency band and the glucose (similar to in Example 1).

[0149] FIG. 23 A represents a Summary of brain region and bands for which there was a significant correlation between the power and glucose level, and whether it was a positive or negative correlation. Positive correlation refers to higher the power in that particular band corresponding to higher the glucose levels; negative correlation refers to less power being associated with higher glucose values. FIG. 23B shows the high negative correlation betweenglucose and EEG behind the ear and the high positive correlation between glucose and EEG occipitally.

[0150] This Example also assessed a variety of time windows prior to the sensed glucose values, as shown in FIGS. 24A and 24B, with the table shown in FIG. 24B providing channels / locations with significant correlation with glucose values in each of the time windows. The significant correlation is related to the table in FIG. 23 A for the particular brain region and band of significance.

[0151] As shown, at least one of occipital 1 and occipital 2 (01 and 02), the location of which is shown in FIG. 22A, shows significant positive correlation in all time ranges from T1 = -13 and T2 = -5. Additionally, as shown, Ml and M2 showed significant correlation in at least some of the time ranges from T1 = -13 and T2 = -5. Additionally, as shown, F7 and F8 showed significant correlation in at least some of the time ranges from T1 = -13 and T2 = -5.

[0152] The data showing significant correlation at these locations and in these bands supports placement of wearable sensors at one or both of Ml and M2 locations (an example of behind the ear) and / or one or both 01 and 02 locations. Additionally, any of the trained models herein may be trained on any of the bands shown in FIG. 23 A and / or specifically related to a particular band. For example, any of the models herein may be trained to receive and predict based on only a subset of frequencies from 0-50 Hz, such as at least some frequencies between 0-20 Hz (but not frequencies greater than 20 Hz), such as at least some frequencies between 0-15 Hz, such as at least some frequencies between 0-12 Hz, such as from 5-12 Hz, or from 5-8 Hz, or from 9-12 Hz (alpha). In the table in FIG. 23 A, the data from 02 (positive correlation at 5-8 Hz) is an example of a feature that is within a particular band (theta), while the data from Ml and M2 (negative correlation at 9-12 Hz) is an example of a feature within a particular band (alpha). In the table in FIG. 23 A, 01 is an example of a brain region with features that span multiple bands, in this case, spanning delta, theta, and alpha. FIGS. 25B, 26B, 27B, and 28B show exemplary plots of raw power and glucose readings for particular electrode locations, at particular frequencies (as shown) and from 11 minutes before to 6 minutes prior to glucose value (T=-l 1 min and T2 = -6 min). CGM levels referenced against the left axis, time is on the bottom axis, and power for that location and frequency range refenced on the right axis. In color, CGM data is blue and power in red. FIGS. 25A, 26A, 27A, and 28A show plots of z-scored power and glucose for select electrode location and frequency range.

[0153] One aspect of this disclosure is related to, after predicting a glucose state, outputting instructions to initiate a communication indicative of the predicted glucose state. In some examples, a computing device (optionally a subject device, such as shown in FIGS. 3 and 21)is adapted to output instructions to create one or more visual communications of information related to the predicted glucose states. In some examples, the predicted glucose states (e.g., predicted values) may be visually presented on a display of the device, which may be controlled by an App on the device.

[0154] FIG. 29 is a non-limiting example of a visual representation of predicted future blood glucose levels on a display of a device, such as a smartphone or other computing device. For example, an App may comprise a computer executable method that is adapted to cause- predicted future blood glucose levels to be displayed, the computer executable method stored in a non-transitory memory, the method comprising: receiving as input extracraniaHy-sensed EEG data or information indicative of extracranially-sensed EEG from a subject, and causing a visual representation of the predicted future blood glucose levels to be displayed on a display of a device. In exemplary FIG. 29, time is on the X-axis and the predicted blood sugar levels on the Y-axis (which may be shown with any unit desired, such as mg / dL). The left side of the X-axis can be considered 3 am, where predicted out or range levels are shown. The software can be adapted to display visual indicators (e.g., the red icons) that can indicate predicted peaks and valleys. The range or any of the ranges herein may be user-adjustable on the display so the user can modify any high or low level for the range that is shown as “inrange.” This is one way to provide for personalized glucose management, which is an improvement in technology.

[0155] FIG. 30 is an illustration of a computing device (e.g., smartphone) with a display, wherein glucose state information is being visually presented on the display. The information communicated in FIG. 30 can be the same or similar to information shown in FIG. 29 and 31.

[0156] FIG. 31 is an additional non-limiting example of a visual representation of predicted blood glucose values being displayed on a display of a device, such as a smartphone or other computing device. FIG. 31 also presents a visual representation of suggested actions to be taken and when, which are correlated with high and low trends on the predicted glucose levels. The suggestions shown are merely exemplary, but provide an exemplary of personalized health management provided by the methods and systems herein.

[0157] FIG. 32 is an illustrate example of some examples herein that can predict and maintain blood glucose levels at desired levels. FIG. 32 illustrates exemplary graphs comparing traditional CGMs (top) and optional systems and methods herein that are adapted to predict and maintain blood glucose levels within (or more closely aligned with) a target range (bottom). It is noted that systems and methods herein do not need to include a management aspect, and may provide significant benefits by provide health information (e.g., predicted glucose state).

[0158] One aspect of this disclosure is related to methods and systems that allow a subject to select or choose (optionally via an App interface) a period of time during which they wish to receive predictions and / or other information alerts such as alerts. For example only, a subject may only wish to receive predictions during sleep, or only near the time and including the time they will consume a meal. The App may similarly be configured to select when to process brain signals and when not to process brain signals. In the example of selecting sleeptime as a time for the prediction process to occur, the subject may simply wish for the system to predict for more near-term events, such as hypoglycemia. In the example of only predicting near meal-time and shortly thereafter, the subject may want information about glucose levels related to the meal, for example. The Apps herein may thus include an input feature that allows the subject to select from one of a predefined time periods during the 24- hour cycle (e.g., from 9am - 12pm) for prediction to occur, or a user selectable feature where the subject can disable and reenable the prediction process (e.g., enable when about to sleep, and turn off when awake). Alternatively, the systems herein may be adapted to detect when the subject has fallen sleep based on the sense brain signals, and then begin to process the brain signals and predict predicted glucose states using the prediction model.

[0159] One aspect of the disclosure is related to methods and systems that allow a subject to choose the prediction window, or the time in advance they wish to receive prediction information about predicted glucose states (relative to one or both of when the brain data was sensed or relative to the current time). For example, a user may wish to only be informed of relatively near-term future predictions. For example, a subject may wish to select a prediction window of 20 minutes. In this example, the x-axis shown in FIGS. 29-31 may only include a time period of 20 minutes (e.g., every minute demarcated on the x-axis with a dash). For example only, a subject may wish to be informed mostly about sleep glucose swings (e.g., to avoid hypoglycemia), and may only wish to focus on near-term predictions. The Apps herein may include a user input feature that is adapted to allow the user to select or choose the prediction window. Choosing a prediction window may optionally be part of an initial user set-up process, wherein it may be beneficial to train a model based on the desired prediction window. Any of the prediction methods herein may thus be adapted to predict 30 minutes in the future (or any number less than 30), and optionally not more than 30 minutes in the future (or any number less than 30). An exemplary use scenario is during sleep, and a subject may receive alerts up to 30 minutes in advance if they are trending towards or in a hypoglycemic state.

[0160] One aspect of predicting glucose states herein is related to predicting prandial blood glucose events, optionally post-prandial. Exemplary post-prandial events include, but are notlimited to, post-meal peaks in glucose levels or time until a post-meal peak in glucose levels. An exemplary rationale for systems and methods adapted to predicting postprandial glucose events is that elevated postprandial glucose concentrations may contribute to suboptimal glycemic control. Additionally, postprandial hyperglycemia is one of the earliest abnormalities of glucose homeostasis associated with type 2 diabetes and is markedly exaggerated in diabetic patients with fasting hyperglycemia.

[0161] As used herein, the term “meal” includes any consumption or ingestion, and may simply be a snack, a drink, etc., and is understood not to be limited to an amount or type of consumption or ingestion.

[0162] After a meal, there is a rise in blood glucose levels. Glucose monitoring data (e.g., from CGMs) has been shown in the literature to be able to identify post-meal glucose peaks. One aspect of this disclosure is a method of training an algorithm or model to predict, using as input extracranially sensed single channel brain activity signals from a subject and glucose values estimated using an existing CGM (but any other type of sensor / monitor may be used to obtain glucose data). A merely exemplary method of training an algorithm may include: obtaining an indicator of consumption by a subject (e.g., an input from a subject that food has been consumed, optionally via an App); obtaining blood glucose measurements from the subject (optionally with a CGM, an implanted glucose monitor; or a noninvasive glucose monitor such as a watch; etc.); and obtaining extracranially sensed brain activity signals from the subject. The method of training can further include temporally identifying on the glucose measurement readings (with glucose values) when consumption has occurred (such as via input from the subject about when food was consumed), and identifying (e.g., visually or automatically identifying) a post-prandial glucose peak subsequent in time to the consumption (e.g., 5 to 90 minutes, for example), such as including the timing postconsumption and / or amplitude of the peak. An algorithm / model can then be trained to identify patterns or other features of the brain activity signals (and optionally other subject measurements, such as heart rate, heart rate variability, and / or others described herein) preceding the glucose peak that are predictive of the subsequent glucose peak. After the algorithm has been trained using the EEG data and monitored glucose data, subsequent brain activity signals can be used as input to the trained model to predict the subject’s postprandial glucose peaks (and which optionally may be performed without monitoring glucose levels with a glucose monitor). Any of the aspects and description above related to training machine learning model is fully incorporated by reference into the post-prandial aspect of the disclosure herein.

[0163] FIG. 33 illustrates an exemplary method 2600 of training a model to predict a prandial event, with a post-prandial event as the example. At step 2601, the method includes providing sensed brain activity data or signals from a subject. Step 2602 includes identifying (manually or automatically) a post-prandial peak (or other prandial event) on glucose values sensed from the subject that have a temporal relationship with the brain activity signals sensed from the subject. Step 2603 includes selecting or identifying one or more features of the brain activity signals correlated with the post-prandial event (or other prandial event). At step 2604, once trained, the trained prediction model is adapted to receive subsequently sensed brain activity signals and predict a post-prandial event.

[0164] In alternative training methods, it may not be necessary to expressly label or identify an event such as a meal or peak, if recorded glucose data is available. For example, in healthy individuals, a relatively large increase in glucose (for example "large" may be about an increase of 1.5-2 mg / dL / min) may be able to be identified without the manual labelling of a meal.

[0165] Subjects may have individualized blood glucose (BG) changes after a meal. For example, one subject may have a post-meal peak 30 minutes after a meal, while a second subject consuming the same meal may have a peak 45 minutes after the meal. Additionally, some subjects may respond with a greater change in glucose levels following meal consumption than others, such as if a first subject does not experience a significant increase in BG while a second subject does in fact experience a significant increase in BG following the same meal. Personalized or individualized knowledge about a post-prandial glucose event (e.g., a BG peak) can provide valuable medical information to the subject. An exemplary but non limiting benefit of an algorithm that is adapted to predict post-prandial BG events such as BG peaks is that a subject can be informed about at least one of the timing or the amplitude of the peak, which may be beneficial to a wide variety of subjects. For example, it may provide information to a subject about when to take insulin and / or what dose of insulin to administer. Additionally, for example, a subject monitoring their overall health may simply want to know when their BG levels may be elevated, such as with a post-meal peak. Additionally, for example, an athlete may benefit from knowing when an elevated or peak BG level is likely to occur, as well as a predicted amplitude of the BG level. Additionally, for example. A subject may wish to avoid certain foods if they know some meals provoke BG levels that are quite elevated. There are many reasons why an individual may benefit from knowing about an elevated BG level, including a peak.

[0166] Any of the methods herein, upon predicting a post-prandial peak, can communicate information related to the peak to the subject, such as visually providing a predicted time of peak and / or the predicted glucose value of the peak on a display of the subject device.

[0167] One aspect of the disclosure continues to utilize input from the subject about meal consumption as a way to modify and / or improve the EEG BG prediction algorithms herein.

[0168] One aspect of the disclosure is the at least partial use or reliance on both of sensed glucose levels and sensed EEG data, the use of which together may, for example, improve the accuracy of algorithms herein. For example only, sensed EEG data may help improve the accuracy of glucose sensors, and / or sensed blood glucose data from glucose sensors may help improve the accuracy of EEG sensors.

[0169] Example 3

[0170] FIGS. 34 - 37 illustrate aspects of a merely exemplary data set that is related to and descriptive of the disclosure herein related to post-prandial or post-meal glucose event prediction (e.g., peaks) and training methods related to the same, such as the exemplary training method in FIG. 6. In this exemplary data set, there were three individuals: one prediabetes, one type 1 diabetes, and 1 non-prediabetes (who may otherwise be referred to as healthy). 2-3 days of EEG data and CGM data were obtained from the subjects at the same time. The healthy subject logged all meals during the time, and thus any meal information on any chart is related to the healthy individual.

[0171] FIG.34 illustrates blood glucose readings for the healthy subject over a period of time, including time of logged meals (again, the logging of meals may not be necessary). BG was recorded with a CGM. Post-prandial glucose peaks are labelled following three logged meals, as well as suspected CGM artifact, which is known to occur to some extent with CGMs.

[0172] FIG. 35 illustrates healthy individual recorded glucose levels vs time, by date, with day and night hours labeled. FIG. 36 illustrates recorded glucose levels vs time for the prediabetic subject, by date, with day and night hours labeled. FIG. 37 illustrates recorded glucose levels vs time for the Type 1 diabetic subject, by date, with day and night hours labeled.

[0173] FIG. 38 shows a series of EEG plots created showing Frequency (Hz) on the y-axis (or left axis) vs Time (s) on the x-axis (bottom axis) vs Power (color scale on the right axis) of how EEG varies in the day vs night from 60 minutes before a known glucose peak to 60 minutes after a known glucose peak. The middle plot for each of day and night is the time of the peak, with 60 minutes before at the top and 60 minutes following the peak at the bottom.

[0174] FIG. 39 illustrates an exemplary blood glucose recording (which is also representative of any sensed glucose data herein), with consumption (in this case food) at 18:28 hours and a post-meal peak 30 minutes later at 19:03 hours.

[0175] FIG. 40, similar to FIG. 38, shows a series of EEG plots with Frequency (Hz) on the y-axis vs Time (s) on the x-axis vs Power (color scale on right), from a time period spanning 60 minutes before a peak until 60 minutes after the peak, at 15 minute increments, (as labeled) with the peak being designated as time “0 minutes” (center plot in FIG. 40). The meal is labeled at 30 minutes before the peak.

[0176] FIGS. 38 and 40 illustrate an aspect of this disclosure, which includes creating or receiving (which may occur automatically) and utilizing 2-dimensional plots relating EEG data with Frequency (Hz), Time, and Power (color scale on right) as an input in any of the training methods or trained algorithms herein. In alternatives, the order of the axes may be modified. The 15-minute time increments in FIG. 40 are exemplary and not limiting in any way. For example, the time increments may be from 5 seconds to 30 minutes, for example.

[0177] In any of the examples herein, methods of predicting include sensing EEG data and predicting postprandial peaks within 5 min - 90 mins of the predicted peaks.

[0178] Once an algorithm is trained on brain signal data and known postprandial peaks, the algorithm can optionally be trained to predict other blood glucose variations outside of the postprandial peaks.

[0179] The system shown in FIG. 7 is illustrative of exemplary systems and methods of using training algorithms and predicting post-meal glucose events.

[0180] One aspect of this disclosure is related to measuring or detecting 1st and 2nd order momentum of EEG data (1stand 2ndderivative) as an input to any of the methods herein. Related thereto is the use of the 1st or 2nd order momentum of EEG data to dynamically determine the optimal time window for extracting EEG information to predict glucose levels. These methods or steps may be performed by a computer executable method, which are described in more details herein.

[0181] One aspect of this disclosure is related to wearable devices and systems that are adapted to sense or record one or more different types of biosignals, and in some particular embodiments they are adapted to sense brain activity signals / data, and optionally adapted to sense additional biosignals. Any of the description that follows may be used in any of the brain activity signal / data sensing and glucose state prediction described herein (future and / or real-time), and vice versa. While the disclosure describes wearables in the context of sensing brain activity signals, it is understood that one or more concepts herein may have broader applicability outside of brain activity signal sensing (e.g., the suction ports described below)and may have applicability to devices that do not include one or more electrodes. The wearable sensing devices and systems herein may be referred to generally as wearable sensors, wearable sensing devices, sensing systems, or wearables. All of the disclosure herein related to wearable devices and systems for sensing the subject’s brain activity signals / data (EEG data) is incorporated by reference into this portion to this disclosure, and vice versa.

[0182] Wearable sensors described herein are, unless indicated the contrary, considered “all- in-one,” or self-contained wearable sensors, which allows them to be worn by the subject as they go about their daily life (e.g., without wires connected to a console) wherein the wearable sensors wirelessly communicate data (e.g., via Bluetooth) to a second device (which may be a subject device or other device such as a care-team device).

[0183] FIG. 41 illustrate a schematically illustrative wearable sensing device (i.e., wearable sensor) 3400, with optional components and features in dashed lines. Any of the features of Sensor 3400 may be incorporated with any other wearable sensor herein, and vice versa.

[0184] The wearable sensing devices herein may be configured with one or two-way wireless communication with a different device (e.g., Bluetooth), such as a subject / user device (e.g., smartphone, smartwatch). In some examples, the sensing device is adapted to continuously sense brain activity signals from the subject, for at least some continuous period or epoch of time, and continuously communicate (for at least some continuous period or epoch of time) in real or substantially real-time the raw or processed EEG data to an App on a subject device, which is described elsewhere herein.

[0185] The wearable sensing devices herein may include one or more memory units adapted to store sensed or processed brain activity signals, at least temporarily (e.g., up to 8 hours). For example, in some embodiments the data can be stored temporarily until it is communicated to a subject device, or until communication can be re-established with a user device if communication was interrupted.

[0186] In one exemplary, non-limiting embodiment shown in FIG. 42 and FIG. 43, wearable sensing device 3510 (which may be considered wearable sensor 3510) includes a wearable sensor 3512. In this example, wearable sensor 3512 includes flexible electrode housing 3513 and a plurality of electrodes (shown) secured to the flexible electrode housing 3513. System 3510 further includes electronics member 3514 that is configured and adapted to be releasably coupled to wearable sensing device 3512 so that, when coupled, it is in communication with the plurality of electrodes. Electronics member 3514 may include connector pins as shown or other electrically conductive elements, each of which are adapted to create an electrical connection to one of the electrodes. Electronics member 3514 may be releasably secured to wearable sensor 3512 in a variety of ways, such as mechanical and / ormagnetic coupling. The views of FIGS. 42 and 43 show first surfaces of each of the plurality of electrodes facing away from the flexible electrode housing 3513 so as to be positioned to be coupled to a surface of a subject, an example of which is shown in FIGS. 44 and 45 (either directly coupled or indirectly coupled to the surface of the subject). The views of FIGS. 42 and 43 also show examples of second surfaces of each of the plurality of electrodes optionally disposed within electrode wells in the flexible electrode housing 3513.

[0187] One of the plurality of electrodes may be positioned and adapted for use as a ground electrode, an example of which is shown in FIG. 42. The plurality of electrodes may also include a plurality of sensing or recording electrodes, examples of which are shown in FIG. 42 (labeled as “recording pair"). The electronics member can be adapted with a processor such that the first and second sensing electrodes can sense in either monopolar mode (each recording with the ground) or in bipolar mode (recording between the recording pair), which can provide an advantage over some devices that can only sense in bipolar mode. In monopolar mode, the third optionally larger electrode as shown is the ground, or referential electrode.

[0188] Wearable sensors herein, such as those in FIGS. 42, 43, 47A-47D, provide the ability for single or dual sensing modalities, compared with some alternative wearable devices configured for only single channel sensing.

[0189] As a mere example, first and second sensing electrodes are optionally spaced 1.0 mm - 4 mm apart. A ground electrode may be spaced from the nearest of the first and second sensing electrodes a greater distance than a spacing between first and second sensing electrodes, an example of which is shown in FIGS. 35 and 36.

[0190] Any of the plurality of electrodes herein may be rigid, flexible, or comprise both rigid and flexible elements. In some non-limiting examples, sensing electrodes herein may comprise one or more of steel, gold, carbon, or silver.

[0191] Any of the electrode housings herein (e.g., 5313 in FIGS. 42 and 43) may comprise a flexible body, wherein at least a portion of the body is flexible, optionally the entirety is flexible. In some embodiments, a flexible electrode housing comprises one or more polymeric materials. In the example of FIGS. 42 and 43, the flexible electrode housing may comprise a silicone, and may consist of a silicone.

[0192] In non-limiting examples, any of the flexible electrode housings herein (e.g., 5313 in FIGS. 42 and 43) optionally has a length of 20 mm - 60 mm, optionally a width of 5 mm - 25 mm, and optionally a height of 1 mm to 8 mm. Dimensions in FIGS. 42 and 43 are understood to be illustrative.

[0193] The wearable sensor 3512 is adapted to be adhered to a subject’s body (e.g., skin, e.g., scalp), exemplary locations for of which are shown in FIGS. 44 and 45, with FIG. 45 illustrating an exemplary behind the ear location. The sensing devices herein (e.g., 3512) may be adapted to be adhered to the subject’s skin with one or more of a wet or dry adhesive, suction, or other non-permanent attachment. Some sensing devices are adapted to be adhered to the subject’s skin without an adhesive, such as if using suction or other non-chemical bonding forces such as Van der Waals forces. Sensing devices herein may optionally include one or more suction ports in a flexible electrode housing (as shown in the example of FIGS. 42 and 43), wherein the suction ports may be configured as wells in the flexible electrode housing, an example of which is shown in FIG. 42. In some alternatives, not shown, the electrodes may be adapted to be adhered to the skin at least partially with suction.

[0194] Suction ports herein may be configured to suck skin in and push air out when a user pushes on the bandage to adhere it to the subject. The electrodes are also pulled towards the skin, providing electrode-skin contact (including indirect contact). Additionally, suction ports may optionally be under the electrodes so that electrodes additionally get pulled or sucked towards the skin.

[0195] While not shown, the sensing devices herein may further comprise an adhesive adapted to at least help adhere the sensing device to the subject’s skin. For example, a flexible electrode housing may include an adhesive layer covering at least a portion of the flexible electrode housing. For example, an adhesive layer may cover one or more of a plurality of electrodes or a flexible electrode housing.

[0196] An exemplary advantage of the wearable system in FIGS. 42-45 is that, if desired or needed, wearable sensor 3512 can be de-coupled from electronics member 3514, and replaced with a new wearable sensor 3512. Additionally, wearable sensor 3512 may optionally be cleaned or replaced and re-coupled to electronics member 3514.

[0197] An exemplary advantage of the system in FIGS. 42-45 is that wearable sensor 3512 may comprise a flexible housing 3513, which may allow better conformability to the skin of a subject (e.g., scalp), which may facilitate better electrode-skin contact.

[0198] One aspect of this disclosure is a wearable device that comprises a plurality of optional suction adhesion ports on a first side of the device. In this aspect, the wearable may or may not include electrodes. An exemplary benefit of such a wearable is that adhesives may irritate the skin of certain subjects, and suction may be a less reactive mechanism to adhere the device to the skin. An additional benefit is the option to remove and replace the device on-demand (described in more detail below) if continuous operation is not required.Additionally, wetness from natural skin perspiration may enhance the suction and thus helpadhere the device to the skin, which may help with electrode-skin contact. Furthermore, suction may allow for some amount of perspiration or other liquid to be retained between the electrode-skin interface, improving the electrical contact.

[0199] In examples in which the wearable device 3512 can be replaced, such as if the housing 3513 or electrodes deteriorate, a plurality of wearable sensors 3512 may be packaged together, similar to a package of bandages. This allows the subject to easily replace an existing wearable sensor with another without a significant interruption in EEG sensing, if that is desired. Maintaining EEG sensing as continuously as possible may be highly desirable, such as if the subject needs to or wants to regularly predict and / or manage their blood glucose levels.

[0200] In some additional examples, the electronics member is configured with a communication module (illustrated in FIG. 41), such as for Bluetooth capabilities to communicate data to a user device / App, such as one or more of a smartphone, a computer, or an insulin pump or other drug delivery device.

[0201] In some embodiments, the optional electronics member is primarily rigid and the wearable sensor 3512 is flexible. The electronics members herein may, however, optionally be at least partially or completely flexible, such as if the electronic member is a mixture of flexible and rigid components to, for example, facilitate conformability. For example only, an ASIC and a power source (e.g., battery) may be rigid, but a printed circuit board (PCB) to which they are attached may be flexible such that the whole package can optionally be curved.

[0202] Flexible electrode housings herein may be manufactured in a variety of skin tones or other colors, or they may be clear.

[0203] In some embodiments, the entire system is completely non-invasive. In some embodiments, the system may further include one or more microneedles, as illustrated in FIG. 41 to optionally sense ISG when worn about the head.

[0204] The wearable sensors herein may be waterproof, even when there is a rechargeable electronics member.

[0205] In any of the embodiments herein, the optional electronics member may optionally include one or more of a memory (which may store at a minimum of 10 hrs, for example); a signal amplifier and processor that is adapted to allow for monopolar and bipolar mode sensing; adaptation for bidirectional communication; Bluetooth capability; include one or more sensors adapted for impedance measurement to monitor adequate skin contact for recording; and / or a rechargeable power supply.

[0206] Any of the wearable systems and / or devices herein may be part of a wireless network of wearable computing devices, which may be referred to as a body area network (BAN), a wireless body area network (WBAN), a body sensor network (BSN), or a medical body area network (MB AN). The systems and devices herein may have a plurality of modes of communication that allows them to communicate with one or more different types of devices, such as other devices in a BAN.

[0207] Any of the wearable sensors herein may include one or more sensors adapted to detect or sense chewing such that the system is adapted as a chew indicator (e.g., a chew counter). For example, one or more sensors adapted to sense an indication of chew may be integrated into one or more components of wearable system 3510 or 400.

[0208] FIG. 46 illustrates exemplary molds for manufacturing exemplary wearable sensing devices, including the housings herein.

[0209] In some applications, the wearable systems (e.g., an example of which is in FIGS. 42- 45) may not necessarily be worn continuously, but may be worn non-continuously, such as being worn only for brief periods of time, such as (without limitation), from 30 seconds to 2 hours (including any range therein). This may be helpful if periodic EEG readings / monitoring are all that is needed or desired, and may be referred to as an on-demand use rather than continuous use. This may be analogized to periodic finger prick glucose reading that are currently used, such as to monitor glucose levels and / or to calibrate a CGM. This approach may be useful, for example only, if a subject is primarily concerned with blood glucose levels at certain times of the day, such as after a meal when glucose levels are expected to increase. In an exemplary use, a subject may place the sensor on their body at the time of or just prior to meal-time, and leave it only for a brief period of time (5 minutes to 2 hours, for example), or long enough for the EEG data to be used to predict or monitor a prandial event. All disclosure herein related to post-prandial glucose level prediction and monitoring is incorporated into these on demand examples.

[0210] This on demand approach may also be used as an alternative to finger-prick testing, as a way to get an immediate real-time blood glucose prediction (through an App, which can present the glucose reading on the display of the user device).

[0211] Additionally, an on-demand use may be used to train an algorithm including periodically wearing the sensor while taking finger-prick glucose measurements, such as wearing the sensor for an hour before and after finger-prick measurements are obtained.

[0212] Additionally, as an example, on demand use may help a subject understand their personalized glycemic profile, such as how their glucose levels change after a meal, or certain types of food or drink.

[0213] In alternative examples, not shown, the optional electronic member (or package) may include one or more reinforcing or support elements that are adapted to reinforce or provide support to the electrode housing (e.g., housing 3513) at the location of the suction ports. The one or more reinforcing or support elements may be sized, configured and adapted to reinforce or bolster the suction port walls and help reduce the tendency of the suction ports to retract and lose their suction, which prolongs skin adhesion. In some merely exemplary embodiments, the one or more reinforcing elements may have pronged configurations, or other configurations that are adapted to provide support and reinforcement to the suction cup walls.

[0214] The disclosure that follows describes alternative wearable sensing devices (wearable sensors), but which may include one or more features from the aforementioned wearable sensing systems (e.g. more than two electrodes, one or which may be a ground electrode, and which may be configured for monopolar and / or bipolar mode sensing).

[0215] FIG. 47A - 47D illustrate a merely exemplary wearable sensing system 400. System 400 includes flexible housing 401 and a plurality of electrodes (as shown) secured to the flexible housing. A portion of each of the plurality of electrodes is within the housing (as shown in the transparent rendering in FIG. 47B), while surfaces of each of the plurality of electrodes are exposed, or facing away from the flexible housing 401, as illustrated in FIG. 47C. While FIGS. 42-45 provides an example of a system with detachable electronics, FIGS. 47A-47D provide an example of a system that is essentially an integrated system, with the electronics embedded therein and not intended to be detachable from a flexible member and reused. Certain features from the exemplary system in FIGS. 42-45, however, may still be present or incorporated into the exemplary system in FIGS. 47A-47D.

[0216] As in the example shown in FIGS. 42-45, exemplary wearable sensor 400 includes a first pair of sensing electrodes (e.g., the two electrodes to the left in the figures), and a ground electrode (the single electrode on the right in the figures). This allows the system to record in both bipolar mode (e.g., between the two sensing electrodes on the left) and monopolar mode (e.g. each of the two left electrodes referenced against the electrode on the right spaced further away as the common ground or reference electrode). The monopolar signal, while more susceptible to artifact due to the voltage measured across a greater distance between sensing electrode and the common reference electrode, has a much greater amplitude and greater sensitivity to neural signals, thus advantageous in wearable systems that record brain activity signal data from a subject. The capability to record in monopolar mode can also help reduce noise, which may be important when recording from localized regions of the head and recording relatively low power brain signals. Systems herein thus provide advantages overbrain activity signal recording systems that have electrodes in close proximity and are adapted to record only in bipolar mode.

[0217] In system 400, the distance between the two sensing electrodes disposed in relatively close proximity to one another may be from 1 mm to 7 mm (measured from center of electrode to center of electrode, as shown).

[0218] In system 400, the distance between a nearest of one of the sensing electrodes and the reference ground electrode may be from 8 mm to 35 mm, an example of which is shown as 15 mm.

[0219] In system 400, flexible electrode housing 401 comprises a skin facing surface 402 configured and positioned to be placed against tissue on a head of subject (directly or indirectly), and a second side 403 spaced from the skin facing surface. The skin facing surface 402 optionally has, as is shown in the side view of FIG. 47B, a concave or generally concave shape, and optionally wherein the second exposed side is at least partially planar, as shown. The merely optional concave curved surface of the skin facing surface may be configured to improve placement behind the ear or on the scalp, but depending on the flexibility of the wearable device, the wearable may not have a curved skin facing surface. In some embodiments, the skin facing surface has a radius of curvature from .5 cm to 4 cm along at least a portion of the surface, optionally with a constant radius of curvature along the entire surface. It some embodiments the radius of curvature may vary to some extent along the surface of the first tissue side.

[0220] Skin facing surfaces of each of the plurality of electrodes face away from the flexible housing to facilitate placement against the tissue, either directly against tissue or indirectly against tissue, as shown in FIG. 47B and 47C. In this particular example, surfaces of each of the plurality of electrodes face away from an optionally concave skin facing surface of the flexible housing, as shown.

[0221] Wearable sensor 400 illustrates an example where the plurality of electrodes are aligned linearly or substantially linearly along a length “L” of the flexible housing, as shown.

[0222] Wearable sensor 400 illustrates an example where the height of the flexible housing (e.g. “H” dimension labeled in FIG.40b) may be from 3 mm to 10 mm.

[0223] Wearable sensor 400 illustrates an example where the length of the flexible housing (e.g. “L” dimension labeled in FIG. 40b) may be from 12 mm to 40 mm.

[0224] Wearable sensor 400 illustrates an example of a flexible electrode housing that has, in a top or bottom view (e.g., FIGS. 47A and 47C), a rectangular shape or generally rectangular shape. The flexible electrode housing may have other shapes, however, in top and bottom views.

[0225] As shown and as described herein, the wearable sensor may further include a power source (e.g. a battery), optionally a rechargeable power source, embedded in the electrode housing (which is optionally flexible). The electrode housing may optionally be adapted to inductively recharge a battery, and optionally via the plurality of electrodes, which can enhance water resistance of the device (i.e. a port is not required for a charging cable).

[0226] With any of the wearable sensors herein, the plurality of electrodes may be rigid, optionally gold plated.

[0227] With any of the wearable sensors herein, the plurality of electrodes may be rigid and may comprise one or more polymeric materials, optionally further comprising conductive epoxy.

[0228] With any of the wearable sensors herein, the wearable may be water resistant, and may comprise embedded epoxy to improve the water resistance.

[0229] With any of the wearable sensors herein, a flexible electrode housing may comprise a flexible overmolded wearable housing, manufactured by overmolding.

[0230] With any of the wearable sensors herein, the plurality of electrodes may extend proud of the skin facing surface or side of the electrode housing.

[0231] With any of the wearable sensors herein, a flexible overmolded wearable housing may comprise a silicone.

[0232] Sensor 400 (or any other wearable sensor herein) may include, as shown, an adhesive patch or layer disposed on a skin facing surface of the electrode housing. The system may include a removable layer (e.g. paper) that when removed from the adhesive exposes the adhesive to facilitate adhering the electrode housing to the skin of a subject. For example, an adhesive layer may be a polyester fabric, such as (without limitation) 3M™ Medical Tape 4578.

[0233] With any of the wearable sensors herein, an adhesive layer may leave exposed skin facing surfaces of the plurality of electrodes, or it may cover the skin facing surface of the electrodes. Adhesive layers may have selectively conductive regions at the location of the electrodes to create a monolithic patch (i.e., no electrode openings). To further improve skinelectrode contact, any of the wearable sensors herein may incorporate one or more spring- loaded or spring-like electrode contacts to accommodate curvature and / or anatomical differences between users.

[0234] Any of the wearable sensors herein may include a power source (e.g. battery) embedded within a flexible electrode housing (e.g., as shown in FIG. 47A and 47B). It may also optionally include kinetic charging capabilities, such as triboelectric generators, to harvest movement energy and extend battery life.

[0235] Any of the wearable sensors herein may include an EEG ASIC embedded within a flexible electrode housing (e.g., as shown in FIG. 47A and 47B).

[0236] Any of the wearable sensors herein may include a Bluetooth module disposed within a flexible electrode housing (e.g., as shown in FIG. 47 A and 47B), the system adapted to communicate information to an external device (e.g., a smartphone).

[0237] Any of the wearable sensors herein may include one or more memory structures disposed within a flexible electrode housing (e.g., within an EEG ASIC) adapted for on-board storage, at least temporary storage of sensed information or information indicative of sensed information.

[0238] Any of the wearable sensors herein may include a processor within a flexible electrode housing, such as within an EEG ASIC.

[0239] Wearable sensor 400 (or any other wearable sensors herein) may be configured for real-time (or near real-time) streaming of sensed brain activity signals to an external device, which optionally has stored thereon any of the Applications described herein.

[0240] Sensor 400 (or any other wearable sensors herein) and / or an Application stored on a subject device that is in communication with sensor 400, may be configured to provide an alert (e.g., a vibratory alert, audible alert) if the system detects an event, such as an abnormality or condition outside of what is expected or desired.

[0241] System 400 and any of the Apps herein may be configured for two-way communication, optionally wherein the Application is adapted to allow a user to provide input on EEG tracings / recordings or other input temporally related to EEG tracings / recordings.

[0242] One optional aspect of the disclosure herein is related to methods of manufacturing a wearable sensing system, which may optionally comprise overmolding a flexible housing at least partially around any of the wearable systems components herein, such as those shown in FIGS. 47A-47D. The electrodes are partially embedded in the overmolded housing such that surfaces of each of the plurality of electrodes are exposed to facilitate their placement against tissue to facilitate sensing brain activity signals. Forming the flexible overmold may comprise forming a curved tissue side of the overmold, and wherein the plurality of electrodes may be proud of the curved tissue side, or they may be flush or substantially flush with the curved tissue side. The method may include securing an adhesive to a tissue side of the flexible overmold.

[0243] Any of the systems herein may be adapted to provide additional functionality and benefits outside of prediction glucose states. For example, the following aspects may be incorporated into any of the devices, systems and methods of use herein: Devicecounters / diagnostics to provide info to the users beyond the EEG. Exemplary diagnostics include, without limitation, counters of one or more of the following occurrences: detection of noise (continuous high frequency / high amplitude signal), which can help troubleshoot EEG signals; detection of prolonged increase in overall power and area under the curve, which may, for example, indicate a neurological event such as a seizure; detection of prolonged (e.g., > 60 sec) lack of EEG signal, which may, for example, indicate that a wearable sensor has come loose or off; detection of high impedance, which may, for example, indicate that the sensor may be loose or electrodes not touching skin; detection of various sleep stages (slow wave — 2-4Hz, high-amplitude; high (>12 Hz) to low (<12 Hz) frequency power ratio); sudden high slope decreases or increases in band-specific or broadband power that may indicate the onset of a stressful event; the anticipation of a highly appetitive or insulin-demanding food intake; or onset of exercise / anticipation of exercise. In any of the examples herein, detecting any of these occurrences may result in an alert (of any type) to the patient or user to indicate that, for example only, the sensor is malfunctioning, needs to be placed differently or a large / significant neurological event has occurred. The alert may also provide an overall indicator of sleep quality after a night’s sleep. The alerts may be communicated to the user via an App (on a smartphone) and / or via the wearable sensor (e.g., a vibratory alert if not positioned satisfactorily).

[0244] Any of the Apps herein may be adapted to initiate an output to the user to indicate a status of the wearable sensing system. For example, the status may comprise a functional characterization of the wearable sensing system, such as if the wearable sensing system is functioning (e.g. green light indicator on a display screen controlled by App) or not (e.g. red light indicator on a display screen controlled by App). The status may comprise a position indicator that indicates if the wearable sensing system is in a satisfactory position on the subject (or otherwise indicates that the position is appropriate for its use to sense brain activity signals). A position indicator may indicate if the wearable sensing system is in a satisfactory position on the subject to facilitate satisfactory EEG sensing for any of the methods or uses herein (e.g., real-time glucose monitoring and / or predictive glucose forecasting).

[0245] Any of the bidirectional communication herein optionally includes adaption so the wearable sensing system can send alerts to the patient or user (e.g., via an App) and the patient or user can also communicate marks (or other indicators) back to the wearable sensing system from the App, which marks an event on the EEG tracings / recordings.

[0246] Any of the wearable sensing systems herein may include preferential storage of events of interest if / when they are detected. Once an event is detected, EEG may be stored for anepoch of time. For example only, if the system (wearable and / or App) detects an event of prolonged overall power, a minute EEG clip may be stored (or for some other time period). A storage buffer of about 30 seconds or other time period may facilitate the preferential storage functionality, similar to a digital video recorder.

[0247] In some embodiments, the sensing systems herein may also include additional sensors adapted to sense other biosignals (as illustrated in FIG. 41), such as, without limitation, skin temperature, heart rate, heart rate variability, movement (e.g. accelerometer), peripheral oxygen saturation (spO2), gastric motility (EGG with abdomen sensor), and galvanic skin response (stress measurement), step counters, chewing counters, fall counters, convulsion counters. The sensing systems herein may also include additional sensors such as one or more of optical sensors (e.g., photoplethysmography sensors), inertial sensors (e.g., accelerometer / gyroscope / magnetometer), or electrodermal activity sensors (e.g. for skin conductance; galvanic response).

[0248] As mentioned above, systems herein may be adapted to detect or sense chewing, data for which may be sensed by the wearable sensor and communicated to a user device (e.g., a chew counter). It is known that glucose levels rise after a meal (post-prandial), and thus the precise time or timing of chewing can be used as input to any of the Apps herein to help predict an increase in BG levels. Wearable sensors adapted to be worn in or around the ear are close to the jaw (examples of which are described herein) and are thus particularly well- placed for sensing chewing. Chew sensors can be adapted to distinguish chewing from talking or teeth grinding, which are not indicators of consumption. For example, chewing may be distinguished from talking due to the relative lack of closure during talking (e.g., teeth not making contact with each other). Teeth grinding may be distinguished due to, for example only, a lack of sufficient repetitive opening and closing movements, which may occur more frequently with chewing.

[0249] Any of the wearable sensing systems herein may be adapted to measure impedance values within the sensing systems.

[0250] Any of the wearable sensors herein may also include a vibratory element that is, upon activation, adapted to cause vibrations, which can optionally be part of an alert system to the wearable.

[0251] In some embodiments herein, the systems herein may also be adapted for sleep metric measurements (e.g., duration, sleep stage measurement, etc.).

[0252] Aspects of this disclosure are related to predicting future glucose states or future glycemic events (or risks thereof). Predicting future glucose states may include predicting future blood glucose levels or risk indicators at some time in the future, such as within anhour (e.g., 2, minutes, 15 minutes, 30 minutes, 45 minutes), one or more hours, such as 1 hour, 2 hours, 3 hours, 4 hours, 5 hours, 6 hours, 7 hours, 8 hours, 9 hours, 10 hours, or more, or any range of time from 30 seconds to 10 or more in the future.

[0253] Any of the methods and systems herein that are adapted to predict a glucose state can also be considered to be configured to predict a glycemic event. For example, if a system herein is stored with thresholds for glucose values, and the system herein is configured to send an alert if a threshold is near or reached, the system or method herein is considered to be predicting a glucose state. A glycemic event as used herein may refer generally to user glucose information, and / or a glucose level that is above or below certain levels, such as above or below standard glucose levels, or other patient selected levels.

[0254] Methods and systems herein may optionally be adapted to provide one or more generalized risk levels or risk indicators for entering or having a future glycemic state within one or more certain periods of time in the future. For example only, a predicted risk level may include a plurality of risk indicators, such as low risk, medium risk, and high risk. Optional risk indicators may optionally be visually presented on a display of a device (examples provided herein, such as a smartphone or smartwatch), such as a green light on a display for time in the future where risk is low, a red light on the display for time in the future where risk is high, and a yellow light on the display for time in the future where risk is medium.

[0255] While some parts of the disclosure herein may be related to predicting a future glucose state, such as levels or risk indicators, an aspect of the disclosure herein is optionally related to predicting real-time glucose states. The phrase “real-time” as used herein refers to actual blood levels that are occurring in the subject’s blood at that time. The systems and methods herein may thus optionally be adapted to predict real-time or future glucose states (such as levels), or information indicative thereof. The systems herein may thus also be considered to be continuous glucose monitors, similar to existing monitors.

[0256] Methods and systems herein may be configured to predict a glucose state. Methods and systems herein may also optionally be configured to manage one or more aspects of the predicted state, including optionally preventing a forecasted glucose state or event from occurring. Some aspects of this disclosure are therefore related to the management of blood glucose levels, wherein the term management as used herein may include one or more variety of steps or actions that provide a benefit to the subject.

[0257] A “forecasted” glucose state (or derivative thereof) as used herein refers to a predicted future glucose state as that phrase is used herein.

[0258] Management may include providing information to a subject via any of the Apps herein, such as with an App output and / or alert. For example, methods and system may be adapted to communicate insulin and / or medication needs to a user based on predicted glucose states. Additionally, for example only, automated meal and exercise recommendations, examples of which are shown in FIG. 31 may be part of management herein.

[0259] In some non-limiting examples, management of blood glucose levels includes preventing a glycemic event, such as a hypoglycemic event or a hyperglycemic event (or at least reducing the likelihood thereof). In some examples, management of blood glucose levels optionally also includes ceasing or minimizing the severity of an existing glycemic event. Methods of management herein may optionally include guiding medication and insulin (long / short acting) dosage based on a predicted glucose state (real-time or future).

[0260] It is understood that aspects of this disclosure that are related to predicting glucose levels may or may not be incorporated with aspects of this disclosure related to management of blood glucose levels.

[0261] A merely exemplary method of glucose management includes closed loop functionality, and optionally may include a pump worn on the subject that may be adapted to deliver one or more agents (e.g., glucagon, insulin, etc.), which are currently worn by some patients and used with existing CGMs. For example, one aspect of the disclosure is a method of controlling future glucose levels in a patient in response to predicting future glucose states (although other patient inputs may be used as part of the controlling process - e.g. HR, skin conductance, or any other biosignal herein). In this exemplary aspect, controlling may include delivering insulin or glucagon to a patient before a glucose level deviates from a desired range / limit. In this aspect, controlling may include delivering insulin to a patient at a time when the glucose level is still in a safe range (e.g., 70 -180, 80-170, etc.). In this aspect, controlling may include maintaining glucose levels within a safe / preferred range. In this aspect, controlling may include delivering an agent (e.g. insulin) before the time at which glucose levels are predicted to deviate from a safe range. In this aspect, controlling may include delivering a particular dose of insulin based on the time the glucose levels are predicted to deviate from a safe range. In this aspect, controlling may include delivering a dose of insulin that is different from the dose that would be administered in response to a real time glucose level monitoring process, such as via a CGM.

[0262] The systems herein may optionally include a multiple-chamber pump, such as a dual chamber pump adapted to deliver insulin and glucagon to manage future glucose levels. The systems herein may be adapted and configured to integrate with Bluetooth-enabled insulin pumps (e.g. Omnipod).

[0263] Any of the systems and methods herein may optionally be adapted to deliver insulin when a hyperglycemic event is predicted (real-time or future). Any of the systems and methods herein may optionally be adapted to deliver glucagon (e.g., intranasally; with a pump, etc.) when a hypoglycemic event is predicted. As an example of Management of hypoglycemia herein, in any of the methods herein, an alert or recommendation to consume, for example, a high sugar drink.

[0264] Any of the systems and methods herein may include one or more of oral medicine delivery, subcutaneous delivery, or delivery via a pump.

[0265] Methods of predicting glucose states herein may optionally be performed on a personal device (or a care-team device), such as a smartphone, tablet, smartwatch, and which may include one or more processors thereon adapted to execute one or more methods / algorithms stored thereon. Any type of executable application may be referred to herein as an “App,” and may be adapted to, in response to receiving or based on raw and / or processed EEG data, including only a subset of sensed data (or information indicative of the raw data, processed data, and / or a subset thereof) that has been sensed from the subject, cause the performance of the glucose state predictions herein. In some alternatives, one or more processing steps may take place within the wearable sensing device (e.g., scalp device or subscalp device), which is an example that methods herein (including portions thereof) may be performed in one or more different devices.

[0266] One or more aspects of a predicted glucose levels may optionally be visually represented or presented on a display of a device, examples of which are shown in FIG. 29, FIG. 30, and FIG. 31 (e.g., smartphone, tablet, smartwatch, electronic ophthalmic device such as a contact lens, or glasses). For example only, an executable application (an “App”) may be adapted to visually present a risk indicator and / or predicted glucose levels for some time period in the future, and which may be updated (continuously or periodically) such that the forecast includes predicted levels for that particular time period in the future (e.g., 1 hour, 2 hours, 3 hours, 4 hours, etc.). Additionally, for example only, an App may be adapted to visually present or indicate a specific time at which a glycemic event is forecasted to occur, examples of which are shown in FIG. 29 with the high and low visual markers, or the App may provide a discrete time for a predicted glucose event (e.g., 4: 17 pm). Additionally, for example only, an App may be adapted to present a timer with a countdown indicating the time remaining before a predicted glycemic event.

[0267] In some examples, the methods herein may optionally be adapted to provide a relatively short-term prediction of future blood glucose values in advance (such as 1 hour in advance, 2 hours in advance, 3 hours in advance, etc.), and a longer-term “risk forecast” ofblood glucose hours in advance (e.g., such as 10 hours in advance, 11 hours in advance, 12 hours in advance, etc.). Any of the methods herein (e.g., an App on a personal device) may optionally be adapted to communicate an actual and / or forecast of future glucose levels and / or risk indicators to a patient / care team, optionally to one or more different devices (an example of which is described herein as a Portal, examples of which are shown in FIG. 21 A and 21B). Any of the methods herein (e.g., an App on a personal device) may optionally be adapted to provide a user / patient with a relatively longer (e.g., 10+ hours) forecast of risk indicators, and optionally may be adapted to provide suggested (e.g., optimal) times of day to perform certain activities, such as exercise, eat, take medication based on the forecast (examples of which are shown in FIG. 31). Any of the methods herein (e.g., an App on a personal device) may optionally be adapted to visually present (e.g., plot) an amount (e.g., percent) of time in optimal / preferred glucose range and time spent out of the optimal / preferred range, wherein the range may be adjustable and / or personalized, optionally wherein the method (e.g. App) is adapted to allow for personalized adjustment and setting of the range via interaction with a display of the personal device.

[0268] An of the Apps herein are optionally adapted to store EEG data (raw and / or processed), and optionally until is transferred to a different device, such as an optional online portal described herein. A subject device may have more data storage, so it may be better suited to store more data, allowing a wearable sensor to have a smaller form factor.

[0269] Any of the Apps on a personal / subject device herein may be adapted to perform any of the methods herein (e.g., executable methods, such as forecasting).

[0270] One aspect of the disclosure is a computer executable method that is adapted to present interactive user input features (e.g., icons on a screen, up / down arrows, audible instructions, etc.) that allow a user (patient and / or care team member(s) to adjust a range of desired glucose levels. Any other aspect of the disclosure herein may be incorporated with this aspect.

[0271] A glycemic event as that used herein is an example of a glucose state.

[0272] Any of the methods and systems herein may optionally be adapted to provide or initiate an alert upon the occurrence of an event for which the system is adapted to initiate and / or provide an alert. For example, systems and methods herein may provide or initiate an alert if an existing glycemic event has been detected and / or when a future glycemic event has been forecast. For example, once a future glycemic event has been forecast any of the alerts herein may be triggered (audible alert, text alert, email, vibratory via the wearable sensor, alert to patient and / or car egiver / care team, etc.). Any of the methods herein (e.g., an App on a personal device) may optionally be adapted to provide or initiate an alert when glucose istrending out of range and / or is currently out of range. Any of the methods herein (e.g., an App on a personal device) may optionally be adapted to provide or initiate a relatively high frequency pitch (47-65 Hz) alert when glucose is trending out of range, which may alert a service animal to the forecast or detected glycemic event. One aspect of this disclosure is an executable method that is adapted to provide or initiate a relatively high frequency pitch (e.g. 47-65 Hz) alert when a medical event has been detected and / or forecast, such as a glucose level trending out of range, a seizure is detected or predicted, or loss of consciousness predicted, etc. In some examples, non-audible methods (e.g. low amplitude vibration (e.g. with piezo)) may be preferred, such as in public settings. Some examples may include bone conduction of an audible signal.

[0273] One or more aspects of the actual and / or forecasted blood glucose levels (or information indicative thereof or related thereto) may optionally be communicated, optionally to the subject and / or third party (e.g., caregiver, family member, examples of which are shown in FIGS. 21 A and 21B). One or more aspects of the actual and / or forecasted blood glucose levels may optionally be communicated to one or more devices, which may be the same as or different than the device that predicts the actual and / or forecast blood glucose information.

[0274] The methods and systems herein may optionally be adapted to continuously stream real-time EEG data to a different device, such as any of the personal devices herein (smartphone, watch, etc.). Any of the apps herein may thus be receiving continuous or near- continuous real-time EEG data that is being sensed from the patient, and using the continuously streamed real-time EEG data to make the predictions.

[0275] The methods and systems herein may optionally be adapted to receive one or more non-EEG patient biosignals (parameters), such as any of the additional optional biosignals herein (e.g., skin conductance, heart rate, HRV, blood pressure, etc.), any of which may be additional inputs used at least partially by and with the brain signals to predict the predicted glucose states.

[0276] Any of the methods herein (e.g., an App on a personal device) may optionally also be adapted to receive one or more inputs, optionally from one or more of a subject or caregiver. In this example the inputs are different that biosignals sensed by one or more sensors on a wearable. For example, any of the methods herein (e.g., an App on a personal device) may optionally be adapted to allow for user-input of information such as insulin administration / dose, intake and / or timing of medication, food related information (when, what), exercise, stress, sleep, illness, or the intake / timing of drugs / sub stances that may affect blood sugar (e.g., any of the substances / medicines listed athttps: / / www.trihealth.com / institutes-and-services / diabetes / living-w- diabetes / medications / medications-that-affect-blood-sugar), for example.

[0277] Any of the methods herein (e.g., an App on a personal device) may optionally be adapted to allow a user (patient and / or caregiver) to mark or otherwise indicate events on raw EEG tracings (or processed EEG data), such as provide an indication when a meal was consumed or the time of exercise.

[0278] Part of this disclosure includes methods that are adapted to use one or more inputs to refine or train (e.g., increase the accuracy thereof) any of the forecasting algorithms herein, and optionally the inputs may alternatively or in addition to refine or train any of the CGMs herein. For example, and without limitation, inputs include, without limitation, inputs that are input manually via a user and / or inputs received from other devices (e.g., pumps, glucose meters, and / or CGMs). For example, Apps herein may be adapted to receive information from a glucose meter and / or CGM (e.g., blood glucose readings / values) and use that information to refine a forecasting algorithm. A user may optionally input blood glucose values. Inputs in this context may further include, without limitation, what food was consumed and / or when they consumed the food; when insulin was administered and / or a dose, or any other input that is related to the life of the user that may be able to refine a forecasting method.

[0279] As mentioned above, there may be benefits to predicting glucose states for patients in an ICU setting, who have their blood glucose levels measured periodically, and typically with a finger prick and tested on a glucose meter, a process which may take 10 minutes to complete. Each result is thus roughly 10 minutes delayed. Being able to predict blood glucose levels could be incredibly helpful in an ICU setting, both from a patient care as well as hospital resource management perspective. ICU patients typically already have EEG electrodes attached, and thus brain activity data can be sensed and analyzed to forecast (or help forecast) future blood glucose levels. Any of the methods and systems described herein may thus advantageously be used in an ICU setting. Currently, insulin drips are typically provided to ICU patients, with a sliding scale of dosage as part of the current care. Management of glucose states could thus be incorporated into the ICU as well, concepts of which are described herein.

[0280] Additionally, the systems herein may be beneficial to individuals who are prediabetic and may want to benefit from being informed of their future predicted glucose levels or existing glucose levels. Additionally still, the concepts herein may be useful for subject’s who may otherwise be considered healthy, and who simply want more information about their body, their health, and their overall well-being. For example, without limitation, athletesmay benefit from wearing a brain activity signals sensor at one more discrete locations on the scalp and have an App that can inform them when it would be a good time to ingest carbohydrates if future blood glucose levels are predicted to start to drop just prior to the start of an athletic event, for example. Because the App may be adapted to predict future glucose levels, they may know 1-2 hours before an athletic event when an optimal or preferred time might be to ingest one or more certain types of food. An additional exemplary scenario is to check the App just prior to the start of the athletic event to check if a predicted low glucose level might occur during the event, and in that case could drink a sugary drink in advance of the low level, including at an optimal time. Additionally, since the Apps may be adapted to predict future levels, the athlete could take the wearable sensor off before the game and already be armed with the knowledge about the future glucose levels with the App by the time the athletic event begins. For example, the subject may remove the sensor, but the App can be configured to track previous predicted states, and continue to show the time when predicted levels are predicted to occur.

[0281] In this regard, one aspect of the disclosure is an App that can utilize extra-cranially sensed brain activity signals and provide information indicative of a subject’s predicted glucose levels or risk levels when the subject is not even wearing a wearable brain activity signal sensor, and when they are optionally not even wearing any type of glucose monitor. Any other feature in this disclosure may be incorporated into this aspect.

[0282] Systems herein optionally include an online portal (Examples of which are shown in FIGS. 21 A and 21B), which is optionally available to the subject and any other individual approved by the user (e.g. physician, caregiver, care team, family member, etc.). The portal may be adapted to display any of the information or data described herein, including any historical data on blood glucose trends and user insights on blood glucose levels. The portal may optionally be adapted to display raw EEG signals. The online portal may have signal processing and / or analyzing capabilities as well.

[0283] As a mere example, an online portal may comprise any of the following functionality or capabilities: communicates the actual and forecast of future glucose levels to patient / care team; communicates a trend of past glucose measurements; communicates insulin and medication information to user based on real-time and forecasted blood glucose: provides alerts when glucose is trending out of range or is out of range: provides user with long-term (e.g., 10+ hour) forecast of “risk states” to suggest optimal times of day to exercise, eat meals, and take medication based on the glucose forecast; adapted to displays outside insulin pump, glucose monitor, and App information from Bluetooth linked systems; adapted to displays raw EEG tracing, with user-input markers; can display any other related health data.

[0284] As described herein, systems, devices, and methods herein may be adapted to be incorporated and used to some extent with existing glucose monitors, such as CGMs, invasive glucose monitors, and non-invasive optical glucose monitors, and / or their methods of use. For example, existing CGMs may be modified and adapted to incorporate sensed EEG data (for example only, any sensing concepts / methods in U.S. Patent No. 6,572,542 and / or US Patent No. 8,118,741) and / or forecasting concepts herein to improve performance. For example only, CGM sensed data can be analyzed with patient EEG data, and the predictive EEG data can train the glucose data (e.g., ISG data), so that the CGM may then be adapted to use ISG readings to better predict future blood glucose states, exemplary method steps of which are shown in FIG. 49, and which may be combined with any other suitable method step herein. For example, a certain pattern of EEG-trained ISG data (readings well before an impending event) can then be used to predict a future glycemic event. It is thus understood that any existing CGM may be modified and adapted to incorporate any of the features or methods herein. For example, a CGM can be adapted to communicate with an App and make an alert that a subject should prepare to drink a sugary drink in a certain period of time, such as 1 hour in the future, or that a hypoglycemic event is likely to occur 2.5 hours in the future. Additionally, for example, a CGM could be modified to deliver insulin at a time much earlier than with previous technologies, and could deliver longer lasting insulin well in advance of a hyperglycemic event. FIG. 50 illustrates merely exemplary method steps in which sensed ISG levels can be used to manage a future glucose state of a subject.

[0285] FIG. 51 illustrates merely exemplary steps that may be included in a method of training ISG data with EEG data, which may be performed to allow a glucose monitor to sense ISG and 1) forecast information indicative of a future glucose state (e.g., FIG. 49) and / or 2) facilitate the management of a future glucose state (e.g., FIG. 50).

[0286] Additionally, and only for example, any of the EEG data and methods herein may optionally be used to help calibrate and / or recalibrate glucose monitors (e.g., CGMs) (which need recalibrating over time), which could avoid the need to use glucose meters and finger pricks to re-calibrate glucose monitors such as CGMs. FIG. 52 illustrates a merely exemplary method of calibrating or recalibrating a glucose monitor, optionally a CGM, comprising: calibrating or recalibrating a glucose monitor using at least one of EEG data sensed from the subject or information indicative of the EEG data sensed from the subject. Additionally, glucose monitors (e.g., CGMs) and / or glucose meters may similarly be used to calibrate any of the EEG forecasting methods (e.g., algorithms) herein, an example of which is shown in FIG. 53.

[0287] As described in more detail herein, however, it is conceivable that forecasting may occur without sensing any EEG data. For example, if a CGM is trained on sensed EEG data, existing CGMs may optionally be modified to be adapted with a trained model to forecast future glucose levels and / or risk indicators based on the existing process of sensing ISG levels. Additionally and for example, one or more biosignals (with or without EEG data) may be sensed / obtained and analyzed as part of the prediction process, such as, without limitation, heart rate (HR), heart rate variability, skin conductance, blood pressure, body temperature, exercise level, etc.

[0288] One aspect of the disclosure is an optional bi-directional calibration method and / or system, an example of which is shown in FIG. 54. In one example, a bi-directional calibration method may include sensing interstitial glucose of a subject with a glucose monitor, optionally a CGM; sensing EEG signals from one or more subjects, optionally with any of the wearable devices herein; and performing at least one of, and optionally both of: calibrating (or re-calibrating) the glucose monitor using the sensed EEG signals and / or information indicative of the sensed EEG signals; or calibrating a method that is adapted to determine an existing blood glucose level or forecast a future blood glucose level from the sensed EEG signals and / or information indicative of the sensed EEG signals using the sensed ISG or information indicative of the sensed ISG.

[0289] Additionally, as mentioned above, thresholds for CGMs are currently reset (when they need to be reset) at an office visit. Sensing EEG data and utilizing the forecasting aspects herein may even allow for resetting thresholds without requiring an office visit, such as by using an online portal.

[0290] Additionally, CGMs can be used in conjunction with any of the other heath data / biosignals (e.g., user inputs) herein (e.g., heart rate, blood pressure, skin conductance - which can be sensed easily by existing devices such as smartwatches, fitbits, etc.) or any of the wearable brain activity sensors herein to optionally predict future glucose states.

[0291] Exemplary CGMs, features and methods of use of which may be incorporated herein include those by Dexcom (e.g., G6 CGM System), Medtronic (e.g., Guardian™ Connect), Abbott (e.g., any FreeStyle Libre), and the Eversense® E3 CGM. Exemplary Glucose Meters (glucometers), features and methods of use of which may be incorporated herein: LifeScan OneTouch®, Accu-Chek®, and FreeStyle Lite by Abbott. Exemplary Insulin pumps, features and methods of use of which may be incorporated herein: Medtronic Minimed™, Pumps from Tandem, and Omnipod® pumps.

[0292] Any example, method, system, device or other aspect from any of the disclosure herein may be combined with any other portion of this disclosure unless it is expressly stated otherwise herein.

[0293] Even if not specifically indicated, one or more methods or techniques described in this disclosure (e.g. any of the computer executable methods) may be implemented, at least in part, in hardware, software, firmware or any combination thereof. For example, various aspects of the techniques or components may be implemented within one or more processors, including one or more microprocessors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), programmable logic circuitry, or the like, either alone or in any suitable combination. The term “processor” or “processing circuitry” may generally refer to any of the foregoing circuitry, alone or in combination with other circuitry, or any other equivalent circuitry.

[0294] Such hardware, software, or firmware may be implemented within one device or within separate devices to support the various operations and functions described in this disclosure. In addition, any of the described units, modules or components may be implemented together or separately as discrete but interoperable logic devices. Depiction of different features as modules or units is intended to highlight different functional aspects and does not necessarily imply that such modules or units must be realized by separate hardware or software components. Rather, functionality associated with one or more modules or units may be performed by separate hardware or software components, or integrated within common or separate hardware or software components.

[0295] When implemented in software, the functionality ascribed to the systems, devices and techniques described in this disclosure may be embodied as instructions on a computer- readable medium such as random access memory (RAM), read only memory (ROM), nonvolatile RAM (NVRAM), electrically erasable programmable ROM (EEPROM), Flash memory, and the like. The instructions may be executed by a processor to support one or more aspects of the functionality described in this disclosure.

[0296] FIG. 48 is a merely exemplary block diagram of an exemplary data processing system, any part or all of which may be used with any embodiments of any of the inventions or embodiments set forth herein. Also note that one or more of the components shown in FIG. 48 may be optional even if it is not stated as such herein. Note that while FIG. 48 illustrates various components of a computer system, it is not intended to represent any particular architecture or manner of interconnecting the components; as such details are not germane to the present invention. It will also be appreciated that network computers, handheldcomputers, mobile devices, tablets, cell phones and other data processing systems which have fewer components, or perhaps more components may also be used with the present invention.

[0297] As shown in FIG. 48, exemplary computer system 4100, which is a form of a data processing system, includes a bus or interconnect 4102 which is coupled to one or more microprocessors 4103 and a ROM 4107, a volatile 4 IM 3505, and a non-volatile memory 4106. The microprocessor 4103 is coupled to cache memory 4104. The bus 4102 interconnects these various components together and also interconnects these components 4103, 4107, 4105, and 4106 to a display controller and display device 4108, as well as to optional input / output (I / O) devices 4110, which may be mice, keyboards, modems, network interfaces, printers, and other devices.

[0298] Typically, the input / output devices 4110 are coupled to the system through input / output controllers 4109. The volatile RAM 4105 is typically implemented as dynamic RAM (DRAM) which requires power continuously in order to refresh or maintain the data in the memory. The non-volatile memory 4106 is typically a magnetic hard drive, a magnetic optical drive, an optical drive, or a DVD RAM or other type of memory system which maintains data even after power is removed from the system. Typically, the non-volatile memory will also be a random access memory, although this is not required.

[0299] While FIG. 48 shows that the non-volatile memory is a local device coupled directly to the rest of the components in the data processing system, the present invention may utilize a non-volatile memory which is remote from the system; such as, a network storage device which is coupled to the data processing system through a network interface such as a modem or Ethernet interface. The bus 4102 may include one or more buses connected to each other through various bridges, controllers, and / or adapters, as is well-known in the art. In one embodiment, the VO controller 4109 includes a USB (Universal Serial Bus) adapter for controlling USB peripherals. Alternatively, VO controller 4109 may include IEEE-1394 adapter, also known as FireWire adapter, for controlling FireWire devices, SPI (serial peripheral interface), I2C (inter-integrated circuit) or UART (universal asynchronous receiver / transmitter), or any other suitable technology. Wireless communication protocols may include Wi-Fi, Bluetooth, ZigBee, near-field, cellular and other protocols.

[0300] Some portions of the preceding detailed descriptions have been presented in terms of algorithms and symbolic representations of operations on data bits within a computer memory. These algorithmic descriptions and representations are the ways used by those skilled in the data processing arts to most effectively convey the substance of their work to others skilled in the art. An algorithm is here, and generally, conceived to be a self-consi stentsequence of operations leading to a desired result. The operations are those requiring physical manipulations of physical quantities.

[0301] It should be borne in mind, however, that all of these and similar terms are to be associated with the appropriate physical quantities and are merely convenient labels applied to these quantities. Unless specifically stated otherwise as apparent from the above discussion, it is appreciated that throughout the description, discussions utilizing terms such as those set forth in the claims below, refer to the action and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (electronic) quantities within the computer system's registers and memories into other data similarly represented as physical quantities within the computer system memories or registers or other such information storage, transmission or display devices.

[0302] The techniques shown in the figures herein can be implemented using code (e.g., computer executable methods, algorithms) and data stored and executed on one or more electronic devices. Such electronic devices store and communicate (internally and / or with other electronic devices over a network) code and data using computer-readable media, such as non-transitory computer-readable storage media (e.g., magnetic disks; optical disks; random access memory; read only memory; flash memory devices; phase-change memory) and transitory computer-readable transmission media (e.g., electrical, optical, acoustical or other form of propagated signals — such as carrier waves, infrared signals, digital signals).

[0303] The processes or methods depicted in the preceding figures may be performed by processing logic that comprises hardware (e.g. circuitry, dedicated logic, etc.), firmware, software (e.g., embodied on a non-transitory computer readable medium), or a combination of both. Although the processes or methods are described above in terms of some sequential operations, it should be appreciated that some of the operations described may be performed in a different order. Moreover, some operations may be performed in parallel rather than sequentially.

[0304] The disclosure above is, in some examples, related to sensing brain activity using a wearable sensor, examples of which are described with respect to FIGS. 42-45, 47A-47D. While some applications above are related to using the brain activity data to determine or predict real-time (or near real-time) and / or future blood glucose state or other indicators of metabolic health (or other health related information), the disclosure that follows is not necessarily limited to applications related to blood glucose. As established above, the disclosure herein related to glucose levels is an example of an application of the inventive concepts below related to sensing electrical biosignals using wearable sensors, wherein the electrical biosignals may include signals from a variety of sources, and methods of extractingor distinguishing brain activity signals from within the sensed biosignal data (e.g., removing non-brain data to extract the desired brain activity data).

[0305] As stated herein, electrical activity sensed by an electrode on a scalp is not solely representative of brain activity. The signal is a composite of numerous electrical activities originating from various sources, such as brain activity, muscle activity, vasculature activity, eye movements, skin conductance and sweating, and external interference. The electrical activity generated by the brain is very small compared to the overall voltage measured by an electrode. Brain waves, for instance, are typically in the range of microvolts. In contrast, muscle activity can produce signals that are much larger, often drowning out the delicate brain signals if not properly addressed. Given the nature of these mixed signals, it is crucial to process and analyze electrical biosignal data correctly to ensure that the observed activity genuinely represents brain dynamics and not other sources of electrical activity. Techniques like filtering, artifact rejection, and Independent Component Analysis (ICA) are often employed to separate and remove non-brain-related signals from traditionally acquired EEG data. A multi-channel EEG refers to an EEG recording setup that uses multiple electrodes placed on the scalp to capture electrical brain activity from various regions of the brain simultaneously. Independent Component Analysis (ICA) is a computational method used to separate a multivariate signal into its additive, independent components. Essentially, ICA tries to "unmix" signals that have been combined together. When applied to EEG data, for instance, it can be used to separate out and identify various sources of activity or noise, such as eye blinks, muscle artifacts, or brain activity.

[0306] ICA is widely used as a denoising technique in multi-channel EEG. However, ICA cannot be used with a single channel for the several reasons, and may not be adequate in other non-traditional EEG applications. For example, ICA operates under the assumption that the observed signals (channels in EEG) are linear mixtures of some unknown independent sources. By using multiple channels, ICA can estimate and recover these independent sources. If there is only one channel, there is nothing for ICA to compare it against, making it impossible to deduce the original independent components. Additionally for example, single channel in essence represents a mix of all the sources of electrical activity. With only one channel, the approach would be similar to trying to determine the ingredients of a smoothie based on a single sip; there simply is not enough information. Additionally, for example, ICA requires the number of observations (channels in the case of EEG) to be equal to or greater than the number of sources one is trying to separate. With only one channel, there is essentially only one source, which contradicts the premise of using ICA in the first place.While ICA is not the only denoising technique used in EEG, other techniques also require several channels to be applied properly.

[0307] Methods herein are related to computational, or mathematical approaches to distinguish brain activity data from non-brain activity data from within sensed electrical biosignal data that may include electrical signals from a plurality of different physiological sources. The sensed biosignals may be sensed from one or more wearable scalp sensors, any one of which is optionally, but not limited to, a single channel sensor.

[0308] FIG. 55 represents a general method of distinguishing brain activity data from nonbrain activity data sensed with one or more wearable scalp sensors, which may be included with any other aspect of the disclosure herein, and which may be performed on or with any one or more of the devices or systems described herein.

[0309] FIG. 56 represents a method that includes sensing electrical biosignals with one or more wearable sensor disposed on a scalp of a subject, and based on the sensed biosignals, distinguishing brain activity data from non-brain activity data. The “distinguishing” step in FIG. 56 may be replaced with a step that is considered to “extract” brain activity data from the sensed electrical biosignal data.

[0310] The examples herein include computational approaches to distinguishing or extracting brain activity data. FIG. 57 represents a method that comprises receiving data sensed from one or more wearable scalp sensors, and performing one or more mathematical computations to at least one of distinguish brain activity data from other non-brain activity data, or extract brain activity data from the sensed data.

[0311] FIG. 58 represents a method of one aspect of the disclosure and includes, receiving or generating biosignals sensed from a subject with one or more wearable scalp sensors, wherein the biosignals include electrical biosignals sensed from a plurality of different physiological sources; determining a plurality of features for each of a plurality of epochs of time; decomposing the plurality of features for each of the plurality of epochs of time into constituent components; and distinguishing brain activity data in the sensed electrical biosignals from non-brain activity data in the sensed electrical biosignals. In FIG. 58, the non-brain activity data refers to data from sources of electrical signal other than the brain, and may include one or more of muscle activity data, vasculature activity data, eye movement data, or skin conductance and sweating data. For example, in FIG. 58, the distinguishing step may include, for example without limitation, distinguishing brain activity data from muscle activity data, or it may include distinguishing brain activity data from all data related to muscle activity data, vasculature activity data, eye movement data, and skin conductance and sweating data.

[0312] FIG. 59 represents a method of one aspect of the disclosure and includes, receiving or generating biosignals sensed from a subject with one or more wearable scalp sensors, wherein the biosignals include electrical biosignals sensed from a plurality of different physiological sources; determining a plurality of features for each of a plurality of epochs of time; decomposing the plurality of features for each of the plurality of epochs of time into constituent components; and distinguishing one or more types of electrical biosignal data in the sensed electrical biosignals from one or more different types of electrical biosignal data from within the sensed electrical biosignals. In FIG. 59, the type of electrical biosignal data refers to different sources of electrical signal, and may include brain activity data, muscle activity data, vasculature activity data, eye movement data, and skin conductance and sweating data. For example, in FIG. 59, the distinguishing step may include, for example without limitation, distinguishing muscle activity data from eye movement data, or distinguishing brain activity data from muscle activity data. FIGS. 58 and 59 may be included with any other aspect of the disclosure herein, and may be performed on or with any one or more of the devices or systems described herein.

[0313] FIGS. 60 and 61 represent methods of aspect of the disclosure and include receiving or generating biosignals sensed from a subject with one or more wearable scalp sensors, wherein the biosignals include electrical biosignals sensed from a plurality of different physiological sources; processing the electrical biosignals; and decomposing one or more aspect of the processed biosignals. The method in FIG. 60 includes distinguishing brain activity data in the sensed electrical biosignals from non-brain activity data in the sensed electrical biosignals, which is also included in the method of FIG. 58. The method in FIG. 61 includes distinguishing one or more types of electrical biosignal data in the sensed electrical biosignals from one or more different types of electrical biosignal data from within the sensed electrical biosignals, which is also included in the method in FIG. 59. FIGS. 60 and 61 may be included with any other aspect of the disclosure herein, and may be performed on or with any one or more of the devices or systems described herein.

[0314] The examples and descriptions that follow are merely representative of one or more steps in one or more of the methods in FIGS. 55-61, and the innovative aspects herein are not limited to these examples. Creating machine-learning algorithms to predict a brain state or other physiological states includes creating or building a set of features based on the sensed electrical biosignals, which may optionally include relying on the signal’s amplitude and power spectral density. Power spectral density is conventionally split into four different categories, based on the frequencies. These “frequency bands” are conventionally divided into delta (5), theta (9), alpha (a), beta (P), and gamma (y). Delta, with a frequency range of0.5-4 Hz, is associated with deep sleep, brain injuries, and overall brain function. It is most prominent during dreamless sleep and in deeper, more restorative stages of sleep. Theta, with a frequency range generally of 4-8 Hz, is associated with drowsiness, early stages of sleep, and certain meditative states. It is also linked to creativity, intuition, and daydreaming. Alpha, with a general frequency range of 8-13 Hz, is associated with relaxation, calmness, and a relaxed alertness. It is prominent when the eyes are closed and during relaxation. Beta, with a general frequency range of 13-30 Hz (sometimes subdivided into Low Beta and High Beta) is associated with active, analytical thought processes, alertness, and concentration. It is dominant when we are awake, alert, and engaged in mental activities. Gamma, with a general frequency range of 30-100 Hz (though often, 30-40 Hz is the primary focus), is associated with higher cognitive processing, perception, problem-solving, and recognition. It is believed to be associated with information binding and consciousness.

[0315] Example. The following example of distinguishing or extracting brain activity data was performed with an illustrative single channel wearable scalp sensor. Electrical biosignals were sensed from subjects with one or more wearable scalp sensors, wherein the biosignals included sensed electrical biosignals from a plurality of different physiological sources. The signals were processed with the following exemplary steps, which individually or in any combination are an example of the “processing” step in the methods of FIGS. 60 and 61 : remove power frequencies (60 Hz in the US, 50 Hz in Europe); filter out frequencies outside of 1-100 Hz; select or determine an epoch duration (e.g., 10 seconds); extract raw amplitude for each epoch; and compute the power of each band (using power spectral density, for example), the amplitude, entropy, the Hjorth mobility, and the Hjorth complexity for each epoch. It is of note that these exemplary processing steps above may be used in traditional EEG analysis.

[0316] This Example also included determining a plurality of features (variables), and in this case a plurality of features for each of a plurality of epochs of time, and which is an examples of the “determining a plurality of features” steps in the methods shown in FIGS. 58 and 59. The features were determined by computation or mathematical approaches, and it is understood that the features described herein are exemplary, and other features may be calculated and used in the methods herein. In this example, four moments (described below) were computed of each signal over each epoch. There was four x (1 amplitude +5 band powers +1 entropy +1 mobility +1 complexity), for a total of 36 features (variables) per epoch of time in this example. Again, other features may be calculated, and these are merely exemplary.

[0317] Methods herein include decomposing one or more aspect of the processed biosignals, as shown in the methods in FIGS. 60 and 61. In this example, the method included decomposing the plurality of features (for each of the plurality of epochs of time) into constituent components, which is shown as a step in each of the methods of FIGS. 58 and 59. In this particular example, the empirical mode decomposition (EMD) process / algorithm was used, and was applied over the thirty-six features (variables) taken over all the epochs, with each variable leading to 10 intrinsic mode functions (on average, there could be fewer or more intrinsic mode functions), leading to 360 features (variables) per epoch of time (approximately, the number of intrinsic mode functions cannot be predicted).

[0318] In this particular example, the four statistical moments used are defined as follows: First Moment: this is the mean of the distribution, p = E(X) where E denotes the expected value and X the variable; Second Moment (Variance): When centered around the mean, this represents the variance of the distribution. g2 = E[(X - E(X))2)] = Var(X); Third Moment (Skewness): when appropriately normalized, it provides a measure of the asymmetry or skewness of the distribution. Positive skewness means the distribution has a longer right tail, and negative skewness means it has a longer left tail. g3 = E[(X - E(X))3)]; Fourth Moment (Kurtosis): when appropriately normalized, it describes the "tailedness" of the distribution. Kurtosis greater than 3 (for a distribution normalized to have variance 1) indicates a distribution with heavier tails than the normal distribution, while a kurtosis less than 3 indicates lighter tails. g4 = E[(X - E(X))4)].

[0319] Empirical Mode Decomposition (EMD) is a method used to decompose a signal into its constituent intrinsic mode functions (IMFs). It was introduced by Norden E. Huang in 1998, primarily for analyzing nonlinear and non-stationary signals, common characteristics in many real-world datasets. The idea behind EMD is to represent signals as a sum of oscillatory functions, each with a distinct time scale.

[0320] Based on the decomposition, and in this example with the approximately 360 features, one or more types of electrical biosignal data in the sensed electrical biosignals can be distinguished from one or more different types of electrical biosignal data from within the sensed electrical biosignals (FIGS. 59 and 61), such as distinguishing brain activity data in the sensed electrical biosignals from non-brain activity data in the sensed electrical biosignals (FIGS. 58 and 60). By way of illustration, this Example illustrates how methods herein can differentiate if a wearable sensor (in this case a single-channel electrical sensor) is localized on the scalp or the neck of a subject, using an unsupervised clustering algorithm called UMAP. UMAP, which stands for Uniform Manifold Approximation and Projection, is a dimensionality reduction technique that can be used for visualization similarly to t-SNE (t-Distributed Stochastic Neighbor Embedding), but it can also be used more broadly as a general -purpose dimension reduction technique. UMAP is based on a mathematical framework rooted in Riemannian geometry and algebraic topology. UMAP aims to capture both the local and global structure of the data in the reduced-dimensional representation. While the use of UMAP herein is primarily to visually illustrate the differentiation between types of electrical signal data sensed from a subject, the concepts can be implemented into any method, system and / or device herein as part of a process to perform any of the distinguishing or extracting steps herein (such as implemented on an App stored on a subject personal device, or other computer device).

[0321] After processing raw EEG data following the exemplary steps described above, the variables were normalized and fed (input) into a UMAP algorithm. FIGS. 63A and 63B represent the output of the UMAP algorithm of two sets of data. In red, the datapoints extracted from a single channel device worn on the neck (i.e., not measuring brain activity) and in blue the datapoints extracted from a single channel device worn on the scalp at a behind the ear location as described herein (i.e., measuring brain activity). The device used in both positions was the same. In FIG. 63 A, without implementing the exemplary inventive process described herein (e.g., without calculating a plurality of features from sensed data and decomposing the features into constituent components), the UMAP algorithm was not able to differentiate clearly if the device was placed on the neck or the scalp. As shown FIG. 46B, however, when implementing the exemplary inventive process described herein (e.g., including any one or more of the methods in FIGS. 55-62), the UMAP algorithm creates separate “clusters” for the “neck” and the “scalp” datasets, illustrating an example of an algorithm that can computationally distinguish brain activity data and non-brain activity data (in this case representative of non-brain activity sensed on a neck location).

[0322] FIGS. 64 A and 64B illustrate the results from the same experiment as described above with respect to FIGS. 63 A and 63B, but conducted with another device on a different user. FIG. 64A illustrates a method without implementing the inventive steps herein, and FIG. 64B illustrates a method when implementing the exemplary inventive process, again visually illustrating how an unsupervised UMAP algorithm can computationally distinguish brain activity data and non-brain activity data (in this case representative of non-brain activity sensed on a neck location).

[0323] In a merely exemplary method of use, the method may further include, outputting the distinguished or extracted data, inputting the distinguished brain activity data or information indicative thereof into a prediction model trained to predict a glucose state of the subject based on the distinguished brain activity data, wherein at least some of the non-brain activitydata is not input into the prediction model, predicting the glucose state based on the distinguished brain activity data, and outputting to a subject device information indicative of the predicted glucose state. Any other method or step herein related to glucose state prediction can understandably be incorporated with this or other similar exemplary applications.

Claims

CLAIMSWhat is claimed:

1. A computational method stored on one or more devices, executable by one or more processors, for distinguishing brain activity data from non-brain activity data from within sensed electrical biosignals that are sensed from a subject using one or more wearable scalp sensors, the method comprising the steps of: receiving or generating electrical biosignals (“biosignals”) sensed from a subject with one or more wearable scalp sensors, wherein the sensed electrical biosignals include electrical biosignals from a plurality of different sources; based on the sensed electrical biosignals, computing a plurality of features for each of a plurality of epochs of time; decomposing the plurality of features for each of the plurality of epochs of time into constituent components; and based on the decomposing step, distinguishing brain activity data in the sensed electrical biosignals from non-brain activity data in the sensed electrical biosignals.

2. The method of Claim 1, wherein the distinguishing step comprises distinguishing brain activity data from one or more of muscle activity data, vasculature activity data, eye movement data, or skin conductance and / or sweating data.

3. The method of Claim 1, wherein distinguishing the brain activity data from non-brain activity data comprises computationally clustering the brain activity data from the non-brain activity data.

4. The method of Claim 1, wherein computing the plurality of features comprises computing one or more statistical moments for each of one or more computed characteristics of the biosignals in each of the plurality of epochs of time.

5. The method of Claim 4, wherein the one or more of statistical moments comprise one or more of a mean, a variance, a skewness, or a kurtosis.

6. The method of Claim 4, wherein the one or more characteristics of the biosignals comprises one or more of power, amplitude, entropy, Hjorth mobility, or Hjorth complexity for each epoch.

7. The method of Claim 6, wherein power comprises power of one or more of delta, theta, alpha, beta, or gamma bands.

8. The method of Claim 1, wherein computing the plurality of features comprises computing one or more characteristics of the biosignals in each of the plurality of epochs of time.

9. The method of Claim 8, wherein the one or more characteristics comprise one or more of power, amplitude, entropy, Hjorth mobility, or Hjorth complexity for each epoch.

10. The method of Claim 9, wherein power comprises power of one or more of delta, theta, alpha, beta, or gamma bands.

11. The method of Claim 9, wherein computing the plurality of features further comprises computing one or more statistical moments for each of the one or more of power, amplitude, entropy, Hjorth mobility, or Hjorth complexity.

12. The method of Claim 11, wherein the one or more of statistical moments comprise one or more of a mean, a variance, a skewness, or a kurtosis.

13. The method of Claim 8, wherein computing the plurality of features further comprises computing one or more statistical moments for each of the one or more computed characteristics of the biosignals.

14. The method of Claim 13, wherein the one or more statistical moments comprise one or more of a mean, a variance, a skewness, or a kurtosis.

15. The method of Claim 1, wherein decomposing the plurality of features for each of the plurality of epochs of time into constituent components comprises applying an empirical mode decomposition (EMD) algorithm over the plurality of variables to create a plurality of intrinsic mode functions.

16. The method of Claim 1, wherein distinguishing the brain activity data from non-brain activity data comprises inputting the constituent components into a clustering algorithm.

17. The method of Claim 16, wherein the clustering algorithm is a UMAP algorithm.

18. The method of Claim 1, further comprising outputting the distinguished brain activity data, and optionally visually presenting on a display the brain activity data distinguished from the non-brain activity data.

19. The method of Claim 18, wherein the brain activity data is visually separately clustered from the non-brain activity data.

20. The method of Claim 1, further comprising, inputting the distinguished brain activity data or information indicative thereof into a prediction model trained to predict a glucose state of the subject based on the distinguished brain activity data, wherein at least some of the non-brain activity data is not input into the prediction model; predicting the glucose state based on the distinguished brain activity data; and outputting to a subject device information indicative of the predicted glucose state.

21. A computational method stored on one or more devices, executable by one or more processors, for distinguishing one or more types of electrical biosignal data from one or more different types of electrical biosignal data from within sensed electrical biosignals that are sensed from a subject using one or more wearable scalp sensors, the method comprising the steps of: receiving or generating electrical biosignals (“biosignals”) sensed from a subject with one or more wearable scalp sensors, wherein the sensed electrical biosignals may include electrical biosignals from a plurality of different physiological sources;based on the biosignals, computing a plurality of features for each of a plurality of epochs of time; decomposing the plurality of features for each of the plurality of epochs of time into constituent components; based on the decomposing step, distinguishing one or more types of electrical biosignal data in the sensed electrical biosignals from one or more different types of electrical biosignal data in the sensed electrical biosignals.

22. The method of Claim 21, wherein the distinguishing step comprises distinguishing brain activity data from non-brain activity data.

23. The method of Claim 21, wherein the distinguishing step comprises distinguishing brain activity data from one or more of muscle activity data, vasculature activity data, eye movement data, or skin conductance and / or sweating data.

24. The method of Claim 21, wherein distinguishing the one or more types of electrical biosignal data comprises computationally clustering the one or more types of data from the one or more different types of electrical biosignal data.

25. The method of Claim 21, wherein computing the plurality of features comprises computing one or more of statistical moments for each of one or more computed characteristics of the biosignals in each of the plurality of epochs of time.

26. The method of Claim 25, wherein the one or more of statistical moments comprise one or more of a mean, a variance, a skewness, or a kurtosis.

27. The method of Claim 25, wherein the one or more characteristics of the biosignals comprises one or more of power, amplitude, entropy, Hjorth mobility, or Hjorth complexity for each epoch.

28. The method of Claim 27, wherein power comprises power of one or more of delta, theta, alpha, beta, or gamma bands.

29. The method of Claim 21, wherein computing the plurality of features comprises computing one or more characteristics of the biosignals in each of the plurality of epochs of time.

30. The method of Claim 29, wherein the one or more characteristics comprise one or more of power, amplitude, entropy, Hjorth mobility, or Hjorth complexity for each epoch.

31. The method of Claim 30, wherein power comprises power of one or more of delta, theta, alpha, beta, or gamma bands.

32. The method of Claim 30, wherein computing the plurality of features further comprises computing one or more statistical moments for each of the one or more of power, amplitude, entropy, Hjorth mobility, or Hjorth complexity.

33. The method of Claim 32, wherein the one or more of statistical moments comprise one or more of a mean, a variance, a skewness, or a kurtosis.

34. The method of Claim 29, wherein computing the plurality of features further comprises computing one or more statistical moments for each of the one or more computed characteristics of the biosignals.

35. The method of Claim 34, wherein the one or more statistical moments comprise one or more of a mean, a variance, a skewness, or a kurtosis.

36. The method of Claim 21, wherein decomposing the plurality of features for each of the plurality of epochs of time into constituent components comprises applying an empirical mode decomposition (EMD) algorithm over the plurality of variables to create a plurality of intrinsic mode functions.

37. The method of Claim 21, wherein distinguishing the one or more types of electrical biosignal data from the one or more different types of electrical biosignal data comprises inputting the constituent components into a clustering algorithm.

38. The method of Claim 37, wherein the clustering algorithm is a UMAP algorithm.

39. The method of Claim 21, further comprising outputting the one or more types of electrical biosignal data, and optionally visually presenting on a display the one or more types of electrical biosignal data distinguished from the one or more different types of electrical biosignal data.

40. The method of Claim 39, wherein the one or more types of electrical biosignal data is visually separately clustered from the one or more different types of electrical biosignal data.

41. The method of Claim 21, further comprising, inputting the distinguished one or more types of electrical biosignal data or information indicative thereof into a prediction model trained to predict a glucose state of the subject based on the distinguished data, wherein at least some of the one or more different types of data is not input into the prediction model; predicting the glucose state based on the distinguished one or more types of electrical biosignal data; and outputting to a subject device information indicative of the predicted glucose state.

42. A method for predicting a glucose state of a subject using brain activity data computationally distinguished from non-brain activity data from within sensed electrical biosignals sensed from a subject using one or more wearable scalp sensors, the method comprising the steps of: receiving or generating electrical biosignals (“biosignals”) sensed from a subject with one or more wearable scalp sensors, wherein the electrical biosignals include electrical biosignals sensed from a plurality of different physiological sources; based on the sensed electrical biosignals, computing a plurality of features for each of a plurality of epochs of time; decomposing the plurality of features for each of the plurality of epochs of time into constituent components;based on the decomposing step, distinguishing brain activity data in the sensed electrical biosignals from non-brain activity data in the sensed electrical biosignals; inputting the distinguished brain activity data or information indicative thereof into a prediction model trained to predict a glucose state of the subject based on the distinguished brain activity data; predicting the glucose state based on the distinguished brain activity data; and outputting to a subject device information indicative of the predicted glucose state.

43. The method of Claim 42, further comprising any one or more steps from any one of Claims 1-40.

45. A computational method stored on a device, executable by one or more processors, for extracting brain activity data from electrical biosignals sensed from a subject using one or more wearable scalp sensors, the method comprising the steps of: receiving or generating electrical biosignals (“biosignals”) sensed from a subject with one or more wearable scalp sensors; based on the biosignals, computing a plurality of features for each of a plurality of epochs of time; decomposing the plurality of features for each of the plurality of epochs of time into constituent components; based on the decomposing step, extracting brain activity data; and outputting the extracted brain activity data.

46. The method of Claim 45, further comprising any one or more steps from any of claims 1-43.

47. A method for analyzing electrical signal data from a wearable sensor, optionally a single-channel device, the method comprising the steps of: a) removing power frequencies corresponding to regional standards, wherein said power frequencies are selected from the group consisting of 60Hz and 50Hz; b) filtering out frequencies from the signal data that are outside a predetermined range of 1-100 Hz; c) selecting an epoch duration; d) extracting a raw amplitude for each epoch from the filtered signal data; e) computing power for predetermined bands within the filtered signal data for each epoch, utilizing a power spectral density or a similar computation technique; f) calculating the amplitude entropy for each epoch; g) determining the Hjorth mobility and the Hjorth complexity for each epoch; h) computing the four moments of each signal over each epoch, resulting in an aggregate of 36 variables for each epoch; i) applying the empirical mode decomposition algorithm to the aggregate 36 variables taken over all epochs, wherein each variable yields an average of approximately 10 intrinsic mode functions, resulting in an approximate total of 360 variables; and j) utilizing the approximate 360 variables for each epoch to characterize brain characteristics.

48. The method of claim 47, wherein the predetermined bands used in step e) are based on standard EEG frequency bands, including delta, theta, alpha, beta, and gamma bands.

49. The method of claim 47, wherein the epoch duration selected in step c) is 10 seconds.

50. The method of claim 47, wherein the four moments computed in step h) include the mean, variance, skewness, and kurtosis of each signal over each epoch.

51. The method of claim 47, further comprising the step of: k) feeding the approximate 360 variables into a machine learning algorithm or pattern recognition system to categorize or predict the state or condition represented by the signal data.

52. A method of extracting brain activity data sensed with a wearable single channel EEG sensor from non-brain bodily activity data sensed with the wearable single channel EEG sensor.

53. The method of Claim 52, wherein extracting brain activity comprises performing one or more mathematical computations subsequent to any or all of the steps a, b, c, d, e, f, and g in claim 47 (e.g., performing steps h, i, and j in claim 47 subsequent to, at least the step of, removing power frequencies corresponding to regional standards from the sensed data and subsequent to filtering out frequencies from the sensed data that are outside a range of 1-100 Hz).

54. The method of Claim 52 or Claim 53, further comprising using the extracted brain activity data as input to determine real-time (or near real-time) glucose levels and / or forecast glucose levels, which may include any or all of the disclosure herein related thereto (including any claims, other text, and / or figures).

55. The method of any of Claims 52 - 54, wherein the method is a computer executable method configured to be executed by a processor, and optionally stored on one or more of the wearable single channel EEG sensor, a personal device configured to be communication with the single channel EEG sensor (e.g., part of an App on a computing device like a smartphone), or a different external device (e.g., a care-team computer).

56. The method of any of Claims 52 - 55, wherein the non-brain bodily activity data comprises electrical activity from or resulting from one or more of muscle activity, the vasculature (e.g., from pulsation), eye movement, or skin conductance and / or sweating.

57. A method, comprising: sensing subject data with a single channel EEG sensor, optionally a wearable sensor disposed on a scalp of a subject; and extracting brain activity data from the sensed subject data.

58. The method of Claim 57, wherein extracting brain activity data comprises performing one or more mathematical computations subsequent to any or all of the steps a, b, c, d,e, f, and g in claim 47 (e.g., performing steps h, I, and j in claim 47 subsequent to, at least the step of, removing power frequencies corresponding to regional standards from the sensed data and subsequent to filtering out frequencies from the sensed data that are outside a range of 1-100 Hz).

59. The method of Claim 57 or Claim 58, further comprising using the extracted brain activity data as input to determine real-time (or near real-time) glucose levels and / or forecast glucose levels, which may include any or all of the disclosure herein related thereto (including any claims, other text, and / or figures).

60. The method of any of Claims 57-59, wherein the method is a computer executable method configured to be executed by a processor, and optionally stored on one or more of the single channel EEG sensor, a personal device configured to be communication with the single channel EEG sensor (e.g., part of an App on computing device like a smartphone), or a different external device (e.g., a care-team computer).

61. The method of any of Claims 57-60, wherein extracting the brain activity data comprises distinguishing the brain activity data from electrical activity from or resulting from one or more of muscle activity, the vasculature (e.g., from pulsation), eye movement, or skin conductance and / or sweating.

62. A method, comprising sensing data with a single channel EEG sensor, optionally a wearable sensor disposed on a scalp of a subject; and distinguishing brain activity data in the sensed data from non-brain bodily activity data in the sensed data.

63. The method of Claim 62, wherein the distinguishing step comprises performing one or more mathematical computations subsequent to any or all of the steps a, b, c, d, e, f, and g in claim 47 (e.g., performing steps h, I, and j in claim 47 subsequent to, at least the step of, removing power frequencies corresponding to regional standardsfrom the sensed data and subsequent to filtering out frequencies from the sensed data that are outside a range of 1-100 Hz).

64. The method of Claim 62 or Claim 63, further comprising using the distinguished brain activity data as input to determine real-time (or near real-time) glucose levels and / or forecast glucose levels, which may include any or all of the disclosure herein related thereto (including any claims, other text, and / or figures).

65. The method of any of Claims 62-64, wherein the method is a computer executable method configured to be executed by a processor, and optionally stored on one or more of the single channel EEG sensor, a personal device configured to be communication with the single channel EEG sensor (e.g., part of an App on computing device like a smartphone), or a different external device (e.g., a care-team computer).

66. The method of any of Claims 62-65, wherein the distinguishing step comprises distinguishing the brain activity data from electrical activity from or resulting from one or more of muscle activity, the vasculature (e.g., from pulsation), eye movement, or skin conductance and / or sweating.

67. A method of extracting brain activity data, comprising: with data that has been sensed with a single channel EEG sensor, optionally from a wearable sensor disposed on a scalp of a subject, extracting brain activity data from the sensed data.

68. The method of Claim 31, further comprising one or more of any of the steps from Claims 58 - 61.

69. A method, comprising: with data that has been sensed with a single channel EEG sensor, optionally from a wearable sensor disposed on a scalp of a subject,distinguishing brain activity data in the sensed data from non-brain bodily activity data in the sensed data.

70. The method of Claim 31, further comprising one or more of any of the steps from Claims 63 - 66.

71. A method, compri sing : with data sensed with a single channel EEG sensor, optionally from a wearable sensor disposed on a scalp of a subject, performing one or more mathematical computations to one or both of: distinguish brain activity data in the sensed data from non-brain bodily activity data in the sensed data, or extract brain activity data from the sensed data.

72. Any of the methods of claims 47-71, wherein data is sensed from the subject using any of the devices, systems and / or methods described in U.S. Pat. No. 11,020,035.

73. A system, comprising: one or more processors; a memory coupled to the one or more processors, the memory storing computerprogram instructions, that, when executed by the one or more processors, perform a computer-implemented method comprising: receiving or generating electrical biosignals (“biosignals”) sensed from a subject with one or more wearable scalp sensors, wherein the sensed electrical biosignals include electrical biosignals from a plurality of different sources; based on the sensed electrical biosignals, computing a plurality of features for each of a plurality of epochs of time; decomposing the plurality of features for each of the plurality of epochs of time into constituent components; based on the decomposing step, distinguishing brain activity data in the sensed electrical biosignals from non-brain activity data in the sensed electrical biosignals; and outputting the distinguished one or more types of electrical biosignal data.

74. A system, comprising:one or more processors; a memory coupled to the one or more processors, the memory storing computerprogram instructions, that, when executed by the one or more processors, perform a computer-implemented method comprising: receiving or generating electrical biosignals (“biosignals”) sensed from a subject with one or more wearable scalp sensors, wherein the sensed electrical biosignals may include electrical biosignals from a plurality of different physiological sources; based on the biosignals, computing a plurality of features for each of a plurality of epochs of time; decomposing the plurality of features for each of the plurality of epochs of time into constituent components; based on the decomposing step, distinguishing one or more types of electrical biosignal data in the sensed electrical biosignals from one or more different types of electrical biosignal data in the sensed electrical biosignals; and outputting the distinguished one or more types of electrical biosignal data.

75. A system, comprising: one or more processors; a memory coupled to the one or more processors, the memory storing computerprogram instructions, that, when executed by the one or more processors, perform a computer-implemented method comprising: receiving or generating electrical biosignals (“biosignals”) sensed from a subject with one or more wearable scalp sensors; based on the biosignals, computing a plurality of features for each of a plurality of epochs of time; decomposing the plurality of features for each of the plurality of epochs of time into constituent components; based on the decomposing step, extracting brain activity data; and outputting the extracted brain activity data.

Images