WO2025059709A1 - Thresholding and treatment titration using a forecasting model - Google Patents

Abstract

Systems and methods for monitoring and treating epilepsy involve a detection model configured to detect seizure events in a patient based upon electroencephalogram (EEG) data, a prediction model configured to predict seizure events in the patient, and a titration model configured to adjust therapeutic treatments for the patient. The detection model, forecasting model, and titration model are interlinked via the provision of further data input/outputs among the models, to enable each model to iteratively adapt to the other models to more accuracy detect, predict, and treat epileptic seizure events in the patient.

Description

THRESHOLDING AND TREATMENT TITRATION USING A FORECASTING MODEL

TECHNICAL FIELD

[0001] The present disclosure relates to systems and methods for monitoring of electroencephalographical activity in a patient and, in particular, to detecting and predicting seizures in a patient based upon therapeutic treatment such as pharmacological treatment and/or neurostimulation applied to the patient.

BACKGROUND OF THE DISCLOSURE

[0002] Epilepsy is considered the world’s most common serious brain disorder, with an estimated 50 million sufferers worldwide and 2.4 million new cases occurring each year. Epilepsy is a condition of the brain characterized by epileptic seizures that vary from brief and barely detectable seizures to more conspicuous seizures in which a sufferer vigorously shakes. Epileptic seizures are unprovoked, recurrent, and due to unexplained causes.

[0003] Epilepsy and seizures can be classified in a variety of ways, but generally can be broken down into two categories: generalized onset epilepsy and focal onset epilepsy. In generalized epilepsy, seizures begin with a widespread, excessive electrical discharge involving both hemispheres (sides) of the brain. By contrast, in focal epilepsy, an abnormal electrical discharge occurs in one small part of the brain. Focal seizures are frequently caused by an underlying structural abnormality in the brain, though the abnormality is not always visible to clinicians using standard imaging techniques. Other causes, such as head trauma, stroke, infection, tumors, can also cause focal epilepsy.

[0004] Diagnosing epilepsy has typically required detailed study of clinical observations and of electrical and/or other signals in the patient’s brain and/or body. Particularly with respect to studying electrical activity in the patient’s brain (e.g., using electroencephalography to produce an electroencephalogram (EEG)), such study usually requires the patient to be monitored for some period of time. The monitoring of electrical activity in the brain requires the patient to have a number of electrodes placed on the scalp, each of which electrodes is typically connected to a data acquisition unit that samples the signals continuously (i.e., at a rate exceeding 200 Hz) to record the signals for later analysis. Medical personnel monitor the patient to watch for outward signs of epileptic or other events, and review the recorded electrical activity signals to determine whether an event occurred, whether the event was epileptic in nature and, in some cases, the type of epilepsy and/or region(s) of the brain associated with the event. Because the electrodes are wired to the data acquisition unit, and because medical personnel must monitor the patient for outward clinical signs of epileptic or other events, the patient is typically confined to a small area (e.g., a hospital or clinical monitoring room) during the period of monitoring, which can last anywhere from several hours to several days. Moreover, where the number of electrodes placed on or under the patient’s scalp is significant, the size of the corresponding wire bundle coupling the sensors to the data acquisition unit may be significant, which may generally require the patient to remain generally inactive during the period of monitoring, and may prevent the patient from undertaking normal activities that may be related to the onset of symptoms.

[0005] While ambulatory encephalograms (aEEGs) allow for longer-term monitoring of a patient outside of a clinical setting, aEEGs are typically less reliable than EEGs taken in the clinical setting, because clinical staff do not constantly monitor the patient for outward signs of epileptic events or check if the electrodes remain affixed to the scalp and, as a result, are less reliable when it comes to determining the difference between epileptic and non-epileptic events.

[0006] The use of EEG in the determination of whether an individual has epilepsy, the type of epilepsy, and its location (or foci) in the brain is fundamental in the diagnostic pathway of individuals suspected of epilepsy. Implant devices may provide a solution for long-term recording of EEG while minimally, if at all, affecting the quality of life of the patient in which the device is implanted. The recording of EEG data via implant devices (and/or other EEG recording devices) can be used to inform algorithm-based models to detect and/or predict (“forecast”) seizures, warn or notify the patient of ongoing or future seizures via alerts, and/or drive or adjust neurostimulation, drug regimes, and/or other therapeutic intervention to treat or prevent seizures. Examples of such devices, techniques and models are described in U.S. Patent Application Publication No. 16/797,315, entitled “Electrode Device for Monitoring and/or Stimulating Activity in a Subject,” the entirety of which is hereby incorporated by reference.

[0007] However, while the EEG offers a rich source of information relating to the disease to create and implement the models described above, EEG data alone may form only an incomplete picture of the current and historical state of the patient. That is, while the models above take into account the current and historical EEG data of the patient, the models traditionally do not take into account the historical and ongoing therapies (e.g., neurostimulation and drug intake) administered to the patient. Further, traditional models typically decouple the aspects of seizure forecasting and seizure detection. That is, algorithms for detecting a seizure in a patient within an EEG monitoring window do not traditionally account for any previous- established predictions regarding whether the patient may suffer a seizure during the same window.

[0008] Treatment of epilepsy is an inexact science. For example, the standard of care for an individual with either suspected or diagnosed epilepsy is to administer one or more anti-epileptic drugs (AEDs) in an effort to minimize or eliminate epileptic seizures in the individual, or to administer neurostimulation to the brain or vagal nerve of the patient using a neurostimulator device. Typically, such drugs are administered in oral form and taken regularly (e.g., daily) at a dosage that is determined by the treating physician (e.g., neurologist). The specific dose and administration frequency that is effective for a particular patient is specific to the patient and is generally determined by titrating the dose until a perceived effective dosage and treatment schedule is determined. In other words, titration may be adjusted so as to optimize the protection of the patient against epileptic seizures. Thus, challenges exist in identifying effective dosages and timings for therapeutic treatments to be administered to the patient. Additionally, just as algorithms for detecting seizures have not traditionally been coupled to forecasting outputs for the patient, the algorithms for predicting and detecting seizures have not traditionally been coupled to information indicating previous treatments administered to the patient, and the adjustments made to such treatment regimes.

[0009] In view of the above challenges, the present disclosure identifies that it would be advantageous to allow aspects of seizure forecasting, seizure detection, and therapeutic treatment to interact with each other to obtain greater precision in each aspect.

[0010] Any discussion of documents, acts, materials, devices, articles or the like which has been included in the present background is not to be taken as an admission that any or all of these matters form part of the prior art base or were common general knowledge in the field relevant to the present disclosure as it existed before the priority date of each claim of this application.

SUMMARY OF THE DISCLOSURE

[0011] In embodiments, a computer-implemented method is provided, the method being implemented via one or more processors. The method may include (1) obtaining a forecasting model configured to generate predictions of likelihood of epileptic seizure events in a patient based at least upon first input data indicative of historical electroencephalogram (EEG) signal data of the patient, (2) obtaining first titration data for the patient, the first titration data indicating administration of therapeutic treatment to the patient responsive to an epileptic condition or an epileptic seizure event experienced by the patient, (3) generating second input data for the forecasting model based upon the first titration data, and/or (4) providing the second input data as further input to the forecasting model to cause the forecasting model to generate first prediction data indicating a likelihood of an epileptic seizure event in the patient over a first prediction window based upon the first and second input data. The method may include additional, fewer, and/or alternate actions, including actions described in the present disclosure.

[0012] In other embodiments, a system for predicting seizure events in a patient is provided. The system may include an electrode array including a reference electrode and a plurality of sensing electrodes spaced linearly along a lead of the electrode array and configured, collectively, to measure EEG signals of a brain of the patient. The electrode array may further include a processing unit including (1) a memory device, (2) a processor communicatively coupled to the electrodes and receiving the EEG signals from plurality of sensing electrodes, the processor configured to store to the memory device data of the received EEG signals, and (3) a transceiver coupled to the processor and configured to transmit data to, and receive data from, an external computing device. One or more computer-executable routines executing on the external computing device may be configured to (1) receive the EEG signals as first input data to a forecasting model, (2) obtain first titration data for the patient, the first titration data indicating administration of therapeutic treatment to the patient, and/or (3) provide second input data to the forecasting model based upon the first titration data to cause the forecasting model to generate first prediction data indicating a likelihood of an epileptic seizure event in the patient over a first prediction window based upon the first and second input data. The system may include additional, fewer, and/or alternate components, including various components descried in the present disclosure. Moreover, the system may be configured to perform additional, fewer, and/or alternate actions, including various actions described in the present disclosure.

[0013] In still other embodiments, another computer-implemented method is provided, the method being implemented via one or more processors. The method may include (1) obtaining a seizure event detection model configured to generate detection data indicating epileptic seizure events experienced by a patient based upon first input data indicative of electroencephalogram (EEG) signal data of the patient over a first EEG monitoring window, (2) obtaining first prediction data for the patient, the first prediction data indicating likelihood of one or more epileptic seizure events in the patient over a first prediction window matching at least a portion of the first EEG monitoring window, the first prediction data being generated based upon titration data indicating therapeutic treatment administered to the patient, (3) generating second input data for the detection model based upon the first prediction data, and/or (4) providing the second input data as further input to the detection model to cause the detection model to generate first detection data indicating an epileptic seizure event experienced by the patient over the first EEG monitoring window based upon the first and second input data. The method may include additional, fewer, and/or alternate actions, including actions described in the present disclosure.

[0014] In still yet other embodiments, another system for predicting seizure events in a patient is provided. The system may include an electrode array comprising a reference electrode and a plurality of sensing electrodes spaced linearly along a lead of the electrode array and configured, collectively, to measure EEG signals of a brain of the patient. The electrode array may further include a processing unit including (1) a memory device, (2) processor communicatively coupled to the electrodes and receiving the EEG signals from plurality of sensing electrodes, the processor configured to store to the memory device data of the received EEG signals, and (3) a transceiver coupled to the processor and configured to transmit data to, and receive data from, an external computing device. One or more computer-executable routines executing on the external computing device may be configured to (1) receive the EEG signals as first input to a seizure event detection model configured to generate detection data indicating epileptic seizure events experienced by the patient over a first EEG monitoring window, (2) obtain first prediction data for the patient, first prediction data indicating likelihood of one or more epileptic seizure events in the patient over a first prediction window matching at least a portion of the first EEG monitoring window, and/or (3) provide second input data to the detection model based upon the first prediction data to cause the detection model to generate first detection data indicating an epileptic seizure event experienced by the patient over the first EEG monitoring window based upon the first and second input data. The system may include additional, fewer, and/or alternate components, including various components descried in the present disclosure. Moreover, the system may be configured to perform additional, fewer, and/or alternate actions, including various actions described in the present disclosure.

BRIEF DESCRIPTION OF THE FIGURES

[0015] Fig. 1 is a block diagram of an example system according to the described embodiments.

[0016] Figs. 2A and 2B show side and top views, respectively, of an example electrode device.

[0017] Figs. 2C through 2E show cross-sectional views of portions of the electrode device of Figs. 2A and 2B.

[0018] Figs. 2F and 2G show top and side views, respectively, of a distal end portion of the electrode device of Figs. 2A and 2B.

[0019] Fig. 2H illustrates an example implantation location of electrodes of an electrode device.

[0020] Fig. 2I illustrates an example implantation location of an electrode device.

[0021] Fig. 3 is a block diagram of an example sensor array including a plurality of electrodes and a local processing device.

[0022] Fig. 4A is a block diagram depicting in greater detail an embodiment of an example system implementing a static model.

[0023] Fig. 4B is a block diagram depicting in greater detail an embodiment of an example system implementing a trained Al model.

[0024] Fig. 5 is a block diagram depicting another example embodiment, in which detection and classification of events take place on an external device.

[0025] Fig. 6A is a block diagram of a first part of an example system for use in creating a trained Al model.

[0026] Fig. 6B is a block diagram of a second part of the example system for use in creating a trained Al model. [0027] Fig. 7A is a block diagram depicting a first Al training set of data according to described embodiments.

[0028] Fig. 7B is a block diagram depicting a second Al training set of data according to described embodiments.

[0029] Fig. 8 is a block diagram depicting an embodiment of outputs of a detection model, a forecasting model, and a titration model.

[0030] Fig. 9 illustrates the general concept of a therapeutic treatment window to target using a titration model.

[0031] Fig. 10 is a block diagram depicting example flow of data among a detection model, a forecasting model, and a titration model.

[0032] Fig. 11 A is a block diagram of an example computer-implemented method of a forecasting model.

[0033] Fig. 11 B is a block diagram of an example computer-implemented method of a detection model.

[0034] Fig. 12 is a block diagram of an example embodiment of a processor device coupled to a separate sensor array according to the description.

[0035] Fig. 13 is a block diagram of another example embodiment of a processor device coupled to a combined unit including the sensor array according to the description.

[0036] Fig. 14A illustrates an example communication scheme between the sensor array and the processor device.

[0037] Fig. 14B illustrates an alternate example of a communication scheme between the sensor array and the processor device.

[0038] Fig. 15A illustrates an example communication scheme between the processor device and external equipment.

[0039] Fig. 15B illustrates an alternate example of a communication scheme between the processor device and external equipment.

[0040] Fig. 15C illustrates yet another example of a communication scheme between the processor device and external equipment.

DETAILED DESCRIPTION

[0041] Embodiments of the present disclosure relate to monitoring, classification, and forecasting of seizure activity in a patient based upon electroencephalography (EEG) data of the patient obtained via one or more EEG monitoring devices. More particularly, in certain embodiments of the present disclosure, EEG data of the patient is monitored using an array of sensors disposed beneath the patient’s scalp (or, in some embodiments, in one or more devices worn externally by the patient) to monitor epileptic brain activity, with the array of sensors being communicatively coupled to one or more processing devices configured to monitor and classify the brain activity to determine when seizure activity occurs and, in some instances, the particular locations (or foci) of the seizure activity.

[0042] Embodiments of the present disclosure include a detection model configured to analyze data received from the EEG monitoring device(s) to identify seizure activity in the patient during a monitoring window, and a forecasting model configured to forecast (/.e., predict) seizure activity in the patient over a future prediction window. Still additionally, embodiments of the present disclosure include a titration model configured to determine and adjust quantities and/ortiming of therapeutic treatment (e.g., neurostimulation and/or drug intake) to be administered to the patient to optimize therapeutic effect while avoiding adverse side effects to the patient.

[0043] In embodiments of the present disclosure, aspects of the detection model, forecasting model, and titration model are coupled to each other to allow outputs of certain ones of the models to function as inputs to other ones of the models. Particularly, in embodiments of the present disclosure, outputs of the forecasting model (e.g., indicating likelihood of a future seizure(s)) are provided as further inputs to the detection model, for example to cause the detection model to decrease required confidence thresholds and increase sensitivity of the detection model to potential seizure activity in the patient for time windows wherein a seizure is predicted to be likely to occur. Additionally, in embodiments of the present disclosure, outputs of the titration model indicating therapeutic treatment administered to the patient are provided as further inputs to the forecasting model, for example to bias the forecasting model to predict a lesser likelihood of seizure activity at times shortly following administration of therapy (i.e., therapeutic treatment), or to predict a greater likelihood of seizure activity at times further on from administration of therapy. Still additionally, in embodiments of the present disclosure, outputs of the detection model are provided as inputs to the titration model to provide for better tailored determinations or adjustments of therapeutic treatment quantities and/or schedules based upon likelihood of seizures suffered by the patient. Still yet further, in embodiments, the outputs of the detection model are provided to the forecasting model to validate the forecasting model and/or to generate subsequent predictions of epileptic seizure events in the patient. By allowing the models to interact with each other in these manners, the embodiments of the present disclosure may improve the reliability of each model to thereby produce more reliable seizure prediction, more reliable seizure detection, and/or more appropriate therapeutic schedules to improve outcomes for the patient.

[0044] Throughout the present disclosure, embodiments are described in which various elements are optional - present in some, but not all, embodiments of the system. Where such elements are depicted in the accompanying figures and, specifically, in figures depicting block diagrams, the optional elements are generally depicted in dotted lines to denote their optional nature.

[0045] EXAMPLE SYSTEM

[0046] Fig. 1 depicts, in its simplest form, a block diagram of a contemplated system 100 directed to detection and classification of neurological events. The system 100 includes an EEG sensor array 102, a processor device 104, and a user interface 106. The sensor array 102 generally provides EEG data, in the form of data representing detected electrical signals, to the processor device 104, which receives the signals and uses the signals to detect and classify seizure events using the electrical signal data. In some embodiments that will be described herein, the sensor array comprises one or more sensors implanted underneath the scalp of the patient. Additionally or alternatively, in some embodiments, the sensor array comprises one or more sensors affixed externally to the patient (e.g., using wearable devices). The user interface 106 may facilitate self-reporting by the patient of any of various data including events perceived by the patient, as well as other data (e.g., medication types, doses, dose times, patient mood, potentially relevant environmental data, and the like). The user interface 106 may also facilitate output of outputs of models described herein, programming of the unit for a particular patient, calibration of the sensor array 102, tuning of various parameters such a gain, filtering, montage settings, etc.

[0047] Additional sensors and/or other equipment 108 may also be present in the system 100. Sensors and/or other equipment 108 may include, for example, a photoplethysmography (PPG) sensor, a microphone, a temperature sensor, an accelerometer, and/or another sensor(s) configured to detect a condition or state of the patient. Additionally or alternatively, sensors and/or other equipment 108 may include one or more therapeutic devices that provide therapeutic support to the patient to treat or mitigate one or more conditions of the patient (e.g., epilepsy and/or other conditions emerging from epileptic activity). Such therapeutic devices may include, for example, (1) a drug pump delivers timed, measured doses of a pharmacological agent (i.e., a drug) to the patient, and/or (2) a neurostimulator device (e.g., a vagal nerve stimulation device, a hypoglossal nerve stimulation device, an epicranial and/or transcranial electrical stimulation device, an intracranial electrical stimulation device, a phrenic nerve stimulator, a cardiac pacemaker, etc.) configured to apply or adjust (e.g., amplitude, frequency of the signal, frequency of the stimulus application, etc.) a neurostimulation signal.

[0048] Generally, an EEG biomarker is a specific pattern, signal, or measurement obtained from an electroencephalogram (EEG) that provides information about a certain condition, disease, or physiological state of the patient. EEG biomarkers are often used to aid in the diagnosis, monitoring, and understanding of various neurological disorders, including epilepsy. These biomarkers can be specific waveforms, frequencies, amplitudes, or other characteristics of the EEG that are associated with particular neurological states. One class of EEG biomarkers is an epileptiform event (or epileptiform). An epileptiform event refers to a transient abnormal electrical discharge in the brain that resembles epileptic activity. These events can manifest as spikes, sharp waves, or other distinctive patterns on an EEG recording. Epileptiform events are often used as markers of potential epilepsy or seizure activity. They indicate abnormal electrical activity in the brain, which may or may not lead to actual seizures, and typically are used by clinicians as potential signals of an ictal event, indicative of the onset, duration, and offset of a seizure event. The shorter duration biomarkers within an epileptiform may include (without limitation) “spikes,” having durations between 30 and 80 milliseconds (ms), and “sharps,” having durations between 70 and 200 ms, although these biomarkers may also occur between seizures. The various biomarkers associated with ictal activity may be indicative of the types of seizures occurring and/or the foci of seizure onset for various seizure events. For example, absence seizures are frequently associated with generalized “spike” activity, though spike activity is not exclusive to absence seizures. Features of epileptiforms may signal additional biomarkers, and interictal (between seizure), pre-ictal, and post-ictal EEG data may provide additional biomarker information related to detection and/or prediction of seizures. In embodiments, photoplethysmography (PPG) data may be collected using one or more additional sensors, and may include PPG biomarker data related to interictal, pre-ictal, post-ictal (and ictal) state of the patient. For instance, oxygen desaturation is known to occur in a significant portion of focal seizures, including those without convulsive activity, before, during, or after a seizure. Similarly, changes in blood pressure, heart rate, or heart rate variability - all detectable within PPG data - can occur before, during, or after a seizure event. In some embodiments, by observing EEG data and PPG data concurrently, over periods of time, additional relationships between biomarkers in EEG data and PPG data can reveal relationships and patterns that facilitate the detection and, perhaps more importantly, prediction of ictal events, and, in some embodiments establish biomarkers relating to drug side-effects and quality of life metrics that may relate to the long-term use of the applied therapeutic treatment(s). Other biomarkers may, in embodiments, be detected from microphone and/or accelerometer data. Together, all of the biomarkers collected may, in some embodiments, be analyzed to determine the types, locations (foci), and other characteristics of seizures.

[0049] SENSOR ARRAY

[0050] Figs. 2A through 21 illustrate an embodiment of a sensor array 102 from Fig. 1 , such as that described in U.S. Patent Application No. 16/797,315, entitled “Electrode Device for Monitoring and/or Stimulating Activity in a Subject,” the entirety of which is hereby incorporated by reference herein. With reference to Figs. 2A and 2B, in one embodiment an electrode device 157 is provided comprising an elongate, implantable body 158 and a plurality of electrodes 160 positioned along the implantable body 158 in the length direction of the implantable body 158. At a proximal end of the implantable body 158, a processing unit 144 is provided for processing electrical signals that can be sent to and/or received from the electrodes 160. Though not required, in some embodiments, an electrical amplifier 163 (e.g., a pre-amp) is positioned in the implantable body 158 between the electrodes 160 and the processing unit 144. In an alternative embodiment, the electrical amplifier 163 may be integrated into the processing unit 144 of the electrode device 157, instead of being positioned in the implantable body 158. A reference electrode 160R may also be included and, in particular, may be integrated into or on the surface of a housing of the processing unit 144. The reference electrode 160R may serve as a voltage reference to facilitate elimination of noise and/or better signal-to-noise ratio.

[0051] With reference to Fig. 2C, which shows a cross-section of a portion of the electrode device 157 adjacent one of the electrodes 160, the electrodes 160 are electrically connected, e.g., to the amplifier 163 and processing unit 144, by an electrical connection 167 that extends through the implantable body 158. A reinforcement device 168 is also provided in the electrode device 157, which reinforcement device 168 extends through the implantable body 158 and limits the degree by which the length of the implantable body 158 can extend under tension.

[0052] In this embodiment, referring back to Figs. 2A and 2B, four electrodes 160 are provided that are spaced along the implantable body 158 between the amplifier 163 and a distal tip 159 of the implantable body 158. The distal tip 159 of the implantable body 158 is tapered. The four electrodes 160 are configured into two electrical pairs 161 , 162 of electrodes, the two most distal electrodes 160 providing a first pair of electrodes 161 and the two most proximal electrodes 160 providing a second pair of electrodes 162. In this embodiment, the electrodes 160 of the first pair 161 are spaced from each other at a distance x of about 40 to 60mm, e.g., about 50 mm (measured from center-to-center of the electrodes 160) and the electrodes 160 of the second pair 162 are also spaced from each other at a distance x of about 40 to 60mm, e.g., about 50 mm (measured from center-to-center of the electrodes 160). The first and second electrode pairs 161 , 162 are spaced from each other at a distance y of about 30 to 50 mm, e.g., about 40 mm (measured from center-to-center of the electrodes of the two pairs that are adjacent each other).

[0053] With reference to Figs. 2D and 2E, which provide cross-sectional views along lines B- B and C--C in Fig. 2C, respectively, the implantable body 158 has a round, e.g., substantially circular or ovate, cross-sectional profile. Similarly, each of the electrodes 160 has a round, e.g., substantially circular or ovate, cross-sectional profile. Each of the electrodes 160 extend circumferentially, completely around a portion of the implantable body 158. By configuring the implantable body 158 and electrodes 160 in this manner, the exact orientation of the implantable body 158 and electrodes 160, when implanted in a subject, is less critical. For example, the electrodes 160 may interact electrically with tissue in substantially any direction. In this regard, the electrodes 160 may be considered to have a 360-degree functionality. The round cross- sectional configuration can also provide for easier insertion of the implantable portions of the electrode device 157 to the target location and with less risk of damaging body tissue. For example, the implantable body 158 can be used with insertion cannulas or sleeves and may have no sharp edges that might otherwise cause trauma to tissue.

[0054] In this embodiment, the implantable body 158 is formed of an elastomeric material such as medical grade silicone. Each electrode 160 comprises an annular portion of conductive material that extends circumferentially around a portion of the implantable body 158. More specifically, each electrode 160 comprises a hollow cylinder of conductive material that extends circumferentially around a portion of the implantable body 158 and, in particular, a portion of the elastomeric material of the implantable body 158. The electrodes 160 may be considered ‘ring’ electrodes.

[0055] Referring back to the embodiment of Figs. 2A and 2B, and with further reference to Figs. 2F and 2G, to strengthen the engagement between the electrodes 160 and the implantable body 158, straps 165 are provided in this embodiment that extend across an outer surface of each electrode 160. In this embodiment, two straps 165 are located on substantially opposite sides of each electrode 160 in a direction perpendicular to the direction of elongation of the implantable body 158. The straps 165 are connected between sections 166a, 166b of the implantable body 158 that are located on opposite sides of the electrodes 160 in the direction of elongation of implantable body, which sections 166a, 166b are referred to hereinafter as side sections. The straps 165 can prevent the side sections 166a, 166b from pulling or breaking away from the electrodes 160 when the implantable body 158 is placed under tension and/or is bent. In this embodiment, the straps 165 are formed of the same elastomeric material as the side sections 166a, 166b. The straps 165 are integrally formed with the side sections 166a, 166b. From their connection points with the side sections 166a, 166b, the straps 165 decrease in width towards a central part of the each electrode 160, minimizing the degree to which the straps 165 cover the surfaces of the electrodes 160 and ensuring that there remains a relatively large amount of electrode surface that is exposed around the circumference of the electrodes 160 to make electrical contact with adjacent body tissue. With reference to Fig. 2D, around a circumference of each electrode 160, at least 75% of the outer electrode surface, at least 80%, at least 85% or at least 90% of the outer electrode surface may be exposed for electrical contact with tissue, for example.

[0056] In alternative embodiments, a different number of straps 165 may be employed, e.g., one, three, four or more straps 165. Where a greater number of straps 165 is employed, the width of each strap 165 may be reduced. The straps 165 may be distributed evenly around the circumference of each electrode 160 or distributed in an uneven manner. Nevertheless, in some embodiments, the straps 165 may be omitted, ensuring that all of the outer electrode surface is exposed for electrical contact with tissue, around a circumference of the electrode 160.

[0057] As indicated above, the implantable body 158 is formed of an elastomeric material such as silicone. The elastomeric material allows the implantable body 158 to bend, flex and stretch such that the implantable body 158 can readily contort as it is routed to a target implantation position and can readily conform to the shape of the body tissue at the target implantation position. The use of elastomeric material also ensures that any risk of trauma to the subject is reduced during implantation or during subsequent use.

[0058] In embodiments of the present disclosure the electrical connection 167 to the electrodes 160 comprises relatively fragile platinum wire conductive elements. With reference to Figs. 2C to 2E, for example, to reduce the likelihood that the platinum wires will break or snap during bending, flexing and/or stretching of the implantable body 158, the electrical connection

167 is provided with wave-like shape and, more specifically, a helical shape in this embodiment, although other non-linear shapes may be used. The helical shape, for example, of the electrical connection 167 enables the electrical connection 167 to stretch, flex and bend in conjunction with the implantable body 158. Bending, flexing and/or stretching of the implantable body 158 typically occurs during implantation of the implantable body 158 in a subject and upon any removal of the implantable body 158 from the subject after use.

[0059] As indicated above, a reinforcement device 168 is also provided in the electrode device 157, which reinforcement device 168 extends through the implantable body 158 and is provided to limit the degree by which the length of the implantable body 158 can extend under tension. The reinforcement device 168 can take the bulk of the strain placed on the electrode device 157 when the electrode device 157 is placed under tension. The reinforcement device

168 is provided in this embodiment by a fiber (e.g., strand, filament, cord or string) of material that is flexible and which has a high tensile strength. In particular, a fiber of ultra-high- molecular-weight polyethylene (UHMwPE), e.g., Dyneema™, is provided as the reinforcement device 168 in the present embodiment. The reinforcement device 168 extends through the implantable body 158 in the length direction of the implantable body 158 and is generally directly encased by the elastomeric material of the implantable body 158.

[0060] The reinforcement device 168 may comprise a variety of different materials in addition to or as an alternative to UHMwPE. The reinforcement device may comprise other plastics and/or non-conductive material such as a poly-paraphenylene terephthalamide, e.g., Kevlar™. In some embodiments, a metal fiber or surgical steel may be used.

[0061] Similar to the electrical connection 167, the reinforcement device 168 also has a wavelike shape and, more specifically, a helical shape in this embodiment, although other non-linear shapes may be used. The helical shape of the reinforcement device 168 is different from the helical shape of the electrical connection 167. For example, as evident from Figs. 2C to 2E, the helical shape of the reinforcement device 168 has a smaller diameter than the helical shape of the electrical connection 167. Moreover, the helical shape of the reinforcement device 168 has a greater pitch than the helical shape of the electrical connection 167.

[0062] When the implantable body 158 is placed under tension, the elastomeric material of the implantable body 158 will stretch, which in turns causes straightening of the helical shapes of both the electrical connection 167 and the reinforcement device 168. As the electrical connection 167 and the reinforcement device straighten 168, their lengths can be considered to increase in the direction of elongation of the implantable body 158. Thus, the lengths of each of the electrical connection 167 and the reinforcement device 168, in the direction of elongation of the implantable body 158, are extendible when the implantable body 158 is placed under tension.

[0063] For each of the electrical connection 167 and the reinforcement device 168, a theoretical maximum length of extension in the direction of elongation of the implantable body 158 is reached when its helical shape (or any other non-linear shape that may be employed) is substantially completely straightened. However, due to the differences in the helical shapes of the electrical connection 167 and the reinforcement device 168, the maximum length of extension of the reinforcement device 168 is shorter than the maximum length of extension of the electrical connection 167. Therefore, when the implantable body 158 is placed under tension, the reinforcement device 168 will reach its maximum length of extension before the electrical connection 167 reaches its maximum length of extension. Indeed, the reinforcement device 168 can make it substantially impossible for the electrical connection 167 to reach its maximum length of extension. Since the electrical connection 167 can be relatively fragile and prone to breaking, particularly when placed under tension, and particularly when it reaches a maximum length of extension, the reinforcement device 168 can reduce the likelihood that the electrical connection 167 will be damaged when the implantable body 158 is placed under tension. In contrast to the electrical connection 167, when the reinforcement device 168 reaches its maximum length of extension, its high tensile strength allows it to bear a significant amount of strain placed on the electrode device 157, preventing damage to the electrical connection 167 and other components of the electrode device 157.

[0064] In consideration of other components of the electrode device 157 that are protected from damage by the reinforcement device 168, it is notable that the implantable body 158 can be prone to damage or breakage when it is placed under tension. The elastomeric material of the implantable body 158 has a theoretical maximum length of extension in its direction of elongation when placed under tension, the maximum length of extension being the point at which the elastomeric material reaches its elastic limit. In this embodiment, the maximum length of extension of the reinforcement device 168 is also shorter than the maximum length of extension of the implantable body 158. Thus, when the implantable body 158 is placed under tension, the reinforcement device 168 will reach its maximum length of extension before the implantable body 158 reaches its maximum length of extension. Indeed, the reinforcement device 168 can make it substantially impossible for the implantable body 158 to reach its maximum length of extension. Since elastomeric material of the implantable body 158 can be relatively fragile and prone to breaking, particularly when placed under tension, and particularly when it reaches its elastic limit, the reinforcement device 168 can reduce the likelihood that the implantable body 158 will be damaged when it is placed under tension.

[0065] In this embodiment, the helical shapes of the reinforcement device 168 and the electrical connection 158 are provided in a concentric arrangement. Due to its smaller diameter, the reinforcement device 168 can locate radially inside of the electrical connection 167. In view of this positioning, the reinforcement device 168 provides a form of strengthening core to the implantable body 158. The concentric arrangement can provide for increased strength and robustness while offering optimal surgical handling properties, with relatively low distortion of the implantable body 158 when placed under tension.

[0066] As indicated, the reinforcement device 168 is directly encased by the elastomeric material of the implantable body 158. The helically-shaped reinforcement device 168 therefore avoids contact with material other than the elastomeric material in this embodiment. The helically shaped reinforcement device is not entwined or intertwined with other strands or fibers, for example (e.g., as opposed to strands of a rope), ensuring that there is a substantial amount of give possible in relation to its helical shape. The helical shape can move to a straightened configuration under tension as a result, for example.

[0067] The arrangement of the reinforcement device 168 is such that, when the implantable body 158 is placed under tension, the length of the reinforcement device 168 is extendible by about 20% of its length when the implantable body 158 is not under tension. Nevertheless, in embodiments of the present disclosure, a reinforcement device 168 may be used that is extendible by at least 5%, at least 10%, at least 15%, at least 20% or at least 25% or otherwise, of the length of the reinforcement device when the implantable body 158 is not under tension. The maximum length of extension of the reinforcement device 168 in the direction of elongation of the implantable body 158 may be about 5%, about 10%, about 15%, about 20% or about 25% or otherwise of its length when the implantable body 158 is not under tension.

[0068] As represented in Fig. 2C, the reinforcement device 168 has a relatively uniform helical configuration along its length. However, in some embodiments, the shape of the reinforcement device 168 can be varied along its length. For example, the reinforcement device 168 can be straighter (e.g., by having a helical shape with smaller radius and/or greater pitch) adjacent the electrodes 160 in comparison to at other portions of the implantable body 158. By providing this variation in the shape of the reinforcement device 168, stretching of the implantable body 158 may be reduced adjacent the electrodes 160, where there could otherwise be a greater risk of the electrodes 160 dislocating from the implantable body 158. This enhanced strain relief adjacent the electrodes 160 can be provided while still maintaining the ability of the reinforcement device 168, and therefore implantable body 158, to stretch to a desirable degree at other portions of the implantable body 158.

[0069] As indicated, the electrical connection 167 in this embodiment comprises relatively fragile platinum wire conductive elements. At least 4 platinum wires are provided in the electrical connection 167 to each connect to a respective one of the four electrodes 160. The wires are twisted together and electrically insulated from each other. Connection of a platinum wire of the electrical connection 167 to the most distal of the electrodes 160 is illustrated in Fig. 2C. As can be seen, the wire is connected to an inner surface 172 of the electrode 160, adjacent a distal end of the electrode 160, albeit other connection arrangements can be used, including connection arrangements that allow for the montage of electrodes to be adjusted (i.e., to allow different pairings of electrodes) in order to, for example, improve signal reliability.

[0070] The reinforcement device 168 extends through the hollow center of each of the electrodes 160. The reinforcement device 168 extends at least from the distal most electrode 160, and optionally from a region adjacent the distal tip 159 of the implantable body 158, to a position adjacent the amplifier 163. In some embodiments, the reinforcement device 168 may also extend between the amplifier 163 and the processing unit 144. In some embodiments, the reinforcement device 168 may extend from the distal tip 159 and/or the distal most electrode 160 of the implantable body 158 to the processing unit 144.

[0071] To prevent the reinforcement device 168 from slipping within or tearing from the elastomeric material of the implantable body 158, a series of knots 169 are formed in the reinforcement device 168 along the length of the reinforcement device 168. For example, with reference to Fig. 2F, a knot 169a can be formed at least at the distal end of the reinforcement device 168, adjacent the distal tip 159 of the implantable body 158, and/or knots 169 can be formed adjacent one or both sides of each electrode 160. The knots may alone provide resistance to movement of the reinforcement device 168 relative to the elastic material of the implantable body 158 and/or may be used to fix (tie) the reinforcement device 168 to other features of the device 157.

[0072] In the present embodiment for example, as illustrated in Fig. 2C, the reinforcement device 168 is fixed, via a knot 169b, to each electrode 160. To enable the reinforcement device 168 to be fixed to the electrode 160, the electrode 160 comprises an extension portion 173 around which knots 169 of the reinforcement device 168 can be tied. As shown in Fig. 2C, the extension portion 173 can include a loop or arm of material that extends across an open end of the hollow cylinder forming the electrode 160.

[0073] With reference to Figs. 2A, 2B, 2F, and 2G, the electrode device 158 comprises at least one anchor 164, and in this embodiment of plurality of anchors 164. The plurality of anchors 164 are positioned along a length of the implantable body 158, each adjacent a respective one of the electrodes 160. Each anchor 164 is configured to project radially outwardly from the implantable body 158 and specifically, in this embodiment, at an angle towards a proximal end of the implantable body 158. Each anchor 164 is in the form of a flattened appendage or fin with a rounded tip 170. The anchors 164 are designed to provide stabilization to the electrode device 157 when it is in the implantation position. When implanted, a tissue capsule can form around each anchor 164, securing the anchor 164 and therefore the implantable body 158 into place. In this embodiment, the anchors 164 are between about 0.5 mm and 2 mm in length, e.g., about 1 mm or 1 .5 mm in length.

[0074] So that the anchors 164 do not impede implantation of the electrode device 157, or removal of the electrode device 157 after use, each anchor 164 is compressible. The anchors 164 are compressible (e.g., foldable) to reduce the degree by which the anchors 164 projects radially outwardly from the implantable body 158. To further reduce the degree by which the anchors 164 project radially outwardly from the implantable body 158 when compressed, a recess 171 is provided in a surface of the implantable body 158 adjacent each anchor 164. The anchor 164 is compressible into the recess 171 . In this embodiment, the anchors 164 project from a bottom surface of the respective recess 171 and the recess 171 extends on both proximal and distal sides of the anchor 164. Accordingly, the anchors 164 can be compressed into the respective recesses in either a proximal or distal direction. This has the advantage of allowing the anchors 164 to automatically move into a storage position in the recess 171 when pulled across a tissue surface or a surface of an implantation tool such as delivery device, in either of a proximal and a distal direction.

[0075] The electrode device 157 of the present embodiment is configured for use in monitoring electrical activity in the brain and particularly for monitoring electrical activity relating to epileptic events in the brain. The electrode device 157 is configured to be implanted at least partially in a subgaleal space between the scalp and the cranium. At least the electrodes 160 and adjacent portions of the implantable body 158 are located in the subgaleal space.

[0076] An illustration of the implantation location of the electrodes 160 is provided in Fig. 2H. As can be seen, the electrodes 160 locate in particular in a pocket between the galea aponeurotica 206 and the pericranium 203. When implanted, the first and second electrode pairs 161 , 162 are located on respective sides of the midline of the head of the subject in a substantially symmetrical arrangement. The first and second electrode pairs 161 , 162 therefore locate over the right and left hemispheres of the brain, respectively. For example, the first electrode pair 161 can be used to monitor electrical activity at right hemisphere of the brain and the second electrode pair 162 can be used to monitor electrical activity at the left hemisphere of the brain, or vice-versa. Independent electrical activity data may be recorded for each of the right and left hemispheres, e.g., for diagnostic purposes. To position the electrode pairs 161 , 162 over the right and left hemispheres of the brain, the implantable body 158 of the electrode device 157 is implanted in a medial-lateral direction over the cranium of the subject’s head. The electrode pairs 161 , 162 are positioned away from the subject’s eyes and chewing muscles to avoid introduction of signal artifacts from these locations. The electrode device 157 implanted under the scalp in a position generally as illustrated in Fig. 2I.

[0077] The local processing device 144 of the sensor array 102 may be implanted under skin tissue of the patient. With reference to Fig. 3, the local processing device 144 can include an electrical amplifier 146, a battery 148, battery charging circuitry 149, a transceiver 150, an analogue to digital converter (ADC) 152, and a processor 154 to process electrical signals received from or transmitted to the electrodes 160. The local processing device 144 can include a memory 156 to store signal processing data. The local processing device 144 may be similar to a processing device of a type commonly used with cochlear implants, although other configurations are possible. A reference electrode 160R may also be electrically coupled to the local processing device 144, in embodiments, and specifically may be physically coupled to or part of the local processing device 144. In specific embodiments, the reference electrode 160R may be part of a housing of the local processing device 144.

[0078] The data processed and stored by the local processing device 144 may be raw EEG data or partially processed (e.g., partially or fully compressed) EEG data, for example. The EEG data may be transmitted from the local processing device 144 wirelessly, or via a wired connection, to the processor device 104 for further processing and analyzing of the data. The processor device 104 may analyze EEG signals (or other electrical signals) to determine if a seizure event has occurred. Data regarding the event may be generated by the processor device 104 on the basis of the analysis, as described further herein. In one example, the processor device 104 may analyze brain activity signals to determine if an epileptic event has occurred and data regarding the epileptic event (e.g., classification of the event, which hemisphere the event was focused in, what area of the brain the event was focused in, etc.) may be generated by the processor device 104 on the basis of the analysis.

[0079] By carrying out data analysis externally to the sensor array 102, using the processor device 104, for example, there may be a reduction in power consumption within the sensor array 102, enabling the sensor array 102 to retain a smaller geometrical form. Moreover, the processor device 104 may have significantly higher processing power than would be possible with any processor included in the sensor array 102. The processor device 104 may run software that continuously records EEG data received from the sensor array 102.

[0080] In various embodiments envisioned herein, some or all processing functions of the processor device 104 may be implemented externally (i.e., not in an implanted device) or, alternatively, via one or more implantable devices (e.g., having on-board memory/processing capability and not requiring connection to external processing components). In particular, at least one such implantable device may be electrode device 157 disposed under the scalp of the patient (e.g., as described with respect to Figs. 2A-2I). In some embodiments, though, some or all functions of the processor device 104 may be implemented via one or more other processing devices implanted in a different portion of the body of the patient. For example, some or all of functions of the processor device 104 may be implemented in one or more neurostimulation devices disposed in any suitable portion of the patient’s body (e.g., proximate to the brain or the spinal cord), and/or in one or more other processing devices in the body of the patient. Moreover, additional sensors and/or other equipment 108 can include sensors and/or other equipment implanted in the body of the patient, and/or devices worn externally by the patient.

[0081] Moreover, although the present description gives examples of an implantable device by which EEG signals are received from a sensor array 102 under the scalp of the patient, it should be envisioned that other structural arrangements are possible without deviating from the modeling techniques of the present description. Particularly, the sensor array 102 may be provided in one or more wearable devices affixed the patient and capable of receiving the EEG data of the patient. For example, referring back to the description of Figs. 2A-2I, although the electrodes 160 are described as being along the implantable body 158 implanted under the scalp of the patient, electrodes 160 can instead be positioned along an elongate body positioned over (i.e., externally to) the scalp of the patient. Likewise, the additional elements of the described electrode device 157 (e.g., the amplifier 163 and the local processing unit 144) can be affixed externally on the patient. These various components of an externally worn electrode device 157 can be positioned similarly to as depicted in Fig. 2I, with the exception that the components are positioned over the scalp of the patient rather than under the scalp. That is, referring to Fig. 2H, the externally worn components are over the skin 201 . For example, externally worn electrodes 160 can be aligned over the scalp in similar positions as previously described relative to the hemispheres of the brain of the patient, so as to facilitate sensitivity to electrical activity in particular hemispheres/foci of the brain as described herein. In any event, an externally worn electrode device 157 can include similar elements to those described above regarding an implantable device, e.g. protective materials to protect the fragile elements of the electrode device 157.

[0082] THE PROCESSOR DEVICE [0083] Turning now to Fig. 4A, the system 100 is presented as a block diagram in greater detail. As depicted in Fig. 4A, the system 100 includes, in embodiments, the sensor array 102, the processor device 104, and the user interface 106. The sensor array 102 may sense or collect EEG data and communicate the EEG data to the processor device 104. As should be understood from the foregoing discussion of Figs. 1 , 2A-2I, and 3, the sensor array 102 may, in embodiments, include an implanted and/or externally wearable array of electrode devices 110 that provide electrical signal data and, in particular, provide electrical signal data indicative of brain activity of the patient (e.g., EEG signal data). As should also be understood in view of the description above, the sensor array 102 may, in embodiments, be disposed beneath the scalp of the patient - on and extending into the cranium 204 - so as to facilitate accurate sensing of brain activity. Alternatively, in other embodiments, the sensor array 102 may be disposed over the scalp of the patient, e.g. as part of one or more wearable devices. The system 100 further includes, in some embodiments, the additional sensor(s) and/or other equipment 108 which, generally speaking, may support the detection of seizure events and/or the delivery of therapeutic treatment of epilepsy and/or associated conditions (e.g., via drug intake and/or neurostimulation).

[0084] In embodiments, additional sensors 108 include a PPG sensor that detects, using a photodetector circuit, light that is transmitted through or reflected from the patient after the light interacts with the blood just beneath the surface of the patient’s skin. The PPG sensor may be any type of PPG sensor suitable for disposal on the patient and, in particular, suitable for operation from a portable power source such as a battery. The PPG sensor may be disposed at any of a variety of positions on the patient including, but not limited to, the patient’s finger, toe, forehead, earlobes, nasal septum, wrist, ankle, arm, torso, leg, hand, or neck. In some embodiments, the PPG sensor may be integrated with the sensor array 102 and placed on or beneath the scalp of the patient with the sensor array 102. In other embodiments, the PPG sensor may be integrated with the processor device 104, and still in others the PPG sensor may be distinct from both the sensor array 102 and the processor device 104. The PPG sensor may be one or more PPG sensors, disposed as connected or distinct units on a variety of positions on the patient (so-called multi-site photoplethysmography). In embodiments implementing multiple PPG sensors, the multiple PPG sensors may be of the same type, or may be different, depending on the location of each on the patient, the environment in which each is disposed, the location of each in the hardware (e.g., separate from other devices or integrated with the processor device 104, for example), etc.

[0085] Optional other sensors 108, such as microphones, detect sound related to the patient and the patient’s environment. Such a microphone may be any type of microphone suitable for disposal on the patient and suitable for operation from a portable power source such as a battery. In particular, the microphone may be a piezoelectric microphone, a MEMS microphone, or a fiber optic microphone. In embodiments, an accelerometer device may be adapted to measure vibrations and, accordingly, to detect sound, rendering the accelerometer device suitable for use as a microphone. The microphone may be disposed at any of a variety of positions on the patient including, but not limited to, the patient’s head, arm, torso, leg, hand, or neck. In some embodiments, the microphone may be integrated with the sensor array 102 and placed on or beneath the scalp of the patient with the sensor array 102, while in others the microphone may be integrated with the processor device 104, and still in others the microphone may be distinct from both the sensor array 102 and the processor device 104. In embodiments, especially those in which the patient’s voice is the primary sensing target for the microphone, the microphone senses sound via bone conduction. In some embodiments, the microphone may be integrated with a hearing or vestibular prosthesis. The microphone may be one or more microphones, disposed as an array in a particular position on the patient, or as distinct units on a variety of positions on the patient. In embodiments implementing multiple microphones, the multiple microphones may be of the same type, or may be different, depending on the location of each on the patient, the environment in which each is disposed (e.g., sub-scalp vs. not), the location of each in the hardware (e.g., separate from other devices or integrated within the processor device 104, for example), etc. Each may have the same or different directionality and/or sensitivity characteristics as the others, depending on the placement of the microphone and the noises or vibrations the microphone is intended to detect.

[0086] The microphone may detect the patient’s voice, in embodiments, with the goal of determining one or more of: pauses in vocalization; stutters; periods of extended silence; abnormal vocalization; and/or other vocal abnormalities that, individually or in combination with data from the sensor array 102, optional accelerometers), and/or self-reported data received via the user interface 106, may assist algorithms executing within the processor device 104 (and/or other processing devices described herein) in determining whether the patient has experienced an event of interest (e.g., a seizure) and, if so, classifying the event as described herein. In embodiments, the microphone may also detect other noises in the patient’s environment that may be indicative that the patient experienced an event of interest. For example, the microphone may detect the sound of glass breaking, which may indicate that the patient has dropped a glass. Such an indication, in conjunction with electrical signals detected by the sensor array 102, may provide corroboration that the patient has, in fact, experienced an event of interest, such as a seizure.

[0087] In embodiments, the additional sensors 108 may include an accelerometer that detects movement and/or orientation of the patient. The accelerometer may be any type of accelerometer suitable for disposal on the patient and suitable for operation from a portable power source such as a battery. In particular, and by way of example, the accelerometer may be a chip-type accelerometer employing MEMS technology, and may include accelerometers employing capacitive, piezoelectric resistive, or magnetic induction technologies. Like the optional microphone, the optional accelerometer may be in any of a variety of positions on the patient including, but not limited to, the patient’s head, arm, torso, leg, hand, or neck. In some embodiments, there may be multiple accelerometers, to detect motions in different parts of the body. In some embodiments, an accelerometer may be integrated with the sensor array 102 and placed on or beneath the scalp of the patient with the sensor array 102, while in others an accelerometer may be integrated with the processor device 104 and still in others the accelerometer may be distinct from both the sensor array 102 and the processor device 104. The accelerometer may be one or more accelerometers, disposed as an array in a particular position on the patient, or as distinct units on a variety of positions on the patient. In embodiments implementing multiple accelerometers, the multiple accelerometers may be of the same type, or may be different, depending on the location of each on the patient, the environment in which each is disposed (e.g., sub-scalp vs. not), the location of each in the hardware (e.g., separate from other devices or integrated within the processor device 104, for example), etc. Each may have the same or different sensitivity characteristics and/or number of detectable axes as the others, depending on the placement of the accelerometer and the motions and/or vibrations the accelerometer is intended to detect.

[0088] The optional accelerometer may detect tremors, pauses in movement, gross motor movement (e.g., during a tonic-clonic seizure), falls (e.g., during an atonic or drop seizure or a tonic seizure), repeated movements (e.g., during clonic seizures), twitches (e.g., during myoclonic seizures), and other motions or movements that, in combination with data from the sensor array 102, the optional microphone, and/or self-reported data received via the user interface 106, may assist algorithms executing via the processor device 104 (and/or via other processing devices described herein) in determining whether the patient has experienced an invent of interest (e.g., a seizure) and, if so, classifying the event. In embodiments, the accelerometer may function as an additional microphone 20 or may act as the only microphone.

[0089] Like the optional microphone, the optional accelerometer may be in any of a variety of positions on the patient including, but not limited to, the patient’s head, arm, torso, leg, hand, or neck. In some embodiments, there may be multiple accelerometers, to detect motions in different parts of the body. In some embodiments, an accelerometer may be integrated with the sensor array 102 and placed on or beneath the scalp of the patient with the sensor array 102, while in others an accelerometer may be integrated with the processor device 104.

[0090] Together, the sensor array 102 and, if present, the PPG sensor(s), microphone(s) and/or accelerometer(s) may provide data from which biomarker data related to the patient(s) may be extracted. The system 100 may be configured to determine a variety of biomarkers depending on the inclusion and/or placement of the various sensor devices (i.e., the sensor array 102 and, if present, the PPG sensor(s), microphone(s), and/or accelerometer(s)). [0091] In embodiments in which the additional sensors and/or other equipment 108 includes one or more therapeutic devices, the one or more therapeutic devices may transmit information to the processor device 104 (and/or other processing devices described herein). Information transmitted via the one or more therapeutic devices may assist algorithms executing via the processor device 104 (and/or other processing devices), for example, to forecast seizure activity in the patient and/or administer therapeutic treatment to the patient according to the techniques of the present disclosure. For example, in embodiments where other equipment 108 includes a drug pump, the drug pump may transmit data indicative of quantities and timings of drugs administered to the patient, and/or future scheduled administrations of drugs to the patient, which data may be provided as input to the forecasting model described herein. Similarly, in embodiments where other equipment 108 includes a neurostimulator device, the neurostimulator device may transmit data indicative of previous or future neurostimulation (e.g., quantity, timing, duration, and/or location thereof), which may similarly be provided as input to the forecasting model, the operation of which will be described further in subsequent portions of the present description.

[0092] The processor device 104 receives data from the sensor array 102, the optional additional sensors and/or other equipment 108, and the user interface 106 and, using the received data, may detect and classify events of interest. The processor device 104 includes communication circuitry 256, a microprocessor 258, and a memory device 260. The microprocessor 258 may be any known microprocessor configurable to execute the routines necessary for detecting and classifying events of interest, including, by way of example and not limitation, general purpose microprocessors (GPUs), RISC microprocessors, digital signal processors (DSPs), application specific integrated circuits (ASICs), and field-programmable gate arrays (FPGAs).

[0093] The communication circuitry 256 may be any transceiver and/or receiver/transmitter pair that facilitates communication with the various devices from which the processor device 104 receives data and/or transmits data. The communication circuitry 256 is communicatively coupled, in a wired or wireless manner, to each of the sensor array 102, the optional sensors 108, and the user interface 106. Additionally, the communication circuitry 256 is coupled to the microprocessor 258, which, in addition to executing various routines and instructions for performing analysis, may also facilitate storage in the memory 260 of data received, via the communication circuity 256, from the sensor array 102, the optional sensors 108, and the user interface 106.

[0094] The memory 260 may include both volatile memory (e.g., random access memory (RAM)) and non-volatile memory, in the form of either or both of magnetic and solid-state media. In addition to an operating system (not shown), the memory 260 may store sensor array data and/or other sensor/equipment data 262 received from the sensor array 102 and/or other optional sensors/equipment 108, and user report data 268 received from the user (and/or other person such as a caregiver) via the user interface 106. In particular, the user report data 268 may include reports from the user, received via the user interface 106, of various types of symptoms and/or other data. By way of non-limiting examples, the information reported via the user interface 106 may include: perceived seizures/epileptic events; characteristics or features of perceived seizures/epileptic events such as severity and/or duration, perceived effects on memory, or other effects on the individual’s wellbeing (such as their ability to hold a cup or operate a vehicle); other types of physiological symptoms (e.g., headaches, impaired or altered vision, involuntary movements, disorientation, falling to the ground, repeated jaw movements or lip smacking, etc.); characteristics or features of other symptoms (e.g., severity and/or duration); medication ingestion information (e.g., medication types, dosages, and/or frequencies/timing, additionally or alternatively to information provided automatically via one or more drug pumps); neurostimulation treatment application information (e.g., types, dosages, target locations, and/or frequency/timing, additionally or alternatively to information provided automatically via one or more neurostimulator devices), perceived medication side-effects; characteristics or features of medication side-effects (e.g., severity and/or duration), and other user reported information (e.g., food and/or drink ingested, activities performed (e.g., showering, exercising, working, brushing hair, etc.), tiredness, stress levels, etc.), as well as the timing of each. These user reports and, in particular, reports related to symptoms, and to medication or neurostimulation, may facilitate use of the software models described herein to detect seizure activity via the detection model, forecast seizure activity via the forecasting model, and/or titrate therapeutic treatment via the titration model.

[0095] As will be described in greater detail below, the memory 260 may also store one or more models 270 for producing output based upon a set of feature values 272 extracted from the sensor array data and/or other sensor/equipment data 262, the data provided by the sensor/equipment 108 data, and the user report data 268. The one or more models 270 are referred to interchangeably herein as a model 270 or models 270, and it should be understood that one or more models may be implemented to make the determinations described herein, and that references to a single model, unless specifically stating that the model is unitary, are intended to mean “one or more models.” One or more models may include, for example the detection model, forecasting model, and/or titration model as described herein. Model outputs 274 output by the models 270 (e.g., relating to detected seizure events, forecasted seizure events, and/or titration parameters) may be stored in the memory 260. A data pre-processing routine 271 may provide pre-processing of the sensor array data and/or other sensor/equipment data 262, the user report data 268 and, if present, the data from the optional sensors and/or other equipment 108. As will be understood (and, in part, described below), the data preprocessing routine 271 may provide a range of pre-processing steps including, for example, filtering and extraction from the data of the feature values 272. Lastly, in embodiments, the memory 260 may store an image generation routine 276 configured to receive the data 262 and generate one or more images from the data 262. For example, the image generation routine 276 may generate EEG waveforms from the raw EEG data recorded by the sensor array 102. The EEG waveforms may, in turn, be analyzed as images, rather than as raw EEG data by the various models and routines disclosed herein.

[0096] Of course, it should be understood that wherever a routine, model, or other element stored in memory is referred to as receiving an input, producing or storing an output, or executing, the routine, model, or other element is, in fact, executing as instructions on the microprocessor 258. Further, those of skill in the art will appreciate that the model or routine or other instructions would be stored in the memory 260 as executable instructions, which instructions the microprocessor 258 would retrieve from the memory 260 and execute. Further, the microprocessor 258 should be understood to retrieve from the memory 260 any data necessary to perform the executed instructions (e.g., data required as an input to the routine or model), and to store in the memory 260 the intermediate results and/or output of any executed instructions.

[0097] In embodiments, the data pre-processing routine 271 may also extract from the sensor array data and/or other sensor/equipment data 262, and the optional other sensor data, one or more biomarkers. The one or more biomarkers may be included among the feature values that are provided as inputs to the models 270, in embodiments, in order for the models 270 to produce at least a portion of the outputs 274.

[0098] The data stored in the sensor array data and/or other sensor/equipment data 262 and the user report data 268 are stored with corresponding time stamps such that the data may be correlated between data types. For example, each value in the data 262 should have a corresponding time stamp such that the data from optional sensors/equipment and the user report data 268 for the same time (and/or different times) can be compared, allowing the various types of data to be lined up and analyzed for any given time period. With respect to the user report data 268, there may be multiple time stamps for any particular user report, including, for example, the time that the user filled out the user report and the time of the event, symptom, or other information that the user was reporting (as reported by the user).

[0099] Events need not be contemporaneous to be relevant or related, or to be feature values input into the models 270. Put another way, the models 270 may consider temporal relationships between non-contemporaneous events in detecting, predicting, and/or classifying an event and/or determining titration of therapeutic treatments. By way of example and not limitation, an electrical activity event (e.g., EEG signals) indicating a seizure may be classified as a particular type of event if preceded by the ingestion of medication, and as a different type of event if not preceded by the ingestion of the medication. Other examples of noncontemporaneous events preceding a seizure that are precursors are patient subjective reports of auras or optical lights, shortness of breath or increased cardiac pulse rate, and acoustic biomarkers suggesting the alteration of speech patterns. Additionally, the system 100 and, in particular, the models 270, may identify pre- and/or post-seizure events, such as unsteady balance, falls, slurred speech, or brain activity patterns that are indicative of a pre- and/or postseizure event.

[0100] Of course, contemporaneous events may also be relevant. For example, accelerometer data indicative of a generalized tonic-clonic (i.e., grand mal) seizure may be classified as such if it is accompanied by contemporaneous electrical activity indicative of such a seizure.

[0101] It is considered that while some objectives of the system 100 may be achieved using the models 270 (e.g. static models) according to known data about the patient and/or the condition, other objectives of the system 100 may implement one or more trained artificial intelligence (Al) models to achieve still additional benefits.

[0102] Fig. 4B is a block diagram of an example system 300 similar to the system 100 of Fig. 4A, but which includes trained artificial intelligence (Al) models 302 instead of the models 270 based on static algorithms. That is, Fig. 4B corresponds generally to Fig. 4A, with the only difference between the system 100 and the system 300 in the respective figures being the inclusion of the trained Al models 302 rather than the models 270 based on static algorithms. The system 300, as depicted in Fig. 4B is the same in all respects as in Fig. 4A, above, except that the trained Al models 302 are created using Al algorithms to search for and identify patterns in training data and, upon implementation in the processor device 104, to receive the data 262 and/or the user report data 268, and to determine from those data feature values 272 from which the trained Al models 302 may perform the evaluative functions to determine the model outputs 274. Like the static models 270, the trained Al models 302 may consider temporal relationships between non-contemporaneous events and/or biomarkers in performing the evaluative functions. Although the system 300 of Fig. 4B and system 100 of Fig. 4A are depicted separately, it should be appreciated that aspects of the systems 300 may be combined. That is, static models 270 and Al algorithms 302 may be combined or may interact with each other, in some embodiments.

[0103] In embodiments, the model(s) 270, 302 may use long-term EEG monitoring across the brain (e.g., across both hemispheres of the brain individually) to detect, track, and train a system that can aid a physician in treating individuals with epilepsy through the identification of seizures (including, in some embodiments, through the determination of hemispheres or foci of seizures). Such models could also incorporate clinical feedback from the patient, for instance from the user report data 268, including the symptoms experienced during each seizure event, and the relative timing and/or severity thereof. New reported events and symptoms could be employed to seek and/or detect new patterns in the EEG associated with seizures, for example a change in types or symptoms of seizures, a change in severity, and/or a move of the seizure foci from one region to another, and the connections between these parameters. For example, when a new EEG seizure event is determined to have a focus in a hemisphere different from previous seizure events, the model(s) 270, 302 may look at the user report data 268 to find the event as reported by the patient and may automatically associate the symptom to the event (e.g., associating a change in focus with a change in symptoms and/or severity). Conversely, the model(s) may identify new clinical symptoms/events reported by the patient (e.g., symptoms and/or severity thereof) and may identify new patterns in the EEG in the hemisphere in which newly detected events are identified. Either way, in embodiments, the model(s) may provide as output feedback to the physician regarding the number of epileptic events, including the events from each hemisphere (reported independently), the symptoms associated with each event, the severity of events, and/or the relationships of the above parameters to each other (e.g., symptoms segregated based upon association with events in each hemisphere, and/or symptoms associated with the most severe events). These data may be used to administer therapeutic treatment specific to particular symptoms, foci, etc., for example to treat the foci either together or independently by prescribing a new drug for the new foci or an adjusted treatment dosage (or delivery timing) based on detected seizure cycles.

[0104] The model(s) 270, 302 may also provide output indicating whether drug and/or neurostimulation treatments are efficacious with respect to particular aspects of seizures (e.g. example in mitigating or eliminating particular symptoms, reducing the severity or frequency of seizures, or by reducing in the prevalence of seizures in particular foci, new foci, or both) by monitoring the occurrence of epileptiform EEG events with respect to each aspect (e.g., foci, severity, and/or symptoms, and the effects of therapeutic treatment with respect to each), rather than looking only at the sum of the detected or reported seizures. That is, in embodiments in which the model(s) are operative to track and associate specific epileptiform activity with specific symptoms and/or foci, the models(s), having associated specific epileptiform features with those foci, can track seizures based on foci, symptoms, or both, and can determine whether a particular treatment is affecting a specific class of seizures and, if so, which class of seizures are targeted using the treatment in question.

[0105] Moreover, as the model(s) 270, 302 become adapted (through feedback and/or retraining) to detect, track, and/or predict particular aspects of seizures (e.g., associated with multiple foci in a patient), the model(s) may track, detect, and predict seizure cycles of each focus, severity, and/or type of seizure independently, allowing, for example, predictions of when to expect the symptoms of a specific seizure associated with one focus or type of seizure versus another focus or type. Similarly, the models 270, 302 may be adapted to determine titration of therapeutic treatment particularly responsive to detected or forecasted seizures or symptoms thereof associated with particular foci (e.g., targeted neurostimulation).

[0106] Returning to Fig. 4B, the trained Al models 302 may be created by an adaptive learning component configured to “train” Al models (e.g., create the trained Al models 302) to produce the model outputs 274 using as inputs raw or pre-processed (e.g., by the data preprocessing routine 271) data from the sensor array data and/or other sensor/equipment data 262 (as such, or converted into EEG waveform images) and, optionally, other data such as medication and/or neurostimulation treatment information. As described herein, the adaptive learning component may use a supervised or unsupervised machine learning program or algorithm. The machine learning programs or algorithms may employ neural networks, such as convolutional neural networks (CNNs), deep learning neural networks, residual neural networks, or combined learning modules or programs that learn in two or more features or feature datasets in particular areas of interest. The machine learning programs or algorithms may also include natural language processing, image recognition, multi-class output, semantic analysis, automatic reasoning, regression analysis, support vector machine (SVM) analysis, decision tree analysis, random forest analysis, K-Nearest neighbor analysis, naive Bayes analysis, clustering, reinforcement learning, and/or other machine learning algorithms and/or techniques. Machine learning may involve identifying and recognizing patterns in existing data (i.e., training data) such as temporal correlations between biomarkers in the data 262, in order to facilitate making predictions for subsequent data and/or evaluating confidence levels of classification outputs (e.g., detection and confidence of seizure events, predictions and confidences of future seizure events, and/or determinations of titration).

[0107] The trained Al models 302 may be created and trained based upon example (e.g., “training data”) inputs or data (which may be termed “features” and “labels”) in order to make valid and reliable predictions for new inputs, such as testing level or production level data or inputs. In supervised machine learning, a machine learning program operating on a server, computing device, or other processor(s), may be provided with example inputs (e.g., “features”) and their associated, or observed, outputs (e.g., “labels”) in order for the machine learning program or algorithm to determine or discover rules, relationships, or other machine learning “models” that map such inputs (e.g., “features”) to the outputs (e.g., “labels”), for example, by determining and/or assigning weights or other metrics to the model across its various feature categories. Such rules, relationships, or other models may then be provided subsequent inputs in order for the model, executing on the server, computing device, or other processors), to predict an expected output based on the discovered rules, relationships, or model.

[0108] In unsupervised learning, the server, computing device, or other processor(s), may be required to find its own structure in unlabeled example inputs, where, for example, multiple training iterations are executed by the server, computing device, or other processor(s) to train multiple generations of models until a satisfactory model (e.g., a model that provides sufficient prediction accuracy when given test level or production level data or inputs) is generated. The disclosures herein may use one or both of such supervised or unsupervised machine learning techniques.

[00100] Throughout this specification, the phrase “evaluative functions” is used to refer to the collective potential outputs of the models including at least: detecting and/or classifying events that are occurring, according to foci, to foci hemisphere, to epileptiform, and/or to seizure symptoms; detecting and/or classifying events that have occurred according to foci, to foci hemisphere, to epileptiform, and/or to seizure symptoms; predicting and/or classifying events that are about to occur according to foci, to foci hemisphere, to epileptiform, and/or to seizure symptoms; determining parameters of titration to respond to detection and/or prediction of events (e.g., administration of drugs and/or neurostimulation); detecting and/or classifying a lateralization of a detected or predicted event (e.g., a hemisphere in which an event is or is predicted to be focused, a focus of a predicted event, an epileptiform type of a predicted event, symptoms of a predicted event); detecting and/or predicting a focus of a detected or predicted event (e.g., a brain region in which onset occurred or is predicted to occur); detecting, tracking, and/or predicting individual cycles of seizure activity associated with a particular focus, with a particular hemisphere, with a particular epileptiform, and/or with a particular seizure type; detecting and/or predicting how a medication or neurostimulation treatment will affect seizure activity and/or focus of one or more types of seizures and/or of seizures associated with one or more foci; and/or detecting new patterns associated with temporal changes in the origin of seizure activity across hemispheres.

[00101] EXAMPLE PROCESSING USING EXTERNAL DEVICE

[00102] Fig. 5 is a block diagram depicting another example embodiment, in which the evaluative functions take place on a device other than the processor device 104 and, specifically, on an external device 273. In the embodiments depicted in Fig. 5, it is contemplated that the models performing the evaluative functions may include static models 270 from Fig. 4A and/or the trained Al models 302 of Fig. 4B and, as a result, Fig. 5 illustrates an alternate embodiment of Figs. 4A and 4B. In the embodiments contemplated within Fig. 5, the processor device 104 generally collects the data from the sensor array 102 (and, optionally, the user interface 106 and/or other sensors/equipment 108). These data are stored in the memory 260 of the processor device 104 as the sensor array data and/or other sensor/equipment data 262, the user report data 268, etc., respectively. While the processor device 104 may be equipped to perform the modeling - that is may have stored in the memory 260 the models 270 or 302 and the data pre-processing routine(s) 271 , and be configured to perform the evaluative functions to output feature values 272 and model outputs 274 - in the embodiments contemplated by Fig. 5, these functionalities are optional. Instead, the microprocessor 258 may be configured to communicate with the external device 273 such that the external device 273 may perform the evaluative functions. Because Fig. 5 depicts embodiments in which either the processor device 104 or the external device 273 (or both) may use the models 270 or 302 to perform the evaluative functions, the model(s) 270, 302 in Fig. 5 are depicted as optional in each of the devices 104, 273. It should be understood, however, that each of the models 270 or 302 must be present on at least one or the other of the devices 104, 273 in each embodiment (e.g., on a same one of the devices 104, 273, or distributed between the devices 104, 273).

[0109] The external device 273 may be a workstation, a server, a cloud computing platform, or the like, configured to receive data from one or more processor devices 104 associated with one or more respective patients. The external device 273 may include communication circuitry 275, coupled to a microprocessor 277 that, in turn, is coupled to a memory 279. The microprocessor 277 may be any known microprocessor configurable to execute the routines necessary for producing the evaluative results, including, by way of example and not limitation, general purpose microprocessors (GPUs), RISC microprocessors, digital signal processors (DSPs), application specific integrated circuits (ASICs), and field-programmable gate arrays (FPGAs).

[0110] The communication circuitry 275 may be any transceiver and/or receiver/transmitter pair that facilitates communication with the various devices from or to which the processor device 273 receives data and/or transmits data. The communication circuitry 256 is coupled to the microprocessor 277, which, in addition to executing various routines and instructions for performing analysis, may also facilitate storage in the memory 279 of data received, via the communication circuity 275, from the processor devices 104 of the one or more patients.

[0111] The memory 279 may include both volatile memory (e.g., random access memory (RAM)) and non-volatile memory, in the form of either or both of magnetic or solid state media. In addition to an operating system (not shown), the memory 279 may store received data 281 received from the processor devices 104, including the data 262 received from the sensor array 102 and/or other sensors/equipment 108, and/or user report data 268 received from the user via the user interface 106.

[0112] Like the processor device 104, the external device 273 may have, stored in its memory 279, the static models 270 and/or the trained Al models 302, as well as data pre-processing routines 271. The microprocessor 277 may execute the data pre-processing routines 271 to refine, filter, extract biomarkers from, etc. the received data 281 and to output feature values 272 (which, in embodiments, include biomarkers or relationships between biomarkers). The microprocessor 277 may also execute the models 270, 302, receiving as inputs the feature values 272 and outputting model outputs 274. One or more reporting routines 283 stored on the memory 279, when executed by the microprocessor 277, may facilitate outputting reports for use by the patient(s) or by medical personnel, such as physicians, to review the data and or treat the patient(s).

[0113] The embodiments depicted in Fig. 5 also contemplate that, even in embodiments in which the processor device 104 executes the models 270 and/or 302 to produce model outputs, the processor device 104 may communicate the model outputs 274, as well as the data 262, 264, 266, 268 upon which the model outputs are based, to the external device 273. The external device 273 may receive such data for one or more patients, and may store the data for those patients for later viewing or analysis by the patient(s), physicians, or others, as necessary. In embodiments in which the external device 273 performs analysis for multiple patients, or for which the external device 273 receives from multiple processor devices 104 data of multiple patients, the external device 273 may store the received data 281 , the model outputs 274, and the feature values 272 for each patient separately in the memory.

[0114] Although the above description envisions various processing functions being carried out via the external device 273, it should be appreciated that, in some embodiments, at least some of the processing functions of the external device 273 (and/or the processor device 104) may be implemented via an implantable device, i.e., including on-board memory and processing capability that does not require connection to an external device. These functionalities may be disposed, for example, in the local processing device 144 as described above with respect to Figs. 2A-2I and Fig. 3.

[0115] MODEL TRAINING

[0116] Fig. 6A is a block diagram of an example system 310 for use in creating trained Al models (e.g., the trained Al models 302). The system 310 includes one or more sets 312Ar 312AN of data collection hardware that, in embodiments, are similar to the system 100 of Fig. 5. That is, each set of data collection hardware 312AI-312A_N includes a corresponding sensor array 102 (including electrode devices 110) and, optionally, one or more other sensors/equipment 108 and/or a user interface 106. The sensor array used in the data collection hardware from which training data are generated may include the sensor array described above with respect to Figs. 2A-2I in some embodiments, but in other embodiments may additionally or alternatively include traditional multi- or many-channel EEG sensor arrays that are not implanted (e.g., worn by the user over the scalp) and which are tethered to recording equipment. In embodiments, each of the sets 312AI-312A_N of data collection hardware also includes a respective processor device 104, including communication circuitry 256, a microprocessor 258, and a memory 260. Of course, some portion of the training data may originate from data collection hardware that is different from that depicted in Fig. 6A, including data collection hardware that does not include on-board processing devices. As in the systems 100, the memory 260 of each set 312AI-312A_N of data collection hardware stores at least the data 262. Each of the sets 312AI-312A_N of data collection hardware is associated with a corresponding patient A1 -AN and, accordingly, each of the sets 312AI-312AN of data collection hardware collects data for a corresponding patient.

[0117] At least some portion of the data collection hardware may not be necessary to collect at least some portions of training data for the models described herein, in some embodiments. For example, in embodiments, one or more models may be initially trained using at least some data that are collected, classified, and labeled and are archived and available for use and/or research generally. These publicly available data may be combined with data obtained using the sensor array 102 and/or other sensors/equipment 108 to further train the models 302, which may, in some instances, enable certain ones of the models 302 to more accurately.

[0118] In any event, unlike the systems 100, 300 depicted in Figs. 4A and 4B, however, the sets 312AI-312A_N of data collection hardware in the system 310 need not necessarily include the models 270, 302 stored in the memory 260, and the memory 260 need not necessarily store feature values 272 or model outputs 274. That is, the sets 312AI-312A_N of data collection hardware in the system 310 need not necessarily be capable of performing the evaluative functions, but may, in embodiments, merely act as collectors of, and conduits for, information to be used as “training data” for to create the trained Al models 302.

[0119] The data collected by the sets 312AI-312A_N of data collection hardware may be communicated to a modeling processor device 314. The modeling processor device 314 may be any computer workstation, laptop computer, mobile computing device, server, cloud computing environment, etc. that is configured to receive the data from the sets 312AI-312A_N of data collection hardware and to use the data from the sets 312AI-312A_N of data collection hardware to create the trained Al model 302. The modeling processor device 314 may receive the data from the sets 312AI-312A_N of data collection hardware via wired connection (e.g., Ethernet, serial connection, etc.) or wireless connection (e.g., mobile telephony, IEEE 802.11 protocol, etc.), directly (e.g., a connection with no intervening devices) or indirectly (e.g., a connection through one or more intermediary switches, access points, and/or the Internet), between the communication circuitry 256 of the processor device 104 and communication circuitry 316 of the modeling processor device 314. Additionally, though not depicted in Fig. 6A, the data may be communicated from one or more of the sets 312AI-312A_N of data collection hardware to the modeling processor device 314 via storage media, rather than by respective communication circuitry. The storage media may include any known storage memory type including, by way of example and not limitation, magnetic storage media, solid state storage media, secure digital (SD) memory cards, USB drives, and the like. [0120] The modeling processor device 314 includes the communication circuitry 316, in embodiments in which it is necessary, a microprocessor 318, and a memory device 320. Though it should be understood, the microprocessor 318 may be one or more stand-alone microprocessors, one or more shared computing resources or processor arrays (e.g., a bank of processors in a cloud computing device), one or more multi-core processors, one or more DSPs, one or more FPGAs, etc. Similarly, the memory device 320 may be volatile or nonvolatile memory, and may be memory dedicated solely to the modeling processor device 314 or shared among a variety of users, such as in a cloud computing environment.

[0121] The memory 320 of the modeling processor device 314 may store as a first one or more Al training sets 322 (depicted in Fig. 7A) the received data 262 (sensor data and optional other sensor/equipment data) received from each of the sets 312AI-312A_N of data collection hardware and, optionally, from publicly available training data 263. As depicted in Fig. 7A, the user report data 268 may include perceived events (e.g., epileptic/seizure events) 350; characteristics or features of perceived events 352 such as severity and/or duration, perceived effects on memory, or other effects on the individual’s well-being (such as their ability to hold a cup or operate a vehicle, etc.); other types of physiological symptoms (e.g., headaches, impaired or altered vision, involuntary movements, disorientation, falling to the ground, repeated jaw movements or lip smacking, etc.) 354; characteristics or features of other symptoms 356 (e.g., severity and/or duration); treatment (e.g., medication ingestion, neurostimulation application) information 358 (e.g., medication types, dosages, and/or frequencies/timing); perceived medication side-effects 360; characteristics or features of medication side effects 362 (e.g., severity and/or duration), and other user reported information 364 (e.g., food and/or drink ingested, activities performed (e.g., showering, exercising, working, brushing hair, etc.), tiredness, stress levels, etc.), as well as the timing of each. An adaptive learning component 324 may comprise instructions that are executable by the microprocessor 318 to implement supervised or unsupervised machine learning programs or algorithms, as described above. One or more data pre-processing routines 326, when executed by the microprocessor 318, may retrieve the data in the first Al training sets 322 (e.g., for respective models described herein), which may be raw recorded data, and may perform various pre-processing algorithms on the data in preparation for use of the data as training data by the adaptive learning component(s) 324. The pre-processing routines 326 may include routines for removing noisy data, cleaning data, reducing or removing irrelevant and/or redundant data, normalization, transformation, and extraction of biomarkers and other features. The pre-processing routines 326 may also include routines for detection of muscle activity in the electrical activity data and particularly in the EEG data by analyzing the spectral content of the signal and/or routines for selection of the channel or channels of the electrical activity data that have the best (or at least better, relatively) signal to noise ratios. The output of the pre-processing routines 326 and, optionally, in embodiments using image recognition of EEG waveform image data, of the image generation routine 276, is an additional training set stored in the memory 320 as a set 328 of feature values.

[0122] In embodiments in which the adaptive learning component 324 implements unsupervised learning algorithms, the adaptive learning component 324, executed by the microprocessor 318, finds its own structure in the unlabeled feature values 328 and, therefrom, generates a first trained Al model(s) 330.

[0123] In embodiments in which the adaptive learning component 324 implements supervised learning algorithms, the memory 320 may also store one or more classification routines 332 that facilitate the labeling of the feature values (e.g., by an expert, such as a neurologist, reviewing the feature values 328 and/or the first Al training sets 322) to create sets of key or label attributes 334, and/or feature values 328 provided for the publicly available training data 263. In embodiments in which EEG data are represented by waveform images, the classification routines 332 may facilitate selection of portions of EEG waveform images, for example. The adaptive learning component 324, executed by the microprocessor 318, may use both the feature values 328 and the key or label attributes 334 to discover rules, relationships, or other “models” that map the features to the labels by, for example, determining and/or assigning weights or other metrics. The adaptive learning component 324 may output the set of rules, relationships, or other models as a first trained Al model(s) 330.

[0124] Regardless of the manner in which the adaptive learning component 324 creates the first trained Al model(s) 330, the microprocessor 318 may use the first trained Al model(s) 330 with at least a portion of the first Al training set 322 and/or a portion of the feature values 328 extracted therefrom that were reserved for validating the first trained Al model(s) 330, in order to provide model outputs 336 for comparison and/or analysis by a trained professional in order to validate the output of the model.

[0125] As should be apparent, the first Al training set 322 may include data from one or more of the sets 312AI-312A_N of data collection hardware and, as a result, from one or more patients. Thus, the adaptive learning component 324 may use data from a single patient, from multiple patients, or from a multiplicity of patients when creating the first trained Al model(s) 330. The population from which the patient or patients are selected may be tailored according to particular demographic (e.g., a particular type of epilepsy, a particular age group, etc.), in some instances, or may be non-selective. In embodiments, at least some of the patients associated with the sets 312AI-312A_N of data collection hardware from which the first Al training sets 322 is created may be patients without any symptoms of the condition(s) in question and, as such, may serve to provide additional control data to the first Al training set 322.

[0126] In embodiments, the first trained Al model(s) 330 may be transmitted to (or otherwise received - e.g., via portable storage media) to another set of data collection hardware (e.g., the system 300 depicted in of Fig. 4B). The set of data collection hardware may implement the first trained Al model(s) 330 to provide model outputs 274 based on data that was not part of the first Al training sets 322 collected by the sets 312AI-312AN of data collection hardware or, alternatively, may simply collect additional data for use by the modeling processor device 314 to iterate the first trained Al model(s) 330.

[0127] Fig. 6B depicts such an embodiment. In Fig. 6B, a system 340 includes a set 342 of data collection hardware for a patient. Like hardware previously described, the set of 342 of data collection hardware includes the sensor array 102, optional other sensors/equipment 108 and/or the user interface 106, and the processor device 104. The processor device 104 includes the communication circuitry 256, the microprocessor 258, and the memory device 260. The memory device 260 has stored thereon at least the sensor array data and/or other sensor/equipment data 262. However, the memory 260 of the processor device 104 in the set 342 of data collection hardware optionally has stored thereon the first trained Al model 330. In such embodiments, the processor device 104 of the set 342 of data collection hardware may implement the data pre-processing routine 271 to extract feature values 272 and may provide associated model outputs 274.

[0128] Any or all of the data stored in the memory device 260 of the set 342 of data collection hardware may be communicated from the set 342 of data collection hardware to the modeling processor device 314. As above, the modeling processor device 314 may receive the data from the set 342 of data collection hardware via wired connection (e.g., Ethernet, serial connection, etc.) or wireless connection (e.g., mobile telephony, IEEE 802.11 protocol, etc.), directly (e.g., a connection with no intervening devices) or indirectly (e.g., a connection through one or more intermediary switches, access points, and/or the Internet), between the communication circuitry 256 of the processor device 104 and the communication circuitry 316 of the modeling processor device 314. Additionally, though not depicted in Fig. 6B, the data may be communicated from the set 342 of data collection hardware to the modeling processor device 314 via storage media, rather than by respective communication circuitry. The storage media may include any known storage memory type including, by way of example and not limitation, magnetic storage media, solid state storage media, secure digital (SD) memory cards, USB drives, and the like.

[0129] The received data may be stored in the memory 320 as second Al training sets 344 (depicted in Fig. 7B). The second Al training sets 344 (e.g., for respective models described herein) may include the sensor array data and/or other sensor/equipment data 262, user report data 268, and the other data 264 received from the set 342 of data collection hardware. As depicted in Fig. 7A, the user report data 268 may include perceived events (e.g., epileptic/seizure events) 350; characteristics or features of perceived events 352 such as severity and/or duration, perceived effects on memory, or other effects on the individual’s wellbeing (such as their ability to hold a cup or operate a vehicle, etc.); other types of physiological symptoms (e.g., headaches, impaired or altered vision, involuntary movements, disorientation, falling to the ground, repeated jaw movements or lip smacking, etc.) 354; characteristics or features of other symptoms 356 (e.g., severity and/or duration); treatment (e.g., medication ingestion, neurostimulation application) information 358 (e.g., medication types, dosages, and/or frequencies/timing); perceived medication side-effects 360; characteristics or features of medication side effects 362 (e.g., severity and/or duration), and other user reported information 364 (e.g., food and/or drink ingested, activities performed (e.g., showering, exercising, working, brushing hair, etc.), tiredness, stress levels, etc.), as well as the timing of each. The adaptive learning component 324 may comprise instructions that are executable by the microprocessor 318 to implement supervised or unsupervised machine learning programs or algorithms, as described above, for iterating the first trained Al model(s) 330, which may have a first error rate associated with its model outputs 336 (e.g., the results of the evaluative functions), to create a second trained Al model(s) 346, which may have a second error rate, reduced from the first error rate, associated its model outputs 348 (e.g., the results of the evaluative functions). The data pre-processing routines 326, when executed by the microprocessor 318, may retrieve the data in the second Al training set 344, which may be raw recorded data, and may perform various pre-processing algorithms on the data in preparation for use of the data as training data by the adaptive learning component 324. The pre-processing routines 326 may include routines for removing noisy data, cleaning data, reducing or removing irrelevant and/or redundant data, normalization, transformation, and extraction of biomarkers and other features. The preprocessing routines 326 may also include routines for detection of muscle activity in the electrical activity data and particularly in the EEG data by analyzing the spectral content of the signal and/or routines for selection of the channel or channels of the electrical activity data that have the best (or at least better, relatively) signal to noise ratios. The output of the preprocessing routines 326 is a final training set stored in the memory 320 as a set 328 of feature values.

[0130] In embodiments in which the adaptive learning component 324 implements unsupervised learning algorithms, the adaptive learning component 324, executed by the microprocessor 318, finds its own structure in the unlabeled feature values 328 and, therefrom, generates a second trained Al model(s) 346.

[0131] In embodiments in which the adaptive learning component 324 implements supervised learning algorithms, the memory 320 may also store one or more classification routines 332 that facilitate the labeling of the feature values (e.g., by an expert, such as a neurologist, reviewing the feature values 328 and/or the second Al training set 344) to create a set of key or label attributes 334. The adaptive learning component 324, executed by the microprocessor 318, may use both the feature values 328 and the key or label attributes 334 to discover rules, relationships, or other “models” that map the features to the labels by, for example, determining and/or assigning weights or other metrics. The adaptive learning component 324 may output an updated set of rules, relationships, or other models as a second trained Al model(s) 346.

[0132] Regardless of the manner in which the adaptive learning component 324 iterates and/or updates the first trained Al model 330(s) to be the second trained Al model(s) 346, the microprocessor 318 may use the second trained Al model(s) 346 with the second Al training sets 344 and/or the feature values 328 extracted therefrom, or on a portion of the second Al training sets 344 and/or a portion of the feature values 328 extracted therefrom that were reserved for validating the second trained Al model(s) 346, in order to provide model outputs 348 for comparison and/or analysis by a trained professional in order to validate the output of the model(s). An error rate of the model outputs 348 output by the second trained Al model(s) 346 will be reduced relative to an error rate of the model outputs 336 output by the first trained Al model 330. The second trained Al model(s) 346 may be programmed into or communicated to the system depicted, for example, in Fig. 4B, for use performing evaluative functions for patients.

[0133] The static models 270, and the trained Al models 302, may each be programmed to perform the evaluative functions by detecting within the received data (e.g., the data 262 and/or other data such as the user report data 268 and the other sensor data 264) relevant biomarkers forthe condition(s) of interest (e.g., epilepsy/seizure activity) and performing the evaluative functions based on the presence, absence, and/or temporal relationships between the relevant biomarkers. In various embodiments, certain ones of the models 270 and/or 302 may be programmed or trained to perform one or more of the following evaluative functions:

[0134] (1) detecting a seizure, including for example a time, duration, and/or level of confidence in the detection;

[0135] (2) classifying a seizure as epileptic or cardiac in origin;

[0136] (3) classifying a seizure as ictal hypoxemic or not;

[0137] (4) predicting a seizure event, including for example a time, duration, and/or level of confidence in the prediction;

[0138] (5) classifying a severity of a detected seizure event or likely severity of a predicted seizure event;

[0139] (6) determining a pre- or post-ictal impact of a seizure event on patient well-being;

[0140] (7) predicting a pre- or post-ictal impact of a seizure event on patient well-being (severity of the event, ictal cardiac changes; types of ictal respiratory changes);

[0141] (8) predicting a recovery time from post-ictal impacts of a seizure event on patient well-being; [0142] (9) detecting a hemisphere in which a detected seizure event originated;

[0143] (10) predicting a hemisphere in which a predicted seizure event will originate;

[0144] (11) determining that the foci of seizure activity are moving within a hemisphere or between hemispheres;

[0145] (12) determining, tracking, and/or predicting changes in seizure focus and/or EEG patterns as a result of medication and/or neurostimulation application;

[0146] (13) determining a therapeutic treatment that is to be administered to the patient, for example by identifying a type of treatment (e.g. drug or neurostimulation), a quantity, and/or a timing/duration of administration of the treatment;

[0147] (14) predicting an effect of an administered or recommended treatment, e.g. in terms of effect on likelihood of a seizure occurring at times following the administration of treatment.

[0148] OUTPUTS OF DETECTION, FORECASTING, TITRATION MODELS

[0149] Fig. 8 depicts sets of model outputs 370 which may correspond, for example, to the model outputs 274, 336, and/or 348 generated using the techniques of the present description. The model outputs 370 include detection model outputs 371 generally relating to detected seizure events, forecasting model outputs 372 generally relating to predicted seizure events, and titration model outputs 373 generally relating to administration of therapeutic treatment (e.g., drugs, neurostimulation, etc.) and the projected effects thereof. Outputs 371 , 372, and/or 373 may be generated, for example, via execution of static of Al-based models described in the foregoing sections of the present description.

[0150] The detection model outputs 371 can include, for example, outputs indicating the time 374 of a detected event, the location 375 (e.g., hemisphere or focus) of a detected event, the duration 376 of a detected event, the severity 377 of a detected event, and/or one or more confidence levels 378 for each detected event. Confidence levels 378 may, for example, represent the degree of certainty that a detected event was a seizure event (or a particular class of seizure), and/or the degree of certainty that other detection model outputs 371 for an event are correct (e.g. confidence in a lateralization of a detected event between the left and right hemispheres of the brain).

[0151] The forecasting model outputs 372 can include, for example, outputs indicating the predictions of the time 381 , location 382, duration 383, and/or severity 384 of a seizure during a future prediction window (e.g., the next one hour, six hours, one day, three days, one week, two weeks, one month, etc.). The forecasting model outputs 372 can additionally include one or more confidence levels 385 indicating a degree of certainty that a seizure event (or a particular type of event) will occur during the prediction window or portion thereof. In some instances, the confidence level(s) 385 may indicate degree of certainty of other aspects of a predicted event, such as the time 381 , location 382, duration 383, and/or severity 384. Collectively, the forecasting model outputs 372 may form a prediction curve over the prediction window, representing degree of likelihood of a seizure event at various times during the prediction window (e.g., at hourly intervals or at all times over a six hour, twelve hour, one day, three day, etc. window following the administration of a particular treatment(s)).

[0152] The titration model outputs 373 may include treatment administration outputs 390 indicating treatment to be administered to the patient, and/or treatment effect outputs 391 indicating predicted effects of an administered treatment(s). Treatment administration outputs 390 may, for example, identify a treatment type 393 (e.g., drug or neurostimulation), time and/or duration 394 of treatment (e.g., duration and timing of neurostimulation, or time at which a drug is to be administered via oral intake, a drug pump etc.), and/or quantity 395 (e.g. quantity of drug intake, and/or total or instantaneous intensity of neurostimulation). In some embodiments, outputs 390 of the titration model may be used to control one or more therapeutic devices to administer treatment according to the outputs 390. Titration model outputs 391 may additionally or alternatively include treatment effect outputs 391 , for example in the form of effect curves 397 representing predicted effects of treatment over a prediction window (e.g., representing diminishing protection against epileptic seizures in minutes, hours, days etc. following a treatment), and/or confidence levels 398 associated therewith.

[0153] In view of the description above, it will also be appreciated that the classification results may also include additional data not explicitly depicted in Fig. 8. By way of example and not limitation, depending on the embodiment, the classification results may include any of the data indicated as potential outputs described herein, such as the presence of multiple seizure foci, the movement of seizure foci between hemispheres in the brain, and associations among aspects of seizures including but not limited to: data associating specific symptoms to specific foci; data associating specific symptoms to specific epileptiforms; data associating specific epileptiforms to specific hemispheres; data associating specific symptoms with events occurring in specific hemispheres; data associating specific epileptiforms to specific foci; data associating and/or characterizing specific treatments (pharmacological and/or neurostimulatory) with specific foci; data associating and/or characterizing specific treatments (pharmacological and/or neurostimulatory) with specific hemispheres; data associating and/or characterizing specific treatments (pharmacological and/or neurostimulatory) with specific epileptiforms; multi-cycle seizure forecasting data (i.e., prediction data for different types - foci, hemispheres, epileptiforms - of seizures); and/or historical data relating to any of the previously mentioned data.

[0154] It should be understood that the system and, in particular, the adaptive learning component 324 (whether implemented in the separate modeling processor device 314, the data collection hardware 342, or even in the processor device 104 alongside the trained Al model 302), may be programmed to analyze the predicted event data (e.g., predicted seizure event data 372) relative to detected event data to determine the accuracy of the predictions made by the trained Al model 302. The results of the analysis may be used by the adaptive learning component 324 to further refine the trained Al model 302. Similarly, the adaptive learning component 324 may be programmed to analyze titration model outputs (e.g. indicating administered treatment and/or predicted effect) relative to detected and/or predicted event data, to determine the actual effectiveness of administered therapeutic treatment (e.g., in comparison to expected effect), to thereby enable further refinement of the titration model to deliver more effective treatment regimens to the patient.

[0155] DATA FLOW AMONG DETECTION, FORECASTING, TITRATION MODELS

[0156] Traditionally, a forecasting model may predict seizure events in the patient, for example, based upon historical EEG data of the patient, and/or based upon data indicating historical epileptic events of the patient (and/or, in some instances, other patients), which can be self-reported or derived from the historical EEG data. These techniques are described, for example, in U.S. Patent Application Publication No. 16/797,315, entitled “Electrode Device for Monitoring and/or Stimulating Activity in a Subject.” A detection model, traditionally, is programmed to detect seizure events in patient based upon monitored EEG data and/or other data indicative of symptoms of the patient (e.g., oxygen levels, accelerometer data, self-reported symptoms, etc.). Still traditionally, a titration model determines treatment to be administered to the patient, for example based upon detection of a seizure event or high likelihood of an impending seizure event, and/or otherwise at routine intervals as part of a modeled treatment schedule.

[0157] Therapeutic treatment can include neurostimulation, drug intake, and/or other therapies, including those described herein. Generally speaking, management of therapeutic treatment for a patient involves determining and adjusting quantities, timings, and durations of therapies so as to provide benefits to the condition(s) of the patient while avoiding adverse side effects to the extent possible. Fig. 9 illustrates the general concept that for a given condition being treated by application of a given therapy, there will be a dose of the therapy below which the therapy has no effect (i.e., a sub-therapeutic range of doses), a range of doses for which the therapy improves the condition of the patient (i.e., a therapeutic window), and a range of doses for which the therapy causes one or more side-effects, which range may overlap one or both of the therapeutic range and the sub-therapeutic range. Optimally, the range of doses for which the therapy causes side-effects, while it may overlap with a portion of the therapeutic window, will not overlap with the entirety of the therapeutic window, and will leave a portion of the therapeutic window as a “side-effect free therapeutic window,” as depicted in Fig. 9.

Accordingly, medium-to-long term management of therapy to the patient can involve iterative administrative and adjustment of treatment types, quantities, durations, timings, etc., for example to increase or decrease a quantity, spread the administration of a dose over greater or lesser time, administer a therapy at different times, or add and/or subtract forms of therapy. In view of Fig. 9, adjustments to therapy are generally made in view of patient responses indicating whether each therapy is effective and/or whether each therapy produces side effects. Specific algorithms associated with administering and adjusting therapies are described in further detail in U.S. Patent Application Publication No. 16/797,315.

[0158] As identified previously in this description, flaws exist in the traditional operation of forecasting, detection, and titration models. For one, traditional forecasting and detection models typically do not account for the historical and ongoing therapies (e.g., neurostimulation and drug intake) administered to the patient. Further, traditional models typically decouple the aspects of seizure forecasting and seizure detection, in the sense that detection of seizure events in the patient is not influenced by any previous forecasted seizure events for the patient.

[0159] In view of the deficiencies of traditional techniques described above, Fig. 10 depicts a flow diagram 400 demonstrating flow of data among elements including a detection model 410, a forecasting model 412, and a titration model 414. Each of the models 410, 412, and 414 may, generally speaking, be acquired, trained, and adjusted according to the techniques described in other portions of the present description. However, the interactions among the models 410, 412, and 414, as illustrated in Fig. 10, include additional flow of information among the models 410, 412, and 414 to produce improvements to seizure detection, seizure forecasting, and titration of therapeutic treatments to a patient.

[0160] In accordance with traditional techniques, the forecasting model 412 can be trained based (at least) upon historical EEG data 422 and historical seizure events 424 of the patient (and/or, in some instances, other patients). Additionally, as with traditional methods, the result of operation of the models 410, 412, and 414 can include administration of therapy such as neurostimulation 428 and/or drug intake 430 to the patient, for example based upon detected seizures as indicated by input 432 to the titration model 414 from the detection model 410. The flow diagram 400, however, illustrates provision of additional inputs and interactions among the models 410, 412, 414 beyond what has been provided according to traditional techniques.

[0161] For one, as illustrated in Fig. 10, the forecasting model 412 receives additional input 432 (e.g., one or more input signals) indicating historical treatment applied to the patient. The historical treatment can include historical neurostimulation 436 and/or historical drug intake 438 by the patient. Thus, by training and operating the same forecasting model while using the additional input 434, the forecasting model 412 provides for prediction of seizure events based upon the actual history of therapeutic treatment provided to the patient.

[0162] Additionally, as illustrated in Fig. 10, the forecasting model 412 may receive still additional inputs 442, 444 from the detection model 410 and the titration model 414, respectively (i.e., corresponding to outputs and/or intermediate data generated by the detection model 410 and the titration model 414). Particularly, inputs 442 to the forecasting model 412 from the detection model 410 may correspond to any of the outputs 371 from Fig. 8 relating to one or more seizure events experienced by the patient, for example during a prediction window previously forecasted by the forecasting model 412, and/or shortly prior to a prediction window for which the forecasting model 412 is to generate predictions. Using the feedback 442, the forecasting model 412 may, for example, validate the predictions of seizure events in the patient against actual seizure events experienced (or not experienced) by the patient, to enable iterative re-tuning of the forecasting model 412 according to validations. Additionally or alternatively, the inputs 442 may influence the predictions generated by the forecasting model 442 for time windows following seizure activity in the patient.

[0163] Input 444 to the forecasting model 412 from the titration model 444 may include any of the outputs 373 from Fig. 8, e.g., indicating therapeutic treatment administered to the patient and/or updates treatment schedules indicating future administration of therapeutic treatment issued to the patient. Inputs 444 may also represent expected conditions of the patient based upon administered therapeutic treatment. For example, with reference to the therapeutic window established in Fig. 9, the supplementary inputs 444 provided to the forecasting model 412 may cause the forecasting model 412 to predict less likelihood of seizure events (or, perhaps, predict less severe events) in the patient during the time corresponding to a peak effect of an administered treatment, but comparatively greater likelihood of seizure events (or more severe events) in the times before and/or after peak effect is reached.

[0164] Effectively, by providing updated, additional inputs to the forecasting model 412, the techniques described with respect to Fig. 10 improve the operation of the forecasting model (e.g., the initial training and the updating of the forecasting model), as compared to traditional techniques not using the inputs 434, 442, and/or 444.

[0165] Still with reference to the flow diagram 400 in Fig. 10, just as the forecasting model 412 receives inputs 442 produced via the detection model 410, the detection model 410 as envisioned herein receives inputs 448 from the forecasting model 412 to influence the detection model 410. Specifically, the inputs 448 correspond to outputs of the forecasting model 412 indicating predicted seizure events during a prediction window (e.g., any of the outputs 371 from Fig. 8). The inputs 448, when provided to the detection model 410, cause the detection model 410 to be tuned to detect seizure events during a prediction window according to the predictions generated by the forecasting model 412 overthe same (or at least overlapping) prediction window. For example, when the inputs 448 indicate greater likelihood of seizure events during the prediction window, the inputs 448 may cause the detection model to lower a confidence threshold associated with detection of seizure events, effectively making the detection model 410 more sensitive to detecting and reporting potential seizure events. That is, the detection model 410 may not require as great a confidence level as otherwise would be required to generate a detection of a seizure event. Conversely, when the inputs 448 indicate less likelihood of seizure events during the prediction window, the outputs 448 may cause the detection model to raise the confidence threshold, effectively making the detection model less sensitive to detecting and reporting potential seizure events. In effect, the detection model 410 accounts for intelligent predictions of seizure events in the patient when generating detections, thereby improving accuracy in outputs of the detection model 410.

[0166] Even further to the improvements to the performance of the models 410, 412, and 414 described above, still further improvements are achieved through the second-order and third- order effects of exchange of inputs/outputs among the models 410, 412, and 414. For example, the titration model 410 provides the inputs 444 that enable the forecasting model to more accurately produce time-variant predictions of likelihood of seizure events based upon the projected time-variant effects over the prediction window from administered therapies. These improved predictions by the forecasting model 412, provided as inputs 448 to the detection model 410, enable the detection model to more accurately produce detections of seizure events in the patient. Accurate outputs of the detection model 410, provided as inputs 432 to the titration model 414, enable the titration model 414 to more effectively model the effectiveness of previous treatments in mitigating seizure events and more appropriately administer treatment to the patient upon detection of seizure events. The models 410, 412, and 414, interacting with each other in this manner, effectively lean upon each other to allow each model to more accurately represent the conditions and biology of the patient.

[0167] EXAMPLE FLOW DIAGRAMS

[0168] In view of the foregoing discussion of Fig. 10, Figs. 11A and 11 B illustrate flow diagrams of methods 510, 530 associated with operation of the detection model, forecasting model, and titration model. Actions of the methods 510, 530 can be implemented via various structures and techniques described in other portions of the present description. For example, actions of the methods 510, 530 may be implemented via the processing device 104 connected to the sensor array 102, and/or via the external device 273. The methods 510, 530 should not be understood to be limiting, as one should appreciate from the present description that actions of the methods 510, 530 can be combined with each other and/or with other actions/techniques described in the present description. It should be appreciated that actions described with respect to Figs. 11 A and 11 B can be performed via one or more devices implanted in the body of the patient (e.g., under the scalp), one or more devices externally worn by the patient, one or more other external devices, and/or combinations thereof, in accordance with the various structural embodiments described herein. [0169] Referring first to Fig. 11 A, the method 510 includes obtaining the forecasting model configured to predict seizure events in the patient (512, e.g. by training a static or Al-based forecasting model via techniques of the present description, particularly using at least historical EEG data of the patient and/or other patients). The method 510 further includes obtaining titration data indicating therapy administered to the patient (514, e.g. indicating neurostimulation/drug intake as described herein), from which further input data for the forecasting model is generated (516). The titration input data is provided as further input to the forecasting model to cause the forecasting model to predict seizure events in the patient over a prediction window (518). The forecasting model may, for example, predict severity or location of seizures. The predictions (e.g., likelihoods of seizures or of severe seizures) may vary over the prediction window, for example based upon anticipated changes in degree of effect of the administered therapy throughout the prediction window.

[0170] In embodiments, the predictions generated by the forecasting model are provided as further inputs to the detection model to detect seizure events (520, e.g. to adjust severity or confidence thresholds based upon the degree to which a seizure is expected to occur during the prediction window). Based upon outputs of the detection model, feedback may be provided back to the forecasting model (522), e.g. to validate/tune the forecasting model and/or to generate future predictions. Additionally or alternatively, outputs of the detection model may be provided to the titration model to administer treatment or adjust a longer-term treatment regimen for the patient (524).

[0171] Referring next to Fig. 11 B, the method 530 includes obtaining the detection model configured to detect seizure events in the patient based at least upon patient EEG data (532, e.g. by training a static or Al-based detection model via techniques of the present description). The method 530 further includes obtaining prediction data indicating likelihood of seizure events in the patient over a prediction window (534, e.g. provided via the forecasting model as described herein). The method 530 further includes generating further input data for the detection model using the prediction data (536). Still further, the method 530 includes providing the prediction data as further input data to the detection model (538), e.g., to cause the detection model to adjust severity thresholds or confidence thresholds associated with detecting seizure events during an EEG monitoring window that at least partially corresponds to the prediction window (e.g., to influence the detection model at corresponding times within the EEG monitoring window). The detection model may detect one or more seizure events experienced by the patient during the EEG monitoring window, and may for example output a time, severity, duration, seizure type, or confidence level(s) associated therewith.

[0172] In embodiments, the method 500 includes causing therapeutic treatment to be administered to the patient based upon the detected seizure event(s), and/or causing adjustment to the titration model based upon the detected seizure event(s) (540). In these embodiments, titration data reflecting the administration of treatment (e.g., administered treatment or scheduled treatment) may be provided back to the forecasting model (542), to enable the forecasting model to act upon the titration data as described for example with respect to Fig. 11 A. Additionally or alternatively, in embodiments, output of the detection model is provided as still further input to the forecasting model (544), for example to validate/tune the forecasting model and/or to influence subsequent prediction of seizure events.

[0173] EXAMPLE HARDWARE AND COMMUNICATION STRUCTURES

[0174] As may by now be understood, the presently disclosed method and system are amenable to a variety of embodiments, many of which have already been explicitly described, with reference to Figs. 12-15C. The communication arrangements that will be described with respect to Figs. 12-15C are also amenable to a variety of structural arrangements, including those that have been described herein. For example, the communication arrangements of Figs. 12-15C may be used regardless of whether the sensor array 102 is implanted under the scalp of the patient or worn externally by the patient, and regardless of which processing functions described herein are chosen to be performed via implantable devices and which are chosen to be performed via wearable devices and/or other devices external to the patient.

[0175] As described throughout the specification, the system 100 includes an EEG sensor array 102. In various embodiments, the EEG sensor array 102 may be separate from, but communicatively coupled to, the processor device 104, as depicted in Fig. 12. Fig. 13 depicts an embodiment in which the processor device 104 is integrated with the EEG sensor 102. The local processor device 104 may be communicatively coupled to external equipment 672, which may be one or more of the modeling processor device 314, the external device 273, etc.

[0176] Various communication schemes are contemplated, as well. Figs. 14A and 14B illustrate possible communication schemes between the sensor array 102 and the processor device 104 and, in particular, Fig. 14A illustrates a wireless connection 682 between the sensor array 102 and the processor device 104 (i.e., between the communication circuitry 150 of the sensor array 102 and the communication circuitry 256 of the processor device 104). The wireless connection 682 may be any known type of wireless connection, including a Bluetooth® connection (e.g., low-energy Bluetooth), a wireless internet connect (e.g., IEEE 802.11 , known as “WiFi”), a near-field communication connection, or similar. Fig. 14B illustrates a wired connection 684 between the sensor array 102 and the processor device 104. The wired connection 674 may be a serial connection, for example.

[0177] The sensor array 102 may communicate data to the processor device 104 as the data are acquired or periodically. For example, the sensor array 102 may store, in the memory 156 of the local processing unit 144, data as it is acquired from the electrode devices 110 that are part of the sensor array 102 and may periodically (e.g., every second, every minute, every half hour, every hour, every day, when the memory 156 is full, etc.) transmit the data to the processor device 104. In other embodiments, the sensor array 102 may store data until the processor device 104 is coupled to the respective device (e.g., via wireless or wired connection). The sensor array 102 may also store the data until the processor device 104 requests the transmission of the data from the respective deice to the processor device 104. In these manners, the sensor array 102 may be optimized, for example, to preserve battery life, etc.

[0178] Figs. 15A-15C illustrate possible communication schemes between the processor device 104 and external equipment or servers 672, regardless of whether or not the processor device 104 is integrated with the sensor array 102 (e.g., as in Fig. 12). In Fig. 15A, for example, the processor device 104 may be coupled by a wireless communication connection to a mobile device 686, such as a mobile telephony device, which may, in turn, be coupled to the external equipment 672 by, for example, the Internet. In Fig. 15B, the processor device 104 is coupled to one or more intermediary devices 688 (e.g., a mobile telephony base station, a wireless router, etc.), which in turn provides connectivity to the external equipment 672 via the Internet. In Fig. 15C, the processor device 104 is itself a mobile device, such as a mobile telephony device, which may be coupled by one or more intermediary devices 688 to the external equipment 672 by way of the Internet.

[0179] The following list of aspects reflects a variety of the embodiments explicitly contemplated by the present disclosure. Those of ordinary skill in the art will readily appreciate that the aspects below are neither limiting of the embodiments disclosed herein, nor exhaustive of all of the embodiments conceivable from the disclosure above, but are instead meant to be exemplary in nature.

[0180] 1 . A computer-implemented method implemented via one or more processors, the method comprising: obtaining a forecasting model configured to generate predictions of likelihood of epileptic seizure events in a patient based at least upon first input data indicative of historical electroencephalogram (EEG) signal data of the patient; obtaining first titration data for the patient, the first titration data indicating administration of therapeutic treatment to the patient responsive to an epileptic condition or an epileptic seizure event experienced by the patient; generating second input data for the forecasting model based upon the first titration data; and providing the second input data as further input to the forecasting model to cause the forecasting model to generate first prediction data indicating a likelihood of an epileptic seizure event in the patient over a first prediction window based upon the first and second input data.

[0181] 2. The computer-implemented method of aspect 1 , wherein the administration of therapeutic treatment comprises administration of neurostimulation to the patient. [0182] 3. The computer-implemented method of aspect 1 or aspect 2, wherein the administration of therapeutic treatment comprises administration of pharmacological treatment to the patient.

[0183] 4. The computer-implemented method of any one of aspects 1 to 3, wherein the first prediction data indicates varying likelihoods of the epileptic seizure event in the patient over times within the first prediction window.

[0184] 5. The computer-implemented method of any one of aspects 1 to 4, wherein the first predication data indicates a predicted hemisphere or focus of the epileptic seizure event over the first prediction window.

[0185] 6. The computer-implemented method of any one of aspects 1 to 5, further comprising implementing a seizure detection model configured to produce first detection data indicating one or more epileptic seizure events in the patient based upon monitored EEG data of the patient over the first prediction window.

[0186] 7. The computer-implemented method of aspect 6, wherein the monitored EEG data comprises EEG data obtained via a sensor array implanted under a scalp of the patient.

[0187] 8. The computer-implemented method of aspect 6, wherein the monitored EEG data comprises EEG data obtained via one or more sensors in one or more wearable devices affixed externally to the patient.

[0188] 9. The computer-implemented method of any one of aspects 6 to 8, wherein implementing the detection model over the first prediction window comprises: providing the first prediction data as input to the detection model to cause the detection model to adjust a seizure severity threshold or a confidence threshold associated with detection of epileptic seizure events during the first prediction window; and generating the first detection data indicating one or more epileptic seizure events occurring during the first prediction window, based upon the monitored EEG data and the first prediction data.

[0189] 10. The computer-implemented method of any one of aspects 6 to 9, wherein the first detection data indicates respective severity or confidence measurements associated with each of the one or more epileptic seizure events.

[0190] 11 . The computer-implemented method of any one of aspects 6 to 10, further comprising: comparing the first detection data to the first prediction data; and validating or tuning the forecasting model based upon the comparing of the first detection data to the first prediction data.

[0191] 12. The computer-implemented method of any one of aspects 6 to 11 , further comprising providing the first detection data as input to the forecasting model to generate subsequent prediction data indicating a likelihood of epileptic seizure events in the patient over a subsequent prediction window.

[0192] 13. The computer-implemented method of any one of aspects 6 to 12, further comprising providing the first detection data as input to a titration model to generate a treatment schedule comprising administration of first therapeutic treatment of at least one of neurostimulation or pharmacological treatment to the patient.

[0193] 14. The computer-implemented method of aspect 13, further comprising causing the first therapeutic treatment to be administered to the patient in accordance with the treatment schedule.

[0194] 15. The computer-implemented method of aspect 14, wherein causing first therapeutic treatment to be administered to the patient comprises controlling a neurostimulation device affixed to or implanted in the patient, to cause the neurostimulation device to administer neurostimulation to the patient.

[0195] 16. The computer-implemented method of aspect 15, wherein controlling the neurostimulation device comprises transmitting signal communications to the neurostimulation device via Bluetooth communications.

[0196] 17. The computer-implemented method of any one of aspects 14 to 16, wherein causing the first therapeutic treatment to be administered to the patient comprises controlling a drug pump to cause the drug pump to administer a pharmacological treatment to the patient.

[0197] 18. The computer-implemented method of any one of aspects 14 to 17, further comprising: receiving an indication that the first therapeutic treatment was administered to the patient; generating, via the titration model, second titration data indicating the administration of the first therapeutic treatment to the patient; and providing the second titration data as input to the forecasting model to cause the forecasting model to generate second prediction data indicating a likelihood of an epileptic seizure event in the patient over a second prediction window subsequent to the first prediction window.

[0198] 19. The computer-implemented method of aspect 18, wherein the second titration data indicates a predicted variance in effect of the first therapeutic treatment over the second prediction window.

[0199] 20. The computer-implemented method of any one of aspects 1 to 19, wherein obtaining the forecasting model comprises training the forecasting model to generate predictions of epileptic seizure events in the patient based upon historical training data comprising the first input data.

[0200] 21 . The computer-implemented method of any one of aspects 1 to 20, wherein at least one of the obtaining the forecasting model, obtaining the first titration data, generating the second input data, or providing the second input data to the forecasting model is performed via one or more computing devices implanted under the scalp of the patient.

[0201] 22. The computer-implemented method of aspect 21 , wherein the obtaining the forecasting model is performed via the one or more computing devices implanted under the scalp of the patient.

[0202] 23. The computer-implemented method of aspect 21 or aspect 22, wherein the obtaining the first titration data is performed via the one or more computing devices implanted under the scalp of the patient.

[0203] 24. The computer-implemented method of any one of aspects 21 to 23, wherein the generating the second input data is performed via the one or more computing devices implanted under the scalp of the patient.

[0204] 25. The computer-implemented method of any one of aspects 21 to 24, wherein the providing the second input data to the forecasting model is performed via the one or more computing devices implanted under the scalp of the patient.

[0205] 26. The computer-implemented method of any one of aspects 1 to 20, wherein at least one of the obtaining the forecasting model, obtaining the first titration data, generating the second input data, or providing the second input data to the forecasting model is performed via one or more computing devices external to the patient.

[0206] 27. The computer-implemented method of aspect 26, wherein the obtaining the forecasting model is performed via the one or more computing devices external to the patient.

[0207] 28. The computer-implemented method of aspect 26 or aspect 27, wherein the obtaining the first titration data is performed via the one or more computing devices external to the patient.

[0208] 29. The computer-implemented method of any one of aspects 26 to 28, wherein the generating the second input data is performed via the one or more computing devices external to the patient.

[0209] 30. The computer-implemented method of any one of aspects 26 to 29, wherein the providing the second input data to the forecasting model is performed via the one or more computing devices external to the patient.

[0210] 31. The computer-implemented method of any one of aspects 1 to 30, wherein any one or more of the positively-recited actions are performed via one or more computing devices implanted under the scalp of the patient. [0211] 32. The computer-implemented method of any one of aspects 1 to 30, wherein any one or more of the positively-recited actions are performed via one or more computing devices external to the patient.

[0212] 33. The computer-implemented method of any one of aspects 1 to 32, combined with any other suitable one of aspects 1 to 32.

[0213] 34. A system for predicting seizure events in a patient, the system comprising: an electrode array comprising: a reference electrode; a plurality of sensing electrodes spaced linearly along a lead of the electrode array and configured, collectively, to measure EEG signals of a brain of the patient; a processing unit, the processing unit comprising: a memory device; a processor communicatively coupled to the electrodes and receiving the EEG signals from plurality of sensing electrodes, the processor configured to store to the memory device data of the received EEG signals; a transceiver coupled to the processor and configured to transmit data to, and receive data from, an external computing device; and one or more computerexecutable routines executing on the external computing device and configured to: receive the EEG signals as first input data to a forecasting model; obtain first titration data for the patient, the first titration data indicating administration of therapeutic treatment to the patient; provide second input data to the forecasting model based upon the first titration data to cause the forecasting model to generate first prediction data indicating a likelihood of an epileptic seizure event in the patient over a first prediction window based upon the first and second input data.

[0214] 35. The system of aspect 34, wherein the administration of therapeutic treatment comprises administration of neurostimulation to the patient.

[0215] 36. The system of aspect 34 or aspect 35, wherein the administration of therapeutic treatment comprises administration of pharmacological treatment to the patient.

[0216] 37. The system of any one of aspects 35 to 37, wherein the first prediction data indicates varying likelihoods of the epileptic seizure event in the patient over times within the first prediction window.

[0217] 38. The system of any one of aspects 34 to 37, wherein the first predication data indicates a predicted hemisphere or focus of the epileptic seizure event over the first prediction window.

[0218] 39. The system of any one of aspects 34 to 38, wherein the one or more computerexecutable routines are further configured to implement a seizure detection model to produce first detection data indicating one or more epileptic seizure events in the patient based upon monitored EEG data of the patient over the first prediction window, the monitored EEG data being monitored via the electrode array. [0219] 40. The system of aspect 39, wherein the electrode array is implanted under a scalp of the patient.

[0220] 41 . The system of aspect 40, wherein the electrode array is disposed in one or more wearable devices affixed externally to the patient.

[0221] 42. The system of any one of aspects 39 to 41 , wherein the one or more computerexecutable routines configured to implement the seizure detection model over the first prediction window are configured to: provide the first prediction data as input to the detection model to cause the detection model to adjust a seizure severity threshold or a confidence threshold associated with detection of epileptic seizure events during the first prediction window; and generate the first detection data indicating one or more epileptic seizure events occurring during the first prediction window, based upon the monitored EEG data and the first prediction data.

[0222] 43. The system of any one of aspects 39 to 42, wherein the first detection data indicates respective severity or confidence measurements associated with each of the one or more epileptic seizure events.

[0223] 44. The system of any one of aspects 39 to 43, wherein the one or more computerexecutable routines are further configured to: compare the first detection data to the first prediction data; and validate or tune the forecasting model based upon the comparing of the first detection data to the first prediction data.

[0224] 45. The system of any one of aspects 39 to 44, wherein the one or more computerexecutable routines are further configured to provide the first detection data as input to the forecasting model to generate subsequent prediction data indicating a likelihood of epileptic seizure events in the patient over a subsequent prediction window.

[0225] 46. The system of any one of aspects 39 to 45, wherein the one or more computerexecutable routines are further configured to provide the first detection data as input to a titration model to generate a treatment schedule comprising administration of first therapeutic treatment of at least one of neurostimulation or pharmacological treatment to the patient.

[0226] 47. The system of aspect 46, wherein the one or more computer-executable routines are further configured to cause the first therapeutic treatment to be administered to the patient in accordance with the treatment schedule.

[0227] 48. The system of aspect 47, wherein the one or more computer-executable routines configured to cause first therapeutic treatment to be administered to the patient are configured to control a neurostimulation device affixed to or implanted in the patient, to cause the neurostimulation device to administer neurostimulation to the patient. [0228] 49. The system of aspect 48, wherein controlling the neurostimulation device comprises transmitting signal communications to the neurostimulation device via Bluetooth communications.

[0229] 50. The system of any one of aspects 47 to 49, wherein the one or more computerexecutable routines configured to cause first therapeutic treatment to be administered to the patient are configured to control a drug pump to cause the drug pump to administer a pharmacological treatment to the patient.

[0230] 51 . The system of any one of aspects 47 to 50, wherein the one or more computerexecutable routines are further configured to: receive an indication that the first therapeutic treatment was administered to the patient; generate, via the titration model, second titration data indicating the administration of the first therapeutic treatment to the patient; and provide the second titration data as input to the forecasting model to cause the forecasting model to generate second prediction data indicating a likelihood of an epileptic seizure event in the patient over a second prediction window subsequent to the first prediction window.

[0231] 52. The system of aspect 51 , wherein the second titration data indicates a predicted variance in effect of the first therapeutic treatment over the second prediction window.

[0232] 53. The system of any one of aspects 34 to 52, wherein the one or more computerexecutable routines configured to obtain the forecasting model are configured to train the forecasting model to generate predictions of epileptic seizure events in the patient based upon historical training data comprising the first input data.

[0233] 54. The system of any one of aspects 34 to 53, configured to perform the method of any suitable one of aspects 1 to 33.

[0234] 55. The system of any one of aspects 34 to 54, in combination with any other suitable one of aspects 34 to 54.

[0235] 56. A computer-implemented method implemented via one or more processors, the method comprising: obtaining a seizure event detection model configured to generate detection data indicating epileptic seizure events experienced by a patient based upon first input data indicative of electroencephalogram (EEG) signal data of the patient over a first EEG monitoring window; obtaining first prediction data for the patient, the first prediction data indicating likelihood of one or more epileptic seizure events in the patient over a first prediction window matching at least a portion of the first EEG monitoring window, the first prediction data being generated based upon titration data indicating therapeutic treatment administered to the patient; generating second input data for the detection model based upon the first prediction data; and providing the second input data as further input to the detection model to cause the detection model to generate first detection data indicating an epileptic seizure event experienced by the patient over the first EEG monitoring window based upon the first and second input data.

[0236] 57. The computer-implemented method of aspect 56, wherein providing the second input data as further input to the detection model causes the detection model to adjust a seizure severity threshold or a confidence threshold associated with detection of the epileptic seizure event during the first EEG monitoring window.

[0237] 58. The computer-implemented method of aspect 56 or aspect 57, wherein the first detection data comprises one or more confidence levels indicating a degree of certainty that a detected epileptic seizure event occurred during the first EEG monitoring window, or a degree of certainty of one or more parameters of the detected epileptic seizure event.

[0238] 59. The computer-implemented method of any one of aspects 56 to 58, wherein the first detection data indicates a severity of a detected epileptic seizure event during the first EEG monitoring window.

[0239] 60. The computer-implemented method of any one of aspects 56 to 58, wherein the first detection data indicates a duration of a detected epileptic seizure event during the first EEG monitoring window.

[0240] 61 . The computer-implemented method of any one of aspects 56 to 60, wherein the first detection data indicates a hemisphere or focus of a detected epileptic seizure event during the first EEG monitoring window.

[0241] 62. The computer-implemented method of any one of aspects 56 to 61 , further comprising causing a therapeutic treatment to be administered to the patient responsive to the detected epileptic seizure event.

[0242] 63. The computer-implemented method of aspect 62, wherein causing the therapeutic treatment to be administered to the patient comprises causing neurostimulation to be administered to the patient via a neurostimulation device affixed to or implanted in the patient.

[0243] 64. The computer-implemented method of aspect 63, wherein causing the neurostimulation to be administered to the patient comprises controlling the neurostimulation device via transmitting Bluetooth signals to the neurostimulation device.

[0244] 65. The computer-implemented method of any one of aspects 62 to 64, wherein causing the therapeutic treatment to be administered to the patient comprises causing a pharmacological treatment to be administered to the patient via a drug pump.

[0245] 66. The computer-implemented method of any one of aspects 56 to 65, wherein the first prediction data is generated via a forecasting model, and the method further comprising: providing the first detection data as feedback to the forecasting model; comparing the first detection data to the first prediction data; and validating or tuning the forecasting model based upon the comparing of the first detection data to the first prediction data.

[0246] 67. The computer-implemented method of any one of aspects 56 to 66, further comprising providing the first detection data as input to a titration model to cause the titration model to adjust a therapeutic treatment schedule for the patient.

[0247] 68. The computer-implemented method of aspect 67, wherein the therapeutic treatment schedule comprises scheduled administrations of neurostimulation to the patient via the neurostimulator device.

[0248] 69. The computer-implemented method of aspect 67 or aspect 68, wherein the therapeutic treatment schedule comprises scheduled administrations of pharmacological treatment to the patient via a drug pump.

[0249] 70. The computer-implemented method of any one of aspects 56 to 69, further comprising: generating further titration data indicating a therapeutic treatment administered to the patient; and providing the further titration data as input to a forecasting model to cause the forecasting model to generate second prediction data indicating a likelihood of an epileptic seizure event in the patient over a second prediction window subsequent to the first prediction window, based upon the detected epileptic seizure event and the further titration data.

[0250] 71 . The computer-implemented method of aspect 70, wherein the titration data indicates a predicted variance in effect of the administered therapeutic treatment over the second prediction window.

[0251] 72. The computer-implemented method of any one of aspects 56 to 71 , wherein obtaining the detection model comprises training the detection model to generate detection data indicating epileptic seizure events in the patient based upon historical training data including historical EEG data of the patient.

[0252] 73. The computer-implemented method of any one of aspects 56 to 72, comprising receiving the first input data by monitoring the EEG signal data via a sensor array implanted under a scalp of the patient.

[0253] 74. The computer-implemented method of any one of aspects 56 to 73, comprising receiving the first input data by monitoring the EEG signal data via one or more sensors in one or more wearable devices affixed externally to the patient.

[0254] 75. The computer-implemented method of any one of aspects 56 to 74, wherein at least one of the obtaining the seizure event detection model, obtaining the first prediction data, generating the second input data, or providing the second input data to the detection model is performed via one or more computing devices implanted under the scalp of the patient. [0255] 76. The computer-implemented method of aspect 75, wherein the obtaining the seizure event detection model is performed via the one or more computing devices implanted under the scalp of the patient.

[0256] 77. The computer-implemented method of aspect 75 or aspect 76, wherein the obtaining the first prediction data is performed via the one or more computing devices implanted under the scalp of the patient.

[0257] 78. The computer-implemented method of any one of aspects 75 to 77, wherein the generating the second input data is performed via the one or more computing devices implanted under the scalp of the patient.

[0258] 79. The computer-implemented method of any one of aspects 75 to 78, wherein the providing the second input data to the detection model is performed via the one or more computing devices implanted under the scalp of the patient.

[0259] 80. The computer-implemented method of any one of aspects 56 to 75, wherein at least one of the obtaining the seizure event detection model, obtaining the first prediction data, generating the second input data, or providing the second input data to the detection model is performed via one or more computing devices external to the patient.

[0260] 81 . The computer-implemented method of aspect 80, wherein the obtaining the seizure event detection model is performed via the one or more computing devices external to the patient.

[0261] 82. The computer-implemented method of aspect 80 or aspect 81 , wherein the obtaining the first prediction data is performed via the one or more computing devices external to the patient.

[0262] 83. The computer-implemented method of any one of aspects 80 to 82, wherein the generating the second input data is performed via the one or more computing devices external to the patient.

[0263] 84. The computer-implemented method of any one of aspects 80 to 83, wherein the providing the second input data to the detection model is performed via the one or more computing devices external to the patient.

[0264] 85. The computer-implemented method of any one of aspects 56 to 75, wherein any one or more of the positively-recited actions are performed via one or more computing devices implanted under the scalp of the patient.

[0265] 86. The computer-implemented method of any one of aspects 56 to 75, wherein any one or more of the positively-recited actions are performed via one or more computing devices external to the patient [0266] 87. The computer-implemented method of any one of aspects 56 to 86, in combination with any suitable one of aspects 1 to 33.

[0267] 88. The computer-implemented method of any one of aspects 56 to 87, performed via the system of any suitable one of aspects 34 to 55.

[0268] 89. The computer-implemented method of any one of aspects 56 to 88, in combination with any other suitable one of aspects 56 to 88.

[0269] 90. A system for predicting seizure events in a patient, the system comprising: an electrode array comprising: a reference electrode; a plurality of sensing electrodes spaced linearly along a lead of the electrode array and configured, collectively, to measure EEG signals of a brain of the patient; a processing unit, the processing unit comprising: a memory device; a processor communicatively coupled to the electrodes and receiving the EEG signals from plurality of sensing electrodes, the processor configured to store to the memory device data of the received EEG signals; a transceiver coupled to the processor and configured to transmit data to, and receive data from, an external computing device; and one or more computerexecutable routines executing on the external computing device and configured to: receive the EEG signals as first input to a seizure event detection model configured to generate detection data indicating epileptic seizure events experienced by the patient over a first EEG monitoring window; obtain first prediction data for the patient, first prediction data indicating likelihood of one or more epileptic seizure events in the patient over a first prediction window matching at least a portion of the first EEG monitoring window; and provide second input data to the detection model based upon the first prediction data to cause the detection model to generate first detection data indicating an epileptic seizure event experienced by the patient over the first EEG monitoring window based upon the first and second input data.

[0270] 91 . The system of aspect 90, wherein the providing of the second input data to the detection model causes the detection model to adjust a seizure severity threshold or a confidence threshold associated with detection of the epileptic seizure event during the first EEG monitoring window.

[0271] 92. The system of aspect 90 or aspect 91 , wherein the first detection data comprises one or more confidence levels indicating a degree of certainty that a detected epileptic seizure event occurred during the first EEG monitoring window, or a degree of certainty of one or more parameters of the detected epileptic seizure event.

[0272] 93. The system of any one of aspects 90 to 92, wherein the first detection data indicates a severity of a detected epileptic seizure event during the first EEG monitoring window. [0273] 94. The system of any one of aspects 90 to 93, wherein the first detection data indicates a duration of a detected epileptic seizure event during the first EEG monitoring window.

[0274] 95. The system of any one of aspects 90 to 94, wherein the first detection data indicates a hemisphere or focus of a detected epileptic seizure event during the first EEG monitoring window.

[0275] 96. The system of any one of aspects 90 to 95, wherein the one or more computerexecutable routines are further configured to cause a therapeutic treatment to be administered to the patient responsive to the detected epileptic seizure event.

[0276] 97. The system of aspect 96, wherein causing the therapeutic treatment to be administered to the patient comprises causing neurostimulation to be administered to the patient via a neurostimulation device affixed to or implanted in the patient.

[0277] 98. The system of aspect 97, wherein causing the neurostimulation to be administered to the patient comprises controlling the neurostimulation device via transmitting Bluetooth signals to the neurostimulation device.

[0278] 99. The system of any one of aspects 96 to 98, wherein causing the therapeutic treatment to be administered to the patient comprises causing a pharmacological treatment to be administered to the patient via a drug pump.

[0279] 100. The system of any one of aspects 90 to 99, wherein the first prediction data is generated via a forecasting model, and wherein the one or more computer-executable routines are further configured to: provide the first detection data as feedback to the forecasting model; compare the first detection data to the first prediction data; and validate or tune the forecasting model based upon the comparing of the first detection data to the first prediction data.

[0280] 101 . The system of any one of aspects 90 to 100, wherein the one or more computerexecutable routines are further configured to provide the first detection data as input to a titration model to cause the titration model to adjust a therapeutic treatment schedule for the patient.

[0281] 102. The system of aspect 101 , wherein the therapeutic treatment schedule comprises scheduled administrations of neurostimulation to the patient via the neurostimulator device.

[0282] 103. The system of aspect 101 or aspect 102, wherein the therapeutic treatment schedule comprises scheduled administrations of pharmacological treatment to the patient via a drug pump.

[0283] 104. The system of any one of aspects 90 to 103, wherein the one or more computerexecutable routines are further configured to: generate further titration data indicating a therapeutic treatment administered to the patient; and provide the further titration data as input to a forecasting model to cause the forecasting model to generate second prediction data indicating a likelihood of an epileptic seizure event in the patient over a second prediction window subsequent to the first prediction window, based upon the detected epileptic seizure event and the further titration data.

[0284] 105. The system of aspect 104, wherein the further titration data indicates a predicted variance in effect of the administered therapeutic treatment over the second prediction window.

[0285] 106. The system of any one of aspects 90 to 105, wherein the one or more computerexecutable routines configured to obtain the detection model are configured to train the detection model to generate detection data indicating epileptic seizure events in the patient based upon historical training data including historical EEG data of the patient.

[0286] 107. The system of any one of aspects 90 to 106, wherein the electrode array is implanted under a scalp of the patient.

[0287] 108. The system of any one of aspects 90 to 107, wherein the electrode array is disposed in one or more wearable devices affixed externally to the patient.

[0288] 109. The system of any one of aspects 90 to 108, configured to perform the method of any suitable one of aspects 56 to 89.

[0289] 110. The system of any one of aspects 90 to 109, in combination with the system of any suitable one of aspects 34 to 55.

[0290] 111. The system of any one of aspects 90 to 110, configured to perform the method of any suitable one of aspects 1 to 33.

[0291] 112. The system of any one of aspects 90 to 111 , in combination with any other suitable one of aspects 90 to 112.

[0292] 113. Any one of aspects 1 to 112, in combination with any other suitable one of aspects 1 to 112.

Claims

What is claimed is:

1 . A computer-implemented method implemented via one or more processors, the method comprising: obtaining a forecasting model configured to generate predictions of likelihood of epileptic seizure events in a patient based at least upon first input data indicative of historical electroencephalogram (EEG) signal data of the patient; obtaining first titration data for the patient, the first titration data indicating administration of therapeutic treatment to the patient responsive to an epileptic condition or an epileptic seizure event experienced by the patient; generating second input data for the forecasting model based upon the first titration data; and providing the second input data as further input to the forecasting model to cause the forecasting model to generate first prediction data indicating a likelihood of an epileptic seizure event in the patient over a first prediction window based upon the first and second input data.

2. The computer-implemented method of claim 1 , wherein the administration of therapeutic treatment comprises administration of neurostimulation to the patient.

3. The computer-implemented method of claim 1 or claim 2, wherein the administration of therapeutic treatment comprises administration of pharmacological treatment to the patient.

4. The computer-implemented method of any one of claims 1 to 3, wherein the first prediction data indicates varying likelihoods of the epileptic seizure event in the patient over times within the first prediction window.

5. The computer-implemented method of any one of claims 1 to 4, wherein the first predication data indicates a predicted hemisphere or focus of the epileptic seizure event over the first prediction window.

6. The computer-implemented method of any one of claims 1 to 5, further comprising implementing a seizure detection model configured to produce first detection data indicating one or more epileptic seizure events in the patient based upon monitored EEG data of the patient over the first prediction window.

7. The computer-implemented method of claim 6, wherein the monitored EEG data comprises EEG data obtained via a sensor array implanted under a scalp of the patient.

8. The computer-implemented method of claim 6, wherein the monitored EEG data comprises EEG data obtained via one or more sensors in one or more wearable devices affixed externally to the patient.

9. The computer-implemented method of any one of claims 6 to 8, wherein implementing the detection model over the first prediction window comprises: providing the first prediction data as input to the detection model to cause the detection model to adjust a seizure severity threshold or a confidence threshold associated with detection of epileptic seizure events during the first prediction window; and generating the first detection data indicating one or more epileptic seizure events occurring during the first prediction window, based upon the monitored EEG data and the first prediction data.

10. The computer-implemented method of any one of claims 6 to 9, wherein the first detection data indicates respective severity or confidence measurements associated with each of the one or more epileptic seizure events.

11 . The computer-implemented method of any one of claims 6 to 10, further comprising: comparing the first detection data to the first prediction data; and validating or tuning the forecasting model based upon the comparing of the first detection data to the first prediction data.

12. The computer-implemented method of any one of claims 6 to 11 , further comprising providing the first detection data as input to the forecasting model to generate subsequent prediction data indicating a likelihood of epileptic seizure events in the patient over a subsequent prediction window.

13. The computer-implemented method of any one of claims 6 to 12, further comprising providing the first detection data as input to a titration model to generate a treatment schedule comprising administration of first therapeutic treatment of at least one of neurostimulation or pharmacological treatment to the patient.

14. The computer-implemented method of claim 13, further comprising causing the first therapeutic treatment to be administered to the patient in accordance with the treatment schedule.

15. The computer-implemented method of claim 14, wherein causing first therapeutic treatment to be administered to the patient comprises controlling a neurostimulation device affixed to or implanted in the patient, to cause the neurostimulation device to administer neurostimulation to the patient.

16. The computer-implemented method of claim 15, wherein controlling the neurostimulation device comprises transmitting signal communications to the neurostimulation device via Bluetooth communications.

17. The computer-implemented method of any one of claims 14 to 16, wherein causing the first therapeutic treatment to be administered to the patient comprises controlling a drug pump to cause the drug pump to administer a pharmacological treatment to the patient.

18. The computer-implemented method of any one of claims 14 to 17, further comprising: receiving an indication that the first therapeutic treatment was administered to the patient; generating, via the titration model, second titration data indicating the administration of the first therapeutic treatment to the patient; and providing the second titration data as input to the forecasting model to cause the forecasting model to generate second prediction data indicating a likelihood of an epileptic seizure event in the patient over a second prediction window subsequent to the first prediction window.

19. The computer-implemented method of claim 18, wherein the second titration data indicates a predicted variance in effect of the first therapeutic treatment over the second prediction window.

20. The computer-implemented method of any one of claims 1 to 19, wherein obtaining the forecasting model comprises training the forecasting model to generate predictions of epileptic seizure events in the patient based upon historical training data comprising the first input data.

21 . The computer-implemented method of any one of claims 1 to 20, wherein at least one of the obtaining the forecasting model, obtaining the first titration data, generating the second input data, or providing the second input data to the forecasting model is performed via one or more computing devices implanted under the scalp of the patient.

22. The computer-implemented method of any one of claims 1 to 20, wherein at least one of the obtaining the forecasting model, obtaining the first titration data, generating the second input data, or providing the second input data to the forecasting model is performed via one or more computing devices external to the patient.

23. A system for predicting seizure events in a patient, the system comprising: an electrode array comprising: a reference electrode; a plurality of sensing electrodes spaced linearly along a lead of the electrode array and configured, collectively, to measure EEG signals of a brain of the patient; a processing unit, the processing unit comprising: a memory device; a processor communicatively coupled to the electrodes and receiving the EEG signals from plurality of sensing electrodes, the processor configured to store to the memory device data of the received EEG signals; a transceiver coupled to the processor and configured to transmit data to, and receive data from, an external computing device; and one or more computer-executable routines executing on the external computing device and configured to: receive the EEG signals as first input data to a forecasting model; obtain first titration data for the patient, the first titration data indicating administration of therapeutic treatment to the patient; provide second input data to the forecasting model based upon the first titration data to cause the forecasting model to generate first prediction data indicating a likelihood of an epileptic seizure event in the patient over a first prediction window based upon the first and second input data.

24. The system of claim 23, wherein the administration of therapeutic treatment comprises administration of neurostimulation to the patient.

25. The system of claim 23 or claim 24, wherein the administration of therapeutic treatment comprises administration of pharmacological treatment to the patient.

26. The system of any one of claims 23 to 25, wherein the first prediction data indicates varying likelihoods of the epileptic seizure event in the patient over times within the first prediction window.

27. The system of any one of claims 23 to 26, wherein the first predication data indicates a predicted hemisphere or focus of the epileptic seizure event over the first prediction window.

28. The system of any one of claims 23 to 27, wherein the one or more computerexecutable routines are further configured to implement a seizure detection model to produce first detection data indicating one or more epileptic seizure events in the patient based upon monitored EEG data of the patient over the first prediction window, the monitored EEG data being monitored via the electrode array.

29. The system of claim 28, wherein the electrode array is implanted under a scalp of the patient.

30. The system of claim 29, wherein the electrode array is disposed in one or more wearable devices affixed externally to the patient.

31 . The system of any one of claims 28 to 30, wherein the one or more computerexecutable routines configured to implement the seizure detection model over the first prediction window are configured to: provide the first prediction data as input to the detection model to cause the detection model to adjust a seizure severity threshold or a confidence threshold associated with detection of epileptic seizure events during the first prediction window; and generate the first detection data indicating one or more epileptic seizure events occurring during the first prediction window, based upon the monitored EEG data and the first prediction data.

32. The system of any one of claims 28 to 31 , wherein the first detection data indicates respective severity or confidence measurements associated with each of the one or more epileptic seizure events.

33. The system of any one of claims 28 to 32, wherein the one or more computerexecutable routines are further configured to: compare the first detection data to the first prediction data; and validate or tune the forecasting model based upon the comparing of the first detection data to the first prediction data.

34. The system of any one of claims 28 to 33, wherein the one or more computerexecutable routines are further configured to provide the first detection data as input to the forecasting model to generate subsequent prediction data indicating a likelihood of epileptic seizure events in the patient over a subsequent prediction window.

35. The system of any one of claims 28 to 34, wherein the one or more computerexecutable routines are further configured to provide the first detection data as input to a titration model to generate a treatment schedule comprising administration of first therapeutic treatment of at least one of neurostimulation or pharmacological treatment to the patient.

36. The system of claim 35, wherein the one or more computer-executable routines are further configured to cause the first therapeutic treatment to be administered to the patient in accordance with the treatment schedule.

37. The system of claim 36, wherein the one or more computer-executable routines configured to cause first therapeutic treatment to be administered to the patient are configured to control a neurostimulation device affixed to or implanted in the patient, to cause the neurostimulation device to administer neurostimulation to the patient.

38. The system of claim 37, wherein controlling the neurostimulation device comprises transmitting signal communications to the neurostimulation device via Bluetooth communications.

39. The system of any one of claims 36 to 38, wherein the one or more computerexecutable routines configured to cause first therapeutic treatment to be administered to the patient are configured to control a drug pump to cause the drug pump to administer a pharmacological treatment to the patient.

40. The system of any one of claims 36 to 39, wherein the one or more computerexecutable routines are further configured to: receive an indication that the first therapeutic treatment was administered to the patient; generate, via the titration model, second titration data indicating the administration of the first therapeutic treatment to the patient; and provide the second titration data as input to the forecasting model to cause the forecasting model to generate second prediction data indicating a likelihood of an epileptic seizure event in the patient over a second prediction window subsequent to the first prediction window.

41 . The system of claim 40, wherein the second titration data indicates a predicted variance in effect of the first therapeutic treatment over the second prediction window.

42. The system of any one of claims 23 to 41 , wherein the one or more computerexecutable routines configured to obtain the forecasting model are configured to train the forecasting model to generate predictions of epileptic seizure events in the patient based upon historical training data comprising the first input data.

43. A computer-implemented method implemented via one or more processors, the method comprising: obtaining a seizure event detection model configured to generate detection data indicating epileptic seizure events experienced by a patient based upon first input data indicative of electroencephalogram (EEG) signal data of the patient over a first EEG monitoring window; obtaining first prediction data for the patient, the first prediction data indicating likelihood of one or more epileptic seizure events in the patient over a first prediction window matching at least a portion of the first EEG monitoring window, the first prediction data being generated based upon titration data indicating therapeutic treatment administered to the patient; generating second input data for the detection model based upon the first prediction data; and providing the second input data as further input to the detection model to cause the detection model to generate first detection data indicating an epileptic seizure event experienced by the patient over the first EEG monitoring window based upon the first and second input data.

44. The computer-implemented method of claim 43, wherein providing the second input data as further input to the detection model causes the detection model to adjust a seizure severity threshold or a confidence threshold associated with detection of the epileptic seizure event during the first EEG monitoring window.

45. The computer-implemented method of claim 43 or claim 44, wherein the first detection data comprises one or more confidence levels indicating a degree of certainty that a detected epileptic seizure event occurred during the first EEG monitoring window, or a degree of certainty of one or more parameters of the detected epileptic seizure event.

46. The computer-implemented method of any one of claims 43 to 45, wherein the first detection data indicates a severity of a detected epileptic seizure event during the first EEG monitoring window.

47. The computer-implemented method of any one of claims 43 to 46, wherein the first detection data indicates a duration of a detected epileptic seizure event during the first EEG monitoring window.

48. The computer-implemented method of any one of claims 43 to 47, wherein the first detection data indicates a hemisphere or focus of a detected epileptic seizure event during the first EEG monitoring window.

49. The computer-implemented method of any one of claims 43 to 48, further comprising causing a therapeutic treatment to be administered to the patient responsive to the detected epileptic seizure event.

50. The computer-implemented method of claim 49, wherein causing the therapeutic treatment to be administered to the patient comprises causing neurostimulation to be administered to the patient via a neurostimulation device affixed to or implanted in the patient.

51 . The computer-implemented method of claim 50, wherein causing the neurostimulation to be administered to the patient comprises controlling the neurostimulation device via transmitting Bluetooth signals to the neurostimulation device.

52. The computer-implemented method of any one of claims 49 to 51 , wherein causing the therapeutic treatment to be administered to the patient comprises causing a pharmacological treatment to be administered to the patient via a drug pump.

53. The computer-implemented method of any one of claims 43 to 52, wherein the first prediction data is generated via a forecasting model, and the method further comprising: providing the first detection data as feedback to the forecasting model; comparing the first detection data to the first prediction data; and validating or tuning the forecasting model based upon the comparing of the first detection data to the first prediction data.

54. The computer-implemented method of any one of claims 43 to 53, further comprising providing the first detection data as input to a titration model to cause the titration model to adjust a therapeutic treatment schedule for the patient.

55. The computer-implemented method of claim 54, wherein the therapeutic treatment schedule comprises scheduled administrations of neurostimulation to the patient via the neurostimulator device.

56. The computer-implemented method of claim 54 or claim 55, wherein the therapeutic treatment schedule comprises scheduled administrations of pharmacological treatment to the patient via a drug pump.

57. The computer-implemented method of any one of claims 43 to 56, further comprising: generating further titration data indicating a therapeutic treatment administered to the patient; and providing the further titration data as input to a forecasting model to cause the forecasting model to generate second prediction data indicating a likelihood of an epileptic seizure event in the patient over a second prediction window subsequent to the first prediction window, based upon the detected epileptic seizure event and the further titration data.

58. The computer-implemented method of claim 57, wherein the titration data indicates a predicted variance in effect of the administered therapeutic treatment over the second prediction window.

59. The computer-implemented method of any one of claims 43 to 58, wherein obtaining the detection model comprises training the detection model to generate detection data indicating epileptic seizure events in the patient based upon historical training data including historical EEG data of the patient.

60. The computer-implemented method of any one of aspects 43 to 59, comprising receiving the first input data by monitoring the EEG signal data via a sensor array implanted under a scalp of the patient.

61 . The computer-implemented method of any one of aspects 43 to 60, comprising receiving the first input data by monitoring the EEG signal data via one or more sensors in one or more wearable devices affixed externally to the patient.

62. The computer-implemented method of any one of claims 43 to 61 , wherein at least one of the obtaining the seizure event detection model, obtaining the first prediction data, generating the second input data, or providing the second input data to the detection model is performed via one or more computing devices implanted under the scalp of the patient.

63. The computer-implemented method of any one of claims 43 to 61 , wherein at least one of the obtaining the seizure event detection model, obtaining the first prediction data, generating the second input data, or providing the second input data to the detection model is performed via one or more computing devices external to the patient.

64. A system for predicting seizure events in a patient, the system comprising: an electrode array comprising: a reference electrode; a plurality of sensing electrodes spaced linearly along a lead of the electrode array and configured, collectively, to measure EEG signals of a brain of the patient; a processing unit, the processing unit comprising: a memory device; a processor communicatively coupled to the electrodes and receiving the EEG signals from plurality of sensing electrodes, the processor configured to store to the memory device data of the received EEG signals; a transceiver coupled to the processor and configured to transmit data to, and receive data from, an external computing device; and one or more computer-executable routines executing on the external computing device and configured to: receive the EEG signals as first input to a seizure event detection model configured to generate detection data indicating epileptic seizure events experienced by the patient over a first EEG monitoring window; obtain first prediction data for the patient, first prediction data indicating likelihood of one or more epileptic seizure events in the patient over a first prediction window matching at least a portion of the first EEG monitoring window; and provide second input data to the detection model based upon the first prediction data to cause the detection model to generate first detection data indicating an epileptic seizure event experienced by the patient over the first EEG monitoring window based upon the first and second input data.

65. The system of claim 64, wherein the providing of the second input data to the detection model causes the detection model to adjust a seizure severity threshold or a confidence threshold associated with detection of the epileptic seizure event during the first EEG monitoring window.

66. The system of claim 64 or aspect 65, wherein the first detection data comprises one or more confidence levels indicating a degree of certainty that a detected epileptic seizure event occurred during the first EEG monitoring window, or a degree of certainty of one or more parameters of the detected epileptic seizure event.

67. The system of any one of claims 64 to 66, wherein the first detection data indicates a severity of a detected epileptic seizure event during the first EEG monitoring window.

68. The system of any one of claims 64 to 67, wherein the first detection data indicates a duration of a detected epileptic seizure event during the first EEG monitoring window.

69. The system of any one of claims 64 to 68, wherein the first detection data indicates a hemisphere or focus of a detected epileptic seizure event during the first EEG monitoring window.

70. The system of any one of claims 64 to 69, wherein the one or more computerexecutable routines are further configured to cause a therapeutic treatment to be administered to the patient responsive to the detected epileptic seizure event.

71 . The system of claim 70, wherein causing the therapeutic treatment to be administered to the patient comprises causing neurostimulation to be administered to the patient via a neurostimulation device affixed to or implanted in the patient.

72. The system of claim 71 , wherein causing the neurostimulation to be administered to the patient comprises controlling the neurostimulation device via transmitting Bluetooth signals to the neurostimulation device.

73. The system of any one of claims 70 to 72, wherein causing the therapeutic treatment to be administered to the patient comprises causing a pharmacological treatment to be administered to the patient via a drug pump.

74. The system of any one of claims 64 to 73, wherein the first prediction data is generated via a forecasting model, and wherein the one or more computer-executable routines are further configured to: provide the first detection data as feedback to the forecasting model; compare the first detection data to the first prediction data; and validate or tune the forecasting model based upon the comparing of the first detection data to the first prediction data.

75. The system of any one of claims 64 to 74, wherein the one or more computerexecutable routines are further configured to provide the first detection data as input to a titration model to cause the titration model to adjust a therapeutic treatment schedule for the patient.

76. The system of claim 75, wherein the therapeutic treatment schedule comprises scheduled administrations of neurostimulation to the patient via the neurostimulator device.

77. The system of claim 75 or claim 76, wherein the therapeutic treatment schedule comprises scheduled administrations of pharmacological treatment to the patient via a drug pump.

78. The system of any one of claims 64 to 77, wherein the one or more computerexecutable routines are further configured to: generate further titration data indicating a therapeutic treatment administered to the patient; and provide the further titration data as input to a forecasting model to cause the forecasting model to generate second prediction data indicating a likelihood of an epileptic seizure event in the patient over a second prediction window subsequent to the first prediction window, based upon the detected epileptic seizure event and the further titration data.

79. The system of claim 88, wherein the further titration data indicates a predicted variance in effect of the administered therapeutic treatment over the second prediction window.

80. The system of any one of claims 64 to 79, wherein the one or more computerexecutable routines configured to obtain the detection model are configured to train the detection model to generate detection data indicating epileptic seizure events in the patient based upon historical training data including historical EEG data of the patient.

81 . The system of any one of claims 64 to 80, wherein the electrode array is implanted under a scalp of the patient.

82. The system of any one of claims 64 to 81 , wherein the electrode array is disposed in one or more wearable devices affixed externally to the patient.