Detection of risk or occurrence of stroke, aneurysm or cerebral infarction

Abstract

The present disclosure relates to a method and system for acquiring and analyzing physiological signals of a subject to detect the risk or occurrence of stroke or cerebral infarction. A wearable wireless device may be used with one or more electrode clusters to record physiological signals during a sleep-wake cycle. The physiological signal may be utilized to perform sleep and wakefulness analysis. In some examples, physiological signals corresponding to each hemisphere of the brain may be compared, e.g., using coherence, to quantify asymmetry across the hemispheres and identify possible stroke, asymptomatic cerebral infarction (SBI), or aneurysm. Biomarkers may be extracted based on sleep and wakefulness analysis as well as coherence. Biomarkers may include an indication of obstructive sleep apnea or apnea risk, a reduction in slow wave sleep duration, or the presence of SBI or aneurysm. A stroke risk score may be generated from the biomarkers using a predictive model for the early detection of individuals that are potentially at risk of stroke.

Description

Cross-references to related applications

[0001] This application claims priority and benefit to U.S. Provisional Application No. 63 / 586793, filed September 29, 2023, entitled “Detection of Risk or Occurrence of Cerebral Infarction Stroke,” the entire contents of which are incorporated herein by reference for all purposes. Background Technology

[0002] Stroke is a leading cause of death worldwide and a leading cause of severe disability in adults. Time is crucial in treating neurodegenerative diseases such as stroke. Common stroke treatments include injections of tissue plasminogen activators (which dissolve blood clots to increase blood flow) or endovascular surgery to remove the clots. However, if these treatments are not administered promptly, neurons can die due to lack of blood supply. Strokes can be classified into two types: ischemic stroke and hemorrhagic stroke, both of which can lead to neuronal death, resulting in further (often irreversible) brain damage and even death.

[0003] However, stroke can be prevented through early diagnosis and timely intervention. Furthermore, people often experience mini-strokes or transient ischemic attacks (TIAs), which may be caused by aneurysms, before an actual stroke. Therefore, dedicated systems capable of automatically screening or identifying patients at risk of stroke or with aneurysms (ideally in home, long-term care, or off-hospital settings) can save lives. Thus, there is a need for an improved system to accurately detect stroke and / or predict an individual's stroke risk, enabling early detection of those who may develop a stroke in the near future. Such systems can improve an individual's quality of life through stroke prevention and timely intervention, and can reduce the economic burden on healthcare systems and individuals. Summary of the Invention

[0004] Some embodiments of this disclosure relate to using physiological signals from a subject to detect the risk or occurrence of stroke in that subject. A computer-implemented method includes accessing physiological data of a subject collected by a physiological data acquisition component over a period of time. The physiological data acquisition component may include sensing devices and one or more electrode clusters. The sensing devices may include accelerometers and gyroscopes. The sensing devices can be used to acquire, process, and transmit signals from one or more electrode clusters. The one or more electrode clusters may include electroencephalography (EEG) electrodes, electromyography (EMG) electrodes, magnetoencephalography (MEG) electrodes, or electrooculography (EOG) electrodes. Each of the one or more electrode clusters includes at least an active electrode. Other electrodes in each cluster may include reference electrodes or bias electrodes. The physiological data may correspond to physiological signals (e.g., EEG, EMG, EOG, MEG) collected during nighttime, rest periods, or multiple previous nighttime or rest periods.

[0005] Sleep structures can be generated by analyzing the physiological data of a subject. For example, generating sleep structures may include extracting a set of features based on a portion of the physiological data corresponding to each of a plurality of time intervals within a given time period. This set of features may be associated with one or more frequency bands of the physiological signal corresponding to each time interval. The set of features may include one or more of the following: delta power, gamma power, standard deviation, maximum amplitude, gamma power / delta power, the time derivative of delta, and the time derivative of gamma power / delta power. The set of features may also include features derived from the spectrograms or normalized spectrograms of one or more frequency bands of the physiological signal for that time interval using component analysis (e.g., principal component analysis (PCA), independent component analysis (ICA)).

[0006] Furthermore, the state can be predicted for the feature set corresponding to each time interval. This state can correspond to one or more sleep stages or wakefulness states. One or more sleep stages can include rapid eye movement (REM) stages and one or more non-rapid eye movement (NOM) stages. In some instances, a sleep classification model can be used to predict this state. The sleep classification model can include supervised machine learning techniques such as decision trees, support vector machines (SVM), random forests, or neural networks trained on labeled sleep data. In other cases, clustering techniques such as K-means clustering, hierarchical clustering, or Gaussian mixture models (GMM) can be used to predict the state for each time interval. Based on the predicted state for each time interval, the sleep pattern, the relative frequency of one or more sleep stages or wakefulness states, and the duration of one or more sleep stages or wakefulness states can be determined. Additionally, a spectrum and time segments can be calculated.

[0007] Furthermore, one or more biomarkers can be determined, at least in part, based on the subject's sleep structure. These biomarkers may include indications of apnea, apnea risk, sleep variability, or transhemispheric variations. In some instances, a first portion of physiological data corresponding to the first hemisphere of the subject's brain can be determined. Similarly, a second portion of physiological data corresponding to the second hemisphere of the subject's brain can be determined. The coherence between the first and second portions of the physiological data can then be calculated. The first and second portions may correspond to segments (or periods) of physiological data (e.g., EEG signals) or to a sleep-wake cycle. Based on apnea, apnea risk, sleep variability, and coherence, the presence or likelihood of hemorrhagic stroke, asymptomatic brain infarction (SBI), transient ischemic attack (TIA), aneurysm, brain tumor, or traumatic brain injury (TBI) can be determined. Additionally, one or more biomarkers may include the presence of obstructive sleep apnea (OSA), the duration of OSA, the intensity of OSA, or a reduction in the duration of slow-wave sleep (SWS).

[0008] Predictive models can be used to predict a subject's stroke risk score based on one or more biomarkers. In some implementations, the stroke risk score is further predicted based on risk factor data including the subject's medical information and activity indicators. Medical information may include body mass index (BMI), atrial fibrillation diagnosis, diabetes diagnosis, history of blood glucose levels, blood pressure data, family history of stroke or TIA, family history of heart attack, or blood cholesterol levels. Activity indicators may include weekly physical activity intensity. Predictive models may include machine learning models, including recurrent neural networks (RNNs), transformer models, or neural networks.

[0009] Subsequently, the condition is determined, at least in part, based on the stroke risk score. For example, the condition could be: whether the subject's (predicted) stroke risk is above a certain threshold; whether the stroke risk score predicts a high risk of future stroke; or whether the stroke risk score is moderate and accompanied by apnea or detected SBI (including small stroke or TIA), etc.

[0010] In addition, one or more actions are triggered based on the fulfillment of certain conditions. These actions may include: alerting the subject, alerting caregivers or clinicians, or outputting results that may provide a basis for recommendations, or include recommendations to perform assessments or interventions to reduce the stroke risk score. One or more actions may also include presenting the results and / or the subject's stroke risk score on a computing device, or transmitting the results to other devices. Results may also include sleep structure, apnea results, interhemispheric comparisons (e.g., coherence), and stroke results to facilitate further research by clinicians and to develop treatment plans for the subject.

[0011] In some embodiments, a system is provided that includes one or more data processors and a non-transitory computer-readable storage medium including instructions that, when executed on the one or more data processors, cause the one or more data processors to perform some or all of the one or more methods disclosed herein.

[0012] In some embodiments, a computer program product is provided, tangibly embodied in a non-transitory machine-readable storage medium, and includes instructions configured to cause one or more data processors to perform some or all of the methods disclosed herein.

[0013] In some embodiments, a system is provided that includes one or more means for performing some or all of the methods or processes disclosed herein.

[0014] The terminology and expressions used are for descriptive purposes only and not for limitation, and their use is not intended to exclude any equivalents of the shown and described features or portions thereof. However, it should be recognized that various modifications are possible within the scope of the claims of this invention. Therefore, it should be understood that although the claims of this invention have been specifically disclosed through embodiments and optional features, those skilled in the art can make modifications and variations using the concepts disclosed herein, and such modifications and variations are considered to be within the scope of the invention as defined in the appended claims. Attached Figure Description

[0015] Various embodiments will be described below with reference to the accompanying drawings. It should be noted that the drawings are not drawn to scale, and elements with similar structures or functions in all drawings are represented by the same reference numerals. It should also be noted that the drawings are for illustrative purposes only and are not intended to be an exhaustive description of this disclosure or a limitation on its scope.

[0016] Figure 1 An overview of examples of systems for detecting the risk or occurrence of stroke in a subject, according to some embodiments of this disclosure, is shown.

[0017] Figure 2 An example placement of an adhesive film, electrode, and sensing device according to some embodiments of the present disclosure on the forehead of a subject is shown.

[0018] Figure 3 An example flowchart illustrating the processing of physiological signals and extraction of physiological features according to some embodiments of the present disclosure is shown.

[0019] Figure 4 An example implementation of a sleep analyzer that performs sleep stage analysis according to some aspects of this disclosure is shown.

[0020] Figure 5 Examples of implementations of sleep apnea detection in subjects using the output of a sleep analyzer, EEG data, and possibly one or more non-EEG sensor data, according to some embodiments of this disclosure, are shown.

[0021] Figure 6 Example diagrams are shown that process EEG data from the left and right hemispheres to detect stroke, mini-stroke, transient ischemic attack (TIA), or asymptomatic stroke (SBI) in the subject's brain.

[0022] Figure 7 An example architecture for generating a stroke risk score for a subject by using a predictive model, according to some embodiments of this disclosure, is shown.

[0023] Figure 8 An example flowchart of a system for detecting the risk or occurrence of cerebral infarction stroke in a subject, according to some embodiments of the present disclosure, is shown.

[0024] Figure 9 Example diagrams of computer systems that can implement various embodiments of the present disclosure are shown. Detailed Implementation

[0025] Embodiments of this disclosure relate to a method and system for acquiring and analyzing physiological signals of a subject to detect the risk or occurrence of stroke or cerebral infarction. Physiological data acquisition components can be used to acquire physiological signals of a subject during a period of time, whether in a sleep or resting state. According to this disclosure, one or more biomarkers or subclinical stroke risk factors can be identified by analyzing the physiological signals. A stroke risk score for the subject can be generated using a predictive model. According to some embodiments, this disclosure provides a solution for the early detection of individuals at risk of stroke or who may have an aneurysm or SBI, preferably in a home setting (or even in a hospital setting).

[0026] According to some aspects of this disclosure, identifying subclinical risk factors can facilitate earlier and potentially more effective preventative measures against diseases such as stroke, aneurysm, TBI, and brain tumors. Asymptomatic stroke is one such potential risk factor. The term "asymptomatic stroke (SBI)" is used to describe a stroke that is visible on a computed tomography (CT) or magnetic resonance imaging (MRI) scan of the brain but without any corresponding stroke event. SBI can refer to an area of ​​brain damage (e.g., necrosis or tissue death) due to insufficient blood flow (ischemia). SBI is often discovered incidentally during imaging examinations (such as MRI or CT scans) performed for other reasons. Subjects or individuals typically do not experience any noticeable symptoms, hence the term "asymptomatic." Although SBI is asymptomatic, it can be associated with cognitive decline, particularly in older adults, and may increase the risk of future stroke or dementia. The immediate potential sequela of SBI is symptomatic stroke. Approximately 25% of people over 80 years of age may experience more than one SBI. The prevalence of SBI may be significantly higher than that of symptomatic stroke. It is estimated that more than 10 episodes of SBI may occur before each symptomatic stroke. Furthermore, SBI is associated with future cognitive decline and stroke incidence. Therefore, a system that can easily detect SBI without requiring an appointment for a CT or MRI scan would be of great benefit and effectiveness in stroke prevention.

[0027] Similarly, one or more “mini-strokes,” also known as transient ischemic attacks (TIAs), may precede an ischemic stroke. A TIA, often called a mini-stroke, is a temporary episode of neurological dysfunction caused by a brief insufficiency of blood supply to the brain. A TIA can last from a few minutes to up to 24 hours and is often a warning sign of a potential future stroke. Common symptoms include sudden numbness or weakness on one side of the body, difficulty speaking, sudden confusion, or vision problems. SBI, mini-strokes, and TIAs can be caused by a variety of factors, including vascular diseases such as aneurysms. According to the disclosed embodiments, SBI, TIA, or mini-strokes can be detected based on physiological data (such as EEG signals), and subjects are advised to undergo a thorough examination by a clinician before an aneurysm ruptures.

[0028] In addition, some other subclinical risk factors for stroke include obstructive sleep apnea (OSA) and reduced slow-wave sleep (SWS) duration. In older adults with largely normal cognitive function, reduced SWS and severe OSA may be associated with an increased abnormal white matter burden, which may lead to an increased risk of cognitive impairment, dementia, and stroke. OSA may also be disproportionately associated with stroke. According to some aspects of this disclosure, sleep analysis or sleep stage analysis can be performed using physiological signals (e.g., EEG signals) or physiological data from subjects. Sleep stage analysis can classify each period or segment of the EEG signal into one of several predefined sleep stages, including wakefulness, REM, and non-REM stages, such as SWS, stage I, and stage II. Sleep stage analysis can generate outputs: wakefulness or micro-awake detection results, microsleep detection results, sleep patterns, relative frequency and duration of each stage in multiple predefined stages, sleep scores, or sleep graphs. OSA typically causes transient awakenings, which can be detected as a sudden change in the EEG signal. Therefore, additional risk factors or subclinical risk factors, such as reduced OSA or SWS, can be identified based on sleep analysis.

[0029] In resting or sleep states, EEG data or signals exhibit bilateral symmetry. Stroke is typically unilateral; therefore, attenuation or deterioration of interhemispheric coherence may predict symptomatic stroke, small stroke, TIA, or SBI. Interhemispheric comparisons, using methods such as statistical techniques, clustering techniques, distance metrics, or coherence, can provide valuable insights into the functional connectivity and synchronicity of brain activity across different regions and hemispheres. In some instances, interhemispheric comparisons can be used to compare physiological signals corresponding to each hemisphere of the brain to quantify interhemispheric asymmetry and detect the presence or likelihood of neurodegenerative abnormalities. Therefore, biomarkers (or subclinical risk factors) can be extracted based on sleep analysis and interhemispheric comparison results. Biomarkers may include the detection or severity of OSA, reduced SWS duration, or the presence of SBI. After detecting the presence of neurodegenerative abnormalities, the techniques disclosed herein can also detect or identify the presence or likelihood of hemorrhagic stroke, minor stroke, TIA, SBI, aneurysm, brain tumor, or TBI based on one or more biomarkers (e.g., apnea, apnea risk, sleep changes, and interhemispheric comparison results).

[0030] Furthermore, predictive models can be used to generate stroke risk scores based on biomarkers for the early detection of individuals potentially at risk of stroke. These predictive models can include machine learning models such as recurrent neural networks (RNNs), long short-term memory (LSTM) models, transformer models, neural networks (NNs), multi-value prediction algorithms, deep learning models, or regression techniques. Stroke risk scores can be numerical (e.g., 0 to 10), binary values, or categories (e.g., mild, moderate, or high risk).

[0031] In some instances, predictive models may utilize other risk factor data besides biomarkers, such as clinical risk factors or non-EEG-based risk factors. These risk factors may include, but are not limited to, body mass index (BMI), atrial fibrillation diagnosis, diabetes diagnosis, history of blood glucose levels, blood pressure data, family history of stroke or TIA, family history of heart attack, or blood cholesterol levels. Activity indicators may include, for example, weekly physical activity intensity.

[0032] After predicting a stroke risk score, one or more actions can be triggered based on the fulfillment of certain conditions. These actions may include: alerting the subject, alerting caregivers or clinicians, or outputting results or data to provide a basis for recommendations or to include recommendations to perform interventions (such as medical assessments and, possibly, MRI) to reduce the stroke risk score. One or more actions may also include presenting the results and / or the subject's stroke risk score on a user device (e.g., a smartwatch, smartphone) or transmitting the results to other devices. Results may also include biomarkers, sleep analysis results, apnea results, comparative and detection results to facilitate further research by clinicians and to develop treatment plans for the subject.

[0033] In some instances, neurodegenerative disease detection systems include a physiological data acquisition component configured to acquire physiological data related to the subject's brain. This physiological data acquisition component may include sensing devices and one or more electrode clusters. Each of the one or more electrode clusters includes at least one active electrode. The one or more electrode clusters may also include a reference electrode or a ground electrode. These electrodes may include electroencephalography (EEG) electrodes, electromyography (EMG) electrodes, magnetoencephalography (MEG) electrodes, electrooculography (EOG) electrodes, electrocardiography (EKG) electrodes, etc. These electrodes may be dry contact electrodes, dry non-contact (capacitively coupled) electrodes, or wet contact electrodes.

[0034] Physiological data acquisition may also include a transmitter and, possibly, a receiver (which may be a single transceiver). The transmitter may be configured to transmit data corresponding to signals recorded by electrodes to a computing device that is part of a neurodegenerative abnormality detection system. The computing device may be operated by the subject, by a healthcare provider associated with the treating subject, or by an entity providing medical monitoring or treatment to the subject. This communication can be performed using any of the various commercially available protocols, such as wireless networks, including short-range connections (e.g., Bluetooth, Bluetooth Low Energy (BTLE), or ultra-wideband connections) or via WiFi networks (e.g., the Internet). In some instances, the receiver is configured to receive instructions or requests from the computing device, such as an instruction to begin recording signals or a request to send data to the computing device.

[0035] Furthermore, the physiological data acquisition component can be implemented as a wearable device, such as a sensing patch. This sensing patch may include an adhesive membrane, electrodes, and sensing devices. In some embodiments, the electrodes and connecting wires (or electrode leads) can be implemented using a flexible printed circuit board (PCB) and can be attached to the subject using an adhesive material (e.g., an adhesive membrane) or some type of gel for better signal acquisition. The sensing devices can be adhered to the flexible printed circuit board and connected to the electrodes via PCB traces. In some other instances, the sensing devices and electrodes can be co-implemented on a flexible PCB to develop a sensing patch. Furthermore, the electrode structure on the flexible PCB (e.g., the number of electrodes or channels, their location, size, etc.) can be controlled during manufacturing. The sensing patch may include at least one active electrode and a microprocessor (e.g., located inside the sensing device) configured to transmit signals collected by the active electrode or a processed version thereof.

[0036] In some instances, physiological data acquisition components may include wearable components such as headbands, one or more straps, one or more bands, caps, helmets, or caps, where each of a plurality of electrodes is positioned at a location intended to be aligned with a specific brain region when the sensing device is worn. The wearable component may have receiving components (e.g., openings for receiving sensing patches or electrodes). The wearable component can help ensure that the electrodes or adhesive membranes are placed in the subject's target location. Furthermore, instructions may be provided to the subject indicating where the electrodes or adhesive membranes are placed.

[0037] The physiological data acquisition component may include a processing component that performs initial processing using the signals recorded by the electrodes. Such processing may be performed by executing software code and / or using hardware components. Initial processing may include amplifying the signals recorded by the electrodes, determining the differential signal, applying filters (e.g., removing signals around 50 Hz or 60 Hz based on geographic region, or focusing on a frequency band of interest), and / or downsampling the signal. The differential signal can be determined by subtracting the signal from one electrode. For example, the signal from a reference electrode may be subtracted from the signal from an active electrode, or the signal from a first active electrode may be subtracted from the signal from a second active electrode.

[0038] In some instances, one or more initial processing actions may be performed alternatively or additionally at the computing device to which the signal is sent. This computing device may include mobile devices (e.g., smartphones), tablets, laptops, desktop computers, computer servers, and so on.

[0039] The various disclosures in this document relate to techniques for using physiological data acquisition components to improve the prediction of an individual's risk of developing or having already developed a stroke, aneurysm, or cerebral infarction. In other instances, the disclosed techniques can be used to assess the effectiveness of interventions or to monitor the recovery of subjects who have developed neurodegenerative diseases such as stroke.

[0040] Figure 1 An example overview of a system for detecting the risk or occurrence of stroke according to some embodiments of this disclosure is shown. Exemplary system 100 includes a sensing device 105, a network 110, a computing device 115, and one or more databases 120. The sensing device 105 may include a transceiver 108 for communicating with the computing device 115. The sensing device 105 may be connected to the computing device 115 via the network 110. The transceiver 108 may be configured to transmit physiological data recorded by the sensing device 105 to the computing device 115 as part of a neurodegenerative abnormality detection system.

[0041] The computing device 115 can be operated by the subject, by a clinician or healthcare provider associated with the subject, or by an entity assisting the subject in medical monitoring or treatment. The computing device 115 may include mobile devices (e.g., smartphones), personal digital assistants (PDAs), tablets, laptops, desktop computers, computer servers, etc. In some instances, transceiver 108 and / or sensing device 105 may be configured to receive instructions or requests from the computing device 115, such as an instruction to begin recording signals or a request to send data to the computing device 115. Furthermore, communication between the sensing device 105 and the computing device 115 can be conducted using network 110, which may be a wireless network based on commercially available communication protocols, such as Bluetooth, Bluetooth Low Energy, Ultra Wideband (UWB), or WiFi networks (e.g., the Internet). Network 110 may include the Internet, intranets, cellular networks, wired LANs (Local Area Networks), wireless LANs (WiLANs), WANs (Wide Area Networks), MANs (Metropolitan Area Networks), PSTNs (Public Switched Telephone Networks), and other types of communication networks. Network 110 may also include communication devices, such as one or more gateways, routers, or bridges. Network 110 can be any type of network familiar to those skilled in the art, capable of supporting data communication using a variety of available protocols, including but not limited to TCP / IP (Transmission Control Protocol / Internet Protocol), SNA (System Network Architecture), IPX (Internet Packet Switching), AppleTalk®, etc.

[0042] According to some embodiments, the physiological data acquisition component may include a sensing device 105, different types of electrodes, and / or wearable components. Different types of electrodes may include, but are not limited to, dry electrodes, wet electrodes, self-adhesive conductive electrodes, electrodes with snap-fit ​​connectors, EEG electrodes, EMG electrodes, EOG electrodes, MEG electrodes, etc.

[0043] In some embodiments, the sensing device 105 is configured to acquire, record, process, and transmit physiological data related to the subject's brain. In some instances, the physiological data includes electrical activity of the brain, which can be recorded using EEG electrodes attached to the subject's scalp or forehead. The sensing device 105 may be configured with at least one active electrode and a reference electrode. The active electrode acts as the primary sensor for detecting electrical activity directly or indirectly generated by neuronal firing in the brain and nervous system. The active electrode picks up the electrical activity generated by the brain and nervous system and transmits these signals to the sensing device 105 for preliminary processing (e.g., signal amplification, analog-to-digital conversion, noise reduction, etc.) and analysis. The reference electrode provides a baseline or common point of comparison for the active electrode. The sensing device 105 may also include a ground (or bias) electrode. The ground or bias electrode may be placed behind the subject's ear or placed together with the active or reference electrode. The bias electrode serves to stabilize the baseline or potential of the EEG system and reduce noise or interference from an external power source. In some instances, physiological data acquisition components can omit grounding or bias electrodes, instead using reference and active electrodes for data recording. This is because modern differential amplifiers can be designed to operate without dedicated grounding electrodes, instead utilizing a virtual ground created internally within the amplifier circuitry.

[0044] In some other embodiments, the sensing device 105 may be configured to record and store physiological data in an encrypted format, and wirelessly transmit it to a remote center or computing device 115 for further display, storage, processing, and analysis. In another aspect of this disclosure, the recorded and stored physiological data may be wirelessly transmitted in real time to the computing device 115, including cellular phones, smartphones, tablets, and / or computers. The recorded and stored physiological data may also be directly transmitted to computers, cellular phones, smartphones, and / or tablets via the Universal Serial Bus (USB) transmission function built into the sensing device 105.

[0045] During physiological data acquisition, initial or pre-amplification can be performed at or near the electrodes to reduce noise. For example, the electrode snap-fit ​​connector assembly may include noise-reducing or noise-eliminating filters in the electrode connection layer to reduce any electrical noise that the wires may pick up. To further improve the physiological data acquisition process, the sensing device 105 or the electrode snap-fit ​​connector assembly may be configured to continuously monitor electrode impedance and may include indicator lights to indicate the current status of the electrode's connection integrity to the scalp or forehead. The sensing device 105 may include hardware and software components (e.g., firmware or signal processing code) and may be used to perform initial processing. Initial processing may include amplifying the signal recorded by the electrodes, determining the differential signal, applying filters (e.g., removing signals around 50-60 Hz or focusing on the band of interest), and / or downsampling the signal. The differential signal can be determined by subtracting the signal from one electrode. For example, the signal from a reference electrode may be subtracted from the signal from an active electrode, or the signal from a first active electrode may be subtracted from the signal from a second active electrode.

[0046] Sensing device 105 may also include a battery-powered assembly, which may include a rechargeable small high-capacity battery. In some instances, the battery-powered assembly may include a disposable battery. Sensing device 105 may also include memory, a processor, and a transceiver 108 for transmitting data to, for example, computing device 115 or one or more databases 120. Sensing device 105 includes power and charging circuitry for receiving power via a power cord and an AC unit. The power cord is connected to sensing device 105 for charging the small high-capacity battery via a port, which may be (but is not limited to) USB, D-Subminiature (DB)-25, etc. Sensing device 105 includes a power switch function for conserving power of the small high-capacity battery when not in use. Sensing device 105 may also include a power switch indicator light for indicating the current status of sensing device 105. In some instances, sensing device 105 can be charged via a USB connection to a computer. In some other instances, sensing device 105 can be charged wirelessly.

[0047] The exemplary system 100 may also include one or more databases 120 for storing and future processing data (e.g., subject's EEG signals). The subject's physiological data may be stored along with metadata. Metadata may include subject information, type, and the placement of each electrode, etc. In some instances, the sensing device 105 may transmit EEG signals with metadata, or in a certain order, to indicate which EEG signals correspond to electrodes to be placed on the contralateral side of the subject. In some instances, the computing device 115 may correlate various signals from active electrodes with different sides of the subject, but does not specifically determine or predict which signals specifically correspond to the "left" side or hemisphere, or the "right" side or hemisphere. One or more databases 120 may be part of a computer storage system or auxiliary storage device (e.g., hard disk, floppy disk, optical disk, or other non-volatile mass storage device). Furthermore, the computing device 115 may be used to perform signal processing techniques or algorithms on the physiological data (previously recorded signals or real-time signals) and store the results in one or more databases 120. In one embodiment, the sensing device 105 may include a memory, and in another embodiment, the sensing device 105 may also have a plurality of processors for computation.

[0048] Figure 2 An example placement 200 of an adhesive membrane 205, electrodes, and sensing device 105 on the forehead of a subject is shown according to some embodiments of the present disclosure. According to this example placement 200, two or more electrodes 210a-n and sensing device 105 can be adhered to the adhesive membrane 205 to capture physiological signals from the left and right hemispheres. In some instances, the adhesive membrane 205 may be a single, long adhesive membrane affixed to the subject's forehead. The two or more electrodes 210a-n may include an active electrode, a reference electrode, and a bias electrode. These electrodes may be placed close to each other on a sensing patch. In some instances, the bias electrode may be connected to the subject's ear (e.g., earlobe or behind the ear) using electrode wires. In some other instances, the bias electrode may be omitted, and the two or more electrodes 210a-n may consist only of an active electrode and a reference electrode. In some instances, the sensing device 105 is not located on the head.

[0049] In some embodiments, the adhesive membrane 205 may include a stretchable material. In some instances, two or more electrodes 210a-n and connecting wires (or electrode leads) may be implemented using a flexible printed circuit board (PCB) and may be attached to the subject using adhesive materials, headbands, glasses (or goggles), or some type of gel for better signal acquisition. The sensing device 105 may be adhered to the flexible printed circuit board and connected to the two or more electrodes 210a-n via PCB traces. In some other instances, the sensing device 105 and the two or more electrodes 210a-n may be jointly implemented on a flexible PCB, which may act as a single sensing patch. Furthermore, the electrode structure on the flexible PCB (e.g., the number of electrodes or channels, their location, size, etc.) may be controlled during manufacturing. Physiological data or signals may be acquired during sleep or resting states.

[0050] The overall composition of the adhesive membrane 205, two or more electrodes 210a-n, and sensing device 105 may be referred to as a sensing patch. The sensing patch may have a surface or adhesive membrane 205 extending in both length and width directions. Adhesive material may be disposed on part or all of the surface of the adhesive membrane 205 (e.g., across part or all of one or more edges of the surface). The length may be, for example, less than 10 cm, less than 8 cm, less than 6 cm, less than 4 cm, less than 2 cm, etc. The width may be, for example, less than 10 cm, less than 8 cm, less than 4 cm, etc. The length may be, for example, greater than 0.5 cm, greater than 1 cm, greater than 2 cm, greater than 4 cm, etc. The width may be, for example, greater than 0.5 cm, greater than 1 cm, greater than 2 cm, greater than 4 cm, etc. The length may be, for example, between 0.5-10 cm, between 1-6 cm, between 2-4 cm, between 2-8 cm, between 2-8 cm, and / or any other semi-enclosed or enclosed range having the thresholds disclosed herein. The width may be, for example, between 0.5-10 cm, between 1-6 cm, between 2-4 cm, between 2-8 cm, between 2-8 cm and / or any other semi-closed or closed range having the thresholds disclosed herein.

[0051] In some other embodiments, two or more sensing patches can be used to acquire physiological signals from the left and right hemispheres of a subject's brain. For example, two sensing patches can be used and attached to two different locations, such as each side of the brain or the forehead. In some instances, unconnected adhesive membranes can be used at different locations. Two or more electrodes 210a-n can be adhered to each of the unconnected adhesive membranes. Similarly, each adhesive membrane can be connected to transceiver 108 or sensing device 105. In some other instances, the sensing patches can include a series of physiological sensors, such as EEG, EOG, EMG, and MEG sensors. These sensing patches can be strategically placed at various sites on the skull to capture comprehensive physiological data. Each sensing patch can operate independently but also communicate with computing device 115, providing continuous monitoring even if one patch experiences a temporary malfunction or interference. For example, one sensing patch can focus on monitoring neural signals (e.g., EEG and MEG), while another can track eye movements (EOG) and muscle activity (EMG).

[0052] In some instances, the physiological data acquisition component may include a wearable component, attached to or replacing the adhesive membrane. The wearable component may include one or more straps, bands, or caps, and may have receiving components (e.g., openings for receiving patches or electrodes). The wearable component may help ensure that two or more electrodes 210a-n and / or membranes are located at a target position on the subject. In some cases, the sensing device 105 may be housed in a wearable component such as a headband, for example, the headband and wireless EEG monitoring system disclosed in U.S. Application No. 17 / 214574, filed March 26, 2021, the entire contents of which are incorporated herein by reference for all purposes. The headband includes straps and fasteners (e.g., Velcro, hooks, buttons, etc.) for customized fit, adjustment, and improved subject comfort. The headband may also include multiple slots for attaching electrodes or electrode snap connectors at specific locations. For example, the bias electrode and the reference electrode can be installed behind the left and right ears of the subject, respectively, while the active electrode can be installed on the subject's forehead.

[0053] In addition, instructions can be provided to the subject indicating where to place the two or more electrodes 210a-n, one or more adhesive films or sensor patches. For example, drawings or photographs can be provided showing the locations where one or more adhesive films are respectively attached to the subject's head (e.g., a first film on the left side of the forehead, a second film on the right side of the forehead, or a long, thin film placed across the subject's forehead).

[0054] Physiological data or signals can be acquired, for example, during sleep or rest. Physiological data can be collected during the night, during rest, or over several previous nights or rest periods. Typically, EEG signals from the right and left hemispheres exhibit synchronized patterns and similar power across different frequency bands, especially in healthy subjects during sleep or rest. This phenomenon is also known as bilateral symmetry of EEG signals. Bilateral symmetry is a characteristic of resting or sleep states that reflects the synchronization of brain activity between the two hemispheres.

[0055] Figure 3 An example flowchart 300 is shown illustrating the processing of physiological signals 305 and the extraction of physiological features according to some embodiments of the present disclosure. After receiving physiological signals 305, for example, from a sensor patch or one or more databases 120, further processing and analysis can be performed on a computing device 115, such as... Figure 3 As shown. Physiological signals 305 (e.g., EEG signals) can be processed using data preprocessor 330. Data preprocessor 330 includes modules such as preprocessing 310, segmentation 315, transformation 320, and feature extraction 325.

[0056] At preprocessing 310, the physiological signal 305 can be processed to remove noise and other signal artifacts. During preprocessing 310, the physiological signal 305 can be selectively processed to remove artifacts, where an artifact refers to any portion of the physiological signal 305 that distorts the expected received data. These artifacts may be caused by factors such as high-frequency noise from muscle activity (e.g., clenching the teeth or head movements), periodic disturbances from cardiac electrical activity, or other environmental artifacts (e.g., electromagnetic interference), thus affecting the accuracy of the recorded physiological data. These artifacts can be removed from the physiological signal 305, for example, by automatically filtering the physiological signal 305 via filtering (e.g., DC filtering), ICA, or data smoothing techniques.

[0057] Physiological signal 305 can also be preprocessed using component analysis, which decomposes it into independent components and identifies and removes artifacts based on spatial and temporal features. Physiological data artifacts can also be removed by estimating the artifact shadow space, using methods such as Principal Component Analysis (PCA) and projecting the physiological signal 305 into an orthogonal subspace. In other instances, template matching can be performed to identify and remove known artifact patterns by comparing the physiological signal 305 with a predefined template. Furthermore, wavelet transform can be applied to decompose the physiological signal 305 into different frequency components and remove artifacts in specific frequency bands.

[0058] Following preprocessing 310, the physiological signal 305 (or EEG signal) can be segmented 315, dividing the signal (or continuous signal) into multiple time-series increments (also referred to herein as periods) of similar or different lengths. During segmentation 315, a scanning window can be used to further segment the time-series increments or periods into distinct parts, where the scanning window defines the different parts of the time-series increment (or period). The scanning window can be moved by skipping windows, thereby producing non-overlapping parts or segments. For example, a one-hour period or time-series increment of the physiological signal 305 can be scanned or segmented in 1-minute increments (i.e., a 1-minute scanning window), resulting in 60 non-overlapping or non-intersecting parts of the one-hour period. A sliding window can be used for the scanning window, where parts (or segments) of the sliding window may have overlapping time series. For example, a one-hour EEG signal can be scanned with a 1-minute scanning window starting every 30 seconds (i.e., a 30-second sliding window), resulting in overlapping 30-second 1-minute scanning windows. Alternatively, the entire time series of the EEG signal can correspond to a period.

[0059] Segments of the physiological signal 305 (e.g., which may include differential EEG signals and / or preprocessed EEG signals) can be transformed from the time domain to the frequency domain by the transform 320 module. For this purpose, the power spectrum can be calculated, for example, by calculating the power spectral density of each segment of the physiological signal 305 (e.g., the EEG signal). The power can be calculated using various techniques, such as multicone transform, Fourier transform, or wavelet transform. In some instances, for each segment of the EEG signal and for each hemisphere, one or more normalizations can be applied in the time and / or frequency domains by the transform 320 module (e.g., according to the SPEARS algorithm, disclosed in U.S. Application No. 11 / 431425, filed May 9, 2006, the entire contents of which are incorporated herein by reference for all purposes). The EEG signal can be adjusted to address power differences by performing normalization. For example, normalization can be performed by time-weighting the spectral power of one or more segments (or time intervals). The normalized power of each segment or time interval at one or more frequencies can help determine a suitable frequency window for extracting information. This normalization can reveal low power and statistically significant power variations at one or more frequency bands. Frequency bands can include those corresponding to Delta, Theta, Gamma, Alpha, Beta, or any other frequency range.

[0060] Physiological signals 305 (e.g., EEG signals) can be characterized by different frequency bands associated with specific cognitive and physiological states. For example, the Delta band, typically between 0.5 Hz and 4 Hz, is characterized by slow waves with high amplitude. Deep sleep (such as stage 3 of non-REM sleep that supports the recovery process) may be associated with the Delta band. Similarly, the Theta band (ranging from approximately [4–8] Hz) includes intermediate frequencies and amplitudes. Light sleep (such as stages 1 and 2 of non-REM sleep), drowsiness, meditation, or similar states may be associated with the Theta band. The Alpha band, ranging from approximately [8–12] Hz, can be characterized by intermediate frequencies and amplitudes lower than the Delta and Theta bands. Stage 2 of sleep can be characterized by the sleep axis, which typically occurs in the [12–15] Hz frequency range. Various states, such as relaxation and wakefulness with eyes closed, may be associated with the Alpha band. Furthermore, the Alpha band may facilitate the transition between wakefulness and sleep. Following the Alpha band is the Beta band, ranging from approximately [12-30] Hz, characterized by higher frequencies and lower amplitudes, which may be associated with active thinking, focus, wakefulness, or similar activities. The relatively higher Gamma band, ranging from approximately [30-100] Hz, is characterized by higher EEG signal frequencies and lower amplitudes. The Gamma band may be associated with higher information processing and perception, such as REM sleep, characterized by vivid dreams and active, wake-like brain activity. In some instances, gamma activity increases when subjects remain alert and engaged in a task, enhancing the brain's ability to focus, process information quickly, and maintain attention—essential for complex cognitive functions such as problem-solving, memory recall, and awareness. Increased gamma activity is typically observed when subjects are fully focused or deeply engaged in tasks requiring higher thinking and concentration. By processing the spectral characteristics of these bands, brain activity labels (e.g., different sleep stages, resting states, wakefulness, etc.) can be assigned to segments of the EEG signal.

[0061] Within these frequencies, one or more frequency bands can be identified and used for further analysis. Feature extraction can be performed on each segment of the physiological signal. Therefore, one or more features can be defined, which may include or be based on the power (or normalized power) in the transformed signal of each of the one or more frequency bands. One or more features may include statistics determined based on one or more power values ​​or weighted power values. For example, features may include the maximum or minimum power (or normalized power) in the spectrum corresponding to a segment, or the standard deviation of power (across frequency bands), etc. As another example, features may include the standard deviation (or weighted power value) of power values ​​across segments associated with a given frequency band. As yet another example, features may include z-scores, which may include normalized units reflecting the amount of power of the signal relative to the signal mean. z-scores can be converted to a mean deviation form by subtracting the mean from each score. These scores can then be normalized relative to the standard deviation. Units normalized by z-scores may have a standard deviation equal to 1.

[0062] Features can be calculated period-by-period using each of the data from one or more periods. As an illustration, features can be defined as including normalized power in low-frequency bands (e.g., Delta, Theta, Alpha bands), normalized power in high-frequency bands (e.g., Gamma bands), the standard deviation of normalized power values ​​across bands within a period, the maximum normalized power value for a period, etc. Furthermore, derived features can be generated based on the information (or normalized features) calculated for each of the data from one or more periods. Derived features can include, but are not limited to, Gamma power / Delta power, Gamma power / Alpha power, the time derivative of Delta, the time derivative of Gamma power / Delta power, and the time derivative of Gamma power / Alpha power. The time derivatives for the previous and next periods can be calculated. The derived features can then be normalized across one or more periods. Various data normalization techniques can be used, including z-scores, min-max scaling, quantile transformations, logarithmic transformations, and other similar techniques. In some instances, normalization is performed using z-scores, a statistical technique that standardizes the range of an independent variable (or feature). It may involve transforming the feature so that its mean is zero and its standard deviation is one. By applying z-scores, different derived features of spectral power data (e.g., Delta power and Gamma / Delta power) can be scaled to a common range, thus eliminating bias.

[0063] Figure 4An example implementation of a sleep analyzer 405 performing sleep stage analysis according to some embodiments of the present disclosure is shown. The sleep analyzer 405 may have direct access to one or more databases 120. The sleep analyzer 405 can assess a subject's recent sleep patterns, quality, and / or sleep duration by accessing physiological data from one or more previous nights or rest periods in one or more databases 120. The sleep analyzer 405 may include a data retrieval unit 410, a data preprocessor 230, and a sleep classifier 415. In some instances, the sleep analyzer 405 may analyze physiological data from rest periods or awake states to assess and interpret a subject's engagement and attention levels in an activity, task, or environment. In some other instances, neural signals and / or eye fixation may be used to predict a subject's level of attention to a situation while awake.

[0064] In some instances, the data retrieval unit 410 can retrieve or obtain physiological data of a subject from one or more databases 120. In other instances, the data retrieval unit 410 can receive data in real-time or near real-time while the subject is sleeping, resting, or awake (e.g., from a sensor patch) for further analysis on the computing device 115. After retrieving the physiological data and metadata, the data can be transferred to the data preprocessor 330. The metadata may include recording duration, recording date, time, electrode placement or location, etc. The data preprocessor 330 can further prepare the physiological data for subsequent analysis. The data preprocessor 330 can clean the data to remove noise or artifacts and segment the data into time windows or periods (e.g., 30 seconds, 2 minutes, 5 minutes, etc.). These segments can then be transformed and features extracted as needed, highlighting important physiological indicators related to sleep stages, which can be used for further downstream analysis.

[0065] The preprocessed physiological data can then be input into a sleep classifier 415. The sleep classifier 415 can categorize each period or segment into one of several predefined sleep stages, including wakefulness, REM sleep, and non-REM sleep stages (such as SWS, stage I, and stage II). Wakefulness can be further divided into quiet wakefulness and active wakefulness stages. The sleep classifier 415 can also detect microawakenings and / or microsleeps, as well as specific markers. The sleep classifier 415 can assess parameters such as the duration of each sleep stage, transitions between stages, and the overall structure of sleep, which can help detect pathological conditions.

[0066] In some embodiments, each segment of physiological signal 305 (or physiological data) may be assigned a sleep stage or state (including wakefulness) according to U.S. Application No. 11 / 431,425 (incorporated herein by reference for all purposes). Furthermore, a sleep classifier 415 may be used to determine the sleep pattern, relative frequency, and duration of each of one or more sleep stages or wakefulness states. The sleep classifier 415 may use a machine learning model. Machine learning models may include, but are not limited to, regression techniques (e.g., linear regression, multinomial regression) or classification techniques such as decision trees, random forests, support vector machines, neural networks, or deep learning models. Training of these models typically uses large labeled datasets obtained from synthetic or augmented data, public sleep datasets, or polysomnography studies. Standard datasets for sleep pattern analysis may include EEG, EOG, EMG, and often additional channels such as ECG or EKG. Physiological signal 305 may include preprocessed physiological data that may be further used to create a labeled dataset for training the model. The dataset used for training the sleep classifier 415 may include physiological data labeled with the correct sleep stages, which may be determined by experts through manual scoring or by automatic scoring. In some instances, unsupervised techniques such as clustering techniques (e.g., K-means clustering, hierarchical clustering, or Gaussian mixture models) can be used to classify each fragment of the physiological signal 305.

[0067] The output of the sleep classifier 415 or sleep analyzer 405 is referred to herein as sleep structure 420. Sleep structure 420 can provide the overall organization, broader structure, and sleep patterns of sleep over a monitored time period (or time period), including the proportion of time occupied by each stage and the progression of the sleep-wake cycle. Sleep analyzer 405 can generate outputs included in sleep structure 420, such as sleep scores, preference or dominant frequency analysis results, wakefulness detection results, microsleep detection results, marker detection results, spectral or temporal fragmentation results, spectral or temporal features, or sleep maps that divide sleep into stages such as wakefulness, REM, and non-REM (including SWS, stage I, and stage II sleep). Sleep structure 420 may also include average or maximum time between wakefulness, or wakefulness score, reflecting the overall quality of sleep. By analyzing sleep quality, duration, and patterns, sleep analyzer 405 can also calculate metrics such as average time between wakefulness.

[0068] According to some embodiments, sleep structure 420 may include sleep scores for each sleep-wake cycle or for the entire night's sleep, intervals between detected awakenings or microawakes, average sleep duration, and / or a sleep graph. Sleep scores may be numerical indicators of sleep quality and may take into account factors such as the duration, depth, and consistency of sleep cycles. Awakening detection results may include records of events in which the subject briefly wakes up or is disturbed during sleep, which may negatively impact sleep quality. A sleep graph can visually represent changes in sleep stages over time, dividing sleep into different stages such as periods of full wakefulness, REM sleep, and non-REM sleep, including SWS, stage I, and stage II sleep. The sleep graph may also include microawakes, microsleeps, etc. In some instances, single-channel EEG data may be retrieved from one or more databases 120 and used to perform sleep stage analysis to generate sleep structure 420.

[0069] Figure 5 An example implementation of apnea detection 515 for a subject is shown, utilizing the output of a sleep analyzer 405 (or sleep structure 420), EEG data 505, and data 510 from one or more non-EEG sensors. The EEG data 505 corresponds to the EEG signal collected from the subject during sleep and is used by the sleep analyzer 405 to generate the sleep structure 420. Arousal detection results can be retrieved from the sleep structure 420 and can indicate brief or micro-arousal moments in the EEG data 505.

[0070] By using brief or micro-awake moments as reference points, the apnea detection 515 can identify abnormalities in the EEG signals around these points that may indicate apnea (e.g., OSA or hypopnea). OSA typically causes brief awakenings, which can be detected by sudden changes in the EEG signal. Similarly, the apnea detection 515 can analyze changes in EEG signal patterns that may occur during an apnea episode. For example, EEG signals or brain activity may exhibit brief bursts of high-frequency activity, amplitude variations during an apnea event, or the elimination of one or more sleep stages.

[0071] In some embodiments, the apnea detection 515 may utilize one or more non-EEG sensor data 510 to inform the detection and / or severity assessment of apnea events. For example, the non-EEG sensor data 510 may be combined with EEG data 505 using one or more models and / or processing techniques (e.g., such that the input dataset includes one or more features or embeddings generated using EEG data 505, and one or more features or embeddings generated using non-EEG data 510). As another example, EEG data 505 and non-EEG sensor data 510 may be processed at different stages of the workflow (e.g., such that one or more features or embeddings generated using EEG data 505 are first processed using a first model or technique to generate a predicted probability or severity of apnea events, and then one or more features or embeddings generated using non-EEG data 510 are processed using a second model or technique to fine-tune, adjust, and / or confirm such predictions).

[0072] In some instances, apnea detection 515 can generate apnea risk based on EEG data 505, sleep structure 420, or one or more non-EEG sensor data 510. One or more non-EEG sensor data 510 can include, for example, EKG sensor data, pulse oximeter data, audio data, video data, chest movement data, or nasal airflow data. Audio data can be collected using microphones on devices such as smartwatches, smartphones, tablets, and laptops. Similarly, video data can be collected using user devices such as smartphones and tablets. One or more non-EEG sensor data 510 can be acquired in conjunction with or simultaneously with EEG data 505. A pulse oximeter measures blood oxygen levels and is typically placed on a finger or earlobe. Pulse oximeters can help identify a decrease in blood oxygen saturation due to apnea or hypoventilation. Additionally, nasal airflow data can identify periods of cessation of breathing (apnea) or reduction in airflow (hypoventilation). Furthermore, chest movement data can be used to assess breathing effort and can be acquired using a chest movement band containing sensors such as accelerometers and strain gauges. The apnea detection 515 can correlate changes in EEG data 505 at or before micro-awakening moments with one or more non-EEG sensor data 510 (if any), such as pulse oxygen saturation, heart rate, nasal airflow, and chest movement. Therefore, the apnea detection 515 can confirm sleep disruption caused by breathing disorders.

[0073] In some instances, sleep apnea detection515 can employ machine learning algorithms to classify and / or predict apnea events (e.g., OSA, hypoventilation) based on EEG features. These algorithms can be trained on labeled datasets of known OSA events to learn patterns associated with the disease (i.e., OSA).

[0074] After detecting apnea events in EEG data 505 corresponding to a specific time period (e.g., multiple previous nighttime time periods), apnea detection 515 can determine the duration and severity of each apnea event. Apnea detection 515 can also calculate the frequency, average severity, maximum severity, average duration, maximum duration, etc., of apnea events. The frequency of apnea events can be calculated per hour of sleep data (or EEG data 505), per sleep-wake cycle, per night, or per time period.

[0075] Apnea outcome 520 may include all the aforementioned values ​​calculated or determined by apnea detection 515, such as timestamps, severity, and duration of apnea and hypopnea episodes. Furthermore, apnea detection 515 may further analyze timestamps to assess whether apnea events occur randomly during the sleep-wake cycle, are more frequent in specific sleep stages (e.g., SWS, REM sleep, etc.), or vary throughout the night. Additionally, apnea outcome may include the Apnea-Hypopnea Index (AHI), which quantifies the number of apnea and hypopnea episodes per hour of sleep. In some instances, based on one or more criteria, such as if the frequency, duration, or severity of apnea events exceeds a corresponding threshold, apnea detection 515 may screen a subject for apnea or hypopnea and may alert the subject or caregiver. AHI can help classify the severity of OSA. An AHI of 5–15 episodes per hour is considered mild OSA, 15–30 episodes per hour is moderate OSA, and an AHI greater than 30 episodes per hour is classified as severe OSA.

[0076] Figure 6 An example diagram is shown illustrating the processing of EEG data 505 from the left and right hemispheres to detect stroke (or hemorrhagic stroke), small stroke, TIA, aneurysm, SBI, brain tumor, or TBI in a subject's brain. In some instances, EEG data 505 retrieved from one or more databases 120 can be separated based on the location of the EEG electrodes in the subject's head using metadata. EEG data from the first hemisphere 605a (e.g., the left hemisphere) and the second hemisphere 605b (e.g., the right hemisphere) can be processed separately using data preprocessors 330a and 330b. Both data preprocessors 330a-b include similar modules such as preprocessing 310, segmentation 315, transformation 320, and feature extraction 325.

[0077] EEG data (or signals) for each hemisphere 605a-605b can be processed by data preprocessors 330a-b to remove noise and artifacts (e.g., muscle activity, eye movement) from the EEG signals. The EEG data for each hemisphere 605a-605b can be segmented into periods or segments and transformed to the frequency domain. In some instances, one or more frequency bands (e.g., Delta, Theta, Alpha, Beta, Gamma) can be identified, and the feature set described in Figure 3 can be extracted from the EEG signals.

[0078] Subsequently, an interhemispheric comparison 610 can be performed on the EEG data from the first hemisphere 605a and the second hemisphere 605b to analyze or quantify transhemispheric asymmetry that may indicate neurodegenerative abnormalities (e.g., SBI, TIA, stroke, TBI, or tumor). In some embodiments, for the interhemispheric comparison 610, transhemispheric coherence can be calculated to take advantage of the bilateral symmetry of brain activity during sleep or resting states. Coherence can provide valuable insights into the functional connectivity and synchronicity of brain activity between different regions and hemispheres. Coherence values ​​range from 0 to 1. “0” indicates that there is no coherence or synchronicity between the two EEG signals (e.g., EEG data from the first hemisphere 605a and EEG data from the second hemisphere 605b), and the signals are completely independent of each other. Coherence values ​​close to “0” or low coherence may reflect weakened interhemispheric connectivity, communication, or a lack of synchronicity between the two EEG signals. Similarly, a value of "1" represents ideal coherence, which can refer to the situation where signals are perfectly synchronized with each other and have the same frequency and phase relationship. A high coherence value (close to 1) indicates effective and coordinated communication between hemispheres, or good synchronization between the two hemispheres.

[0079] In resting or sleep states, EEG data 505 or signals exhibit bilateral symmetry. Neurodegenerative abnormalities are typically unilateral; therefore, attenuation or deterioration of transhemispheric coherence can indicate stroke, small stroke, TIA, SBI, brain tumor, or TBI. Transhemispheric coherence can be calculated for each segment or period of EEG data (or signals) for each hemisphere 605a-605b, or even for each frequency at each time point. In some instances, transhemispheric coherence can be calculated using the output of sleep analyzer 405 (e.g., sleep structure 420) corresponding to EEG data (or signals) for each hemisphere 605a-605b, such as sleep-wake cycles or different sleep states. In some examples, coherence in different frequency bands (e.g., Delta, Theta, Alpha, Beta, Gamma) can also be calculated to reveal different aspects of brain activity. Coherence can be calculated using a variety of techniques, but not limited to: amplitude squared coherence (MSC), cross spectral density (CSD) (e.g., for general coherence analysis), short-time Fourier transform (STFT), and wavelet transform (e.g., for analyzing time and frequency coherence).

[0080] In some other embodiments, for the interhemispheric comparison 610, multiple feature sets (or feature clusters) can be used, which are extracted separately for each time interval, segment, or period of the EEG signal and for each of the two hemispheres using data preprocessors 330a-b. These feature sets can represent the temporal characteristics of neural activity and can be used to quantify the differences between the left and right hemispheres.

[0081] In some instances, to analyze asymmetry or abnormality in neural activity between the two hemispheres of the brain, a multidimensional distribution can be generated using feature sets corresponding to each hemisphere. Statistical analysis can be performed on these multidimensional distributions to determine whether they originate from the same underlying distribution or whether there are significant differences indicative of neurodegenerative abnormalities. Statistical analysis can utilize one or more statistical tests, such as the Kolmogorov-Smirnov test, the t-test, or the Wilcoxon signed-rank test. A statistical test yields a p-value, which quantifies the probability that the observed differences between the distributions are due to chance. A smaller p-value (e.g., less than 0.05) may indicate asymmetry in neural activity between the hemispheres and the presence of neurodegenerative abnormalities.

[0082] In some other instances, clustering techniques can be applied to determine asymmetry between EEG data from the first hemisphere 605a and the second hemisphere 605b. Clustering techniques can include, but are not limited to, k-means clustering, hierarchical clustering, or Gaussian mixture models (GMMs). Each feature set corresponding to a given period and a given hemisphere (representing EEG data from either the first hemisphere 605a or the second hemisphere 605b) can be assigned to a cluster. Clustering can group similar feature vectors, thereby identifying patterns and similarities within and between the first and second hemispheres. After the clustering process, the clustering or the clustering results can be used to quantify the degree of correlation or difference in neural activity patterns between the two hemispheres. A probabilistic measure or index of asymmetry can be calculated by comparing the distributions of feature vectors assigned to clusters in the first and second hemispheres. This comparison may involve statistical indicators such as the Kolmogorov-Smirnov test, which assesses the difference in feature vector distributions. To quantify clustering asymmetry, probability values ​​or scores can be derived based on the degree of difference in clusters between the left (e.g., first) and right (e.g., second) hemispheres. The greater the difference between the two hemispherical clusters, the lower the similarity score and the higher the probability value (indicating a significant difference in distribution). The probability value may correspond to neurodegenerative abnormalities.

[0083] In some other instances, the feature sets or EEG data of the first hemisphere 605a and the second hemisphere 605b can be projected into a multidimensional space, where each point represents a feature vector from either the left or right hemisphere. For inter-hemispheric comparisons 610, one or more inter-hemispheric distance metrics can be calculated based on the distances between feature vectors from different hemispheres. These metrics can be obtained by calculating various distance measurements, such as Euclidean distance, Mahalanobis distance (covariance between features), or cosine similarity (measured angular difference). These metrics capture the similarity or difference between feature vectors at corresponding points in different hemispheres. Furthermore, one or more intra-hemispheric distance metrics can be calculated within each hemisphere to reflect the distances between feature vectors from the same hemisphere. These intra-hemispheric metrics provide a baseline for similarity comparisons with inter-hemispheric distances.

[0084] These distance metrics can be used to develop a comprehensive score. One approach is to define a score that includes interhemispheric distance (reflecting asymmetry) and intrahemispheric distance (reflecting bilateral similarity). For example, the ratio or difference between interhemispheric and intrahemispheric distances can quantify the degree of asymmetry; a larger ratio or difference indicates more severe interhemispheric asymmetry and a higher risk of neurodegenerative abnormalities. Furthermore, statistical techniques, such as calculating z-scores or normalized distances relative to a reference distribution (e.g., healthy controls), can further refine the assessment of asymmetry. Statistical normalization can account for individual differences and improve the sensitivity of detecting abnormal asymmetry patterns that suggest neurodegenerative diseases.

[0085] Anomaly detection 615 can be used to analyze the results of interhemispheric comparison 610, such as coherence patterns, p-values, probabilities, or scores, and to detect the presence of neurodegenerative abnormalities in the subject's brain. For example, p-values ​​generated using statistical analysis can be compared to p-value thresholds. Smaller p-values ​​(e.g., p-values ​​less than 0.05) may indicate interhemispheric neural activity asymmetry and the presence of neurodegenerative abnormalities. Furthermore, by comparing a composite score to a predefined threshold, anomaly detection 615 can determine the presence and severity of interhemispheric neural activity asymmetry. Additionally, probability values ​​(e.g., from cluster analysis) can be compared to predefined probability thresholds. Probability values ​​greater than the probability threshold indicate the presence of neurological dysfunction or lesions. For healthy subjects (i.e., without neurodegenerative abnormalities), clustering can reveal that feature sets extracted from left and right hemisphere EEG signals form cohesive clusters, indicating bilateral symmetry in neural activity patterns.

[0086] After detecting the presence of a neurodegenerative abnormality, anomaly detection 615 can further detect or identify the presence or likelihood of hemorrhagic stroke, small stroke, TIA, SBI, aneurysm, brain tumor, or TBI based on the results of apnea, apnea risk, sleep changes, and interhemispheric comparison 610 (e.g., coherence, p-value, probability value, score, etc.). In some embodiments, anomaly detection 615 can compare coherence values ​​or patterns (e.g., in different sleep-wake cycles, different sleep stages, time derivatives, or patterns) with baseline data from a control group or healthy individual. Anomaly detection 615 can utilize sleep structure 420 to retrieve one or more biomarkers, such as indications of apnea, apnea risk, sleep changes (e.g., decreased SWS), etc. Anomaly detection 615 can utilize sleep structure 420 and interhemispheric comparison results to differentiate between different neurodegenerative abnormalities. For example, brain occlusion can alter coherence. Tumors often locally slow down EEG velocity, thus disproportionately affecting high-frequency coherence. In addition, TBI disrupts sleep patterns, typically leading to reduced SWS (stage 3) time and alterations in stage 2 sleep. If these abnormalities are not observed (e.g., tumor- and TBI-related abnormalities), and apnea and SWS suppression are present in addition, and the subject does not exhibit any impairment, the anomaly detection 615 can identify or detect an aneurysm or SBI. If the subject exhibits impairment, it could be a mini-stroke, TIA, or stroke. The anomaly detection 615 can also perform longitudinal analysis of the subject's past or historical data to identify each type of neurodegenerative abnormality, such as stroke, mini-stroke, TIA, SBI, brain tumor, or TBI.

[0087] In some other embodiments, EEG data 505 (or EEG data from the first hemisphere 605a and the second hemisphere 605b) may be collected from multiple subjects in a prospective study design. At baseline, multiple subjects may represent individuals at stroke risk who have no history of stroke, mini-stroke, TIA, or SBI. Subjects with three or more stroke risk factors on a stroke risk scorecard may be considered at stroke risk and may be selected. Stroke risk factors on a stroke risk scorecard may include: blood pressure greater than 120 / 80 mm / Hg; a confirmed diagnosis of atrial fibrillation; blood glucose greater than 100 mg / dL; and BMI greater than 25 kg / m². 2 High levels of saturated fat, trans fat, sugary drinks, salt, and excessive calories in the diet; total blood cholesterol greater than 160 mg / dL; diagnosed with diabetes; less than 150 minutes of moderate to vigorous activity per week; smoking; age between 40 and 75 years; not diagnosed with dementia or cognitive impairment that would prevent participation in prospective studies; family history of stroke, transient ischemic attack, or heart attack.

[0088] EEG data 505 and neuroimaging examinations (e.g., MRI, CT) can be collected at baseline and at one or more future time intervals. Neuroimaging results can serve as the basis for classifying individuals (or EEG data 505) into two or more groups, such as healthy individuals (or controls) and affected individuals. Affected individuals can be further classified into stroke individuals, minor stroke (or TIA) individuals, and SBI individuals. Interhemispheric comparison 610 results, apnea results 520, and sleep structure 420 can be calculated for these groups using the corresponding EEG signals and used to train a machine learning model. Anomaly detection 615 can utilize the trained machine learning model to screen individuals who may have recently developed SBI or minor stroke. The machine learning model can include, but is not limited to, deep learning models, Transformer models, decision trees (DT), random forests (RF), support vector machines (SVM), neural networks (NN), etc.

[0089] Furthermore, the anomaly detection 615 can output comparison and detection results 620. Comparison and detection results 620 may include values ​​indicating the presence, likelihood, or severity of stroke, small stroke, TIA, SBI, aneurysm, brain tumor, or TBI. Additionally, the results of interhemispheric comparisons 610 may also be included in comparison and detection results 620 for further analysis, for example, by a clinician. In some instances, one or more biomarkers based on sleep structure 420 or apnea results 520 may also be included in comparison and detection results 620.

[0090] Figure 7An example architecture is shown that generates a stroke risk score for a subject using a predictive model 710. The predictive model 710 can predict the stroke risk score using the subject's sleep structure 420, apnea results 520, comparison and detection results 620, and risk factor data 705. Sleep structure 420, apnea results 520, and comparison and detection results 620 can be obtained based on the analysis of the subject's physiological signals 305 (or EEG data 505). One or more biomarkers can be identified from the sleep structure 420, apnea results 520, comparison and detection results 620, and risk factor data (e.g., other risk factors or non-EEG-based risk factors). One or more biomarkers may include indications of apnea or SBI. One or more biomarkers also include the presence of OSA, duration of OSA, intensity of OSA, or a reduction in SWS duration, a reduction in REM duration, the presence of one or more SBIs, detection of TIA, reduced interhemispheric concordance, etc.

[0091] Risk factor data 705 may be obtained from electronic health records (EHRs) that include the subject's historical medical information, with the consent of the subject or their guardian. Additionally, with the consent of the subject or their guardian, risk factor data 705 may also be obtained from the subject's devices that record their daily activities (e.g., smartwatches, smartphones). Risk factor data 705 may include the subject's medical information and activity indicators. Medical information may include, but is not limited to, BMI, diagnosis of atrial fibrillation, diagnosis of diabetes, history of blood glucose levels, blood pressure data, family history of stroke or transient ischemic attack, family history of heart attack, or blood cholesterol levels. Activity indicators may include, for example, weekly physical activity intensity.

[0092] Predictive model 710 can be trained on a dataset that can be used such as Figure 6The prospective study design explained herein is generated. Physiological signals 305 and neuroimaging data (e.g., MRI, CT) can be collected at baseline and at one or more future time intervals. Data on future stroke events (including minor stroke, TIA, SBI) can also be collected, and individuals can be categorized into affected or high-risk groups. In some instances, multiple groups can be defined based on risk factors, neuroimaging data, and clinical assessment. Physiological signals 305 can then be analyzed using the techniques disclosed in this disclosure to obtain one or more biomarkers. After training, the predictive model 710 can be used to generate a stroke risk score for the subject, preferably at home, for early stroke detection. In some instances, the predictive model 710 can generate a stroke risk score based on one or more biomarkers (e.g., duration, severity, frequency, etc. of the biomarker). In some other instances, risk factor data 705 (if available) can be used in conjunction with one or more biomarkers to predict a subject's stroke risk score. In other instances, the disclosed techniques can be used to assess the effectiveness of interventions or monitor the recovery of subjects who have experienced neurodegenerative diseases such as stroke.

[0093] Predictive models 710 may include machine learning models such as recurrent neural networks (RNNs), long short-term memory (LSTM) models, Transformer models, neural networks (NNs), multi-value prediction algorithms, deep learning models, or regression techniques. A stroke risk score indicates the likelihood or probability that a subject will experience a stroke or minor stroke within a future timeframe (e.g., the next week, month, months, or year). Stroke risk scores may include, for example, numbers (e.g., integers or real numbers chosen from a range such as 0-10) or categories. Categories may be mild, moderate, or high risk. In some cases, stroke risk scores may be binary and may be used to differentiate or screen individuals at risk of stroke.

[0094] In addition, the neurodegenerative abnormality detection system can perform one or more actions based on the subject's stroke risk score. One or more actions include, but are not limited to: generating an alarm, notifying relevant departments (e.g., medical personnel, relatives and / or the subject) to conduct a full medical evaluation to confirm the occurrence or risk of stroke, mini-stroke, TIA, or the presence of one or more SBIs or aneurysms.

[0095] Figure 8An example flowchart of a system for detecting the risk or occurrence of stroke or infarction in a subject, according to some embodiments of this disclosure, is shown. The blocks in flowchart 800 are shown in a specific order, but this order can be modified; for example, some blocks may be executed before others, and some blocks may be executed simultaneously. These blocks may be executed by hardware, software, or a combination thereof. The process at block 805 may include accessing physiological data of the subject collected by a physiological data acquisition component over a period of time. The physiological data may correspond to physiological signals (e.g., EEG, EMG, EOG, EKG, MEG) collected during the night or multiple previous nights. The physiological data acquisition component may include sensing devices and one or more electrode clusters. Each of the one or more electrode clusters includes at least an active electrode. In some instances, these electrodes may be EEG electrodes for recording EEG signals from the left and right hemispheres of the subject's brain.

[0096] At box 810, a sleep structure 420 can be generated by analyzing the subject's physiological data. In some instances, to generate the sleep structure 420, a feature set can first be extracted based on a portion of the physiological data corresponding to each of a plurality of time intervals within the time period. The state of each in the feature set corresponding to each time interval can be predicted. This state can correspond to any of one or more sleep stages or wakefulness states. One or more sleep stages can include REM stages and one or more non-REM stages (including SWS stages). Based on the predicted state for each time interval, the relative frequency and duration of sleep patterns, one or more sleep or wakefulness states (e.g., quiet wakefulness and active wakefulness) can be determined.

[0097] At box 815, one or more biomarkers can be determined at least in part by using the subject's sleep structure 420. In some instances, one or more biomarkers can also be determined based on a first and second portion of physiological data (such as EEG data 505 corresponding to the subject's left and right hemispheres). The coherence between the first and second portions of the physiological data can then be calculated. The presence or likelihood of hemorrhagic stroke, SBI, TIA, aneurysm, brain tumor, or TBI can be determined based on coherence, apnea, apnea risk, and sleep variability. Furthermore, one or more biomarkers may also include the presence of obstructive sleep apnea (OSA), the duration of OSA, the intensity of OSA, or a reduction in slow-wave sleep (SWS) duration.

[0098] At box 820, a predictive model 710 can be used to predict a subject's stroke risk score based on one or more biomarkers. In some embodiments, in addition to one or more biomarkers, the predictive model 710 may also utilize risk factor data 705 (based on non-EEG) to generate the stroke risk score. Risk factor data 705 may include the subject's medical information and activity indicators. The predictive model 710 may include a machine learning model, such as an RNN model, a Transformer model, a deep learning model, or a neural network.

[0099] The procedure at box 825 can determine whether a condition is met, at least in part, based on a stroke risk score. For example, the condition could be: whether the subject's (predicted) stroke risk is above a certain threshold; whether the stroke risk score indicates a high risk of future stroke; or whether the stroke risk score is moderate and there is apnea or a detected SBI (including a small stroke or TIA), etc.

[0100] Finally, at box 830, one or more actions can be triggered based on the fulfillment of certain conditions. These actions may include alerting the subject, alerting caregivers or clinicians, or outputting results or data that provide a basis for recommendations or include recommendations to perform assessments or interventions to reduce the stroke risk score. One or more actions may also include displaying the subject's results and / or stroke risk score on computing device 115, or transmitting the results to other devices. Output results or data may also include sleep structure 420, apnea results 520, and comparison and detection results 620 to facilitate further investigation by clinicians (whether human or not) to recommend treatment plans for the subject.

[0101] Figure 9 Example diagrams of a computing system 900 that can implement various embodiments of the present disclosure are shown. The computing system 900 can be used as... Figure 1 The computing device 115 described herein. For example, the techniques described above for detecting stroke risk or occurrence using physiological data (or EEG signals) can be implemented with computer-executable instructions (e.g., organized in program module 904). Program module 904 may include routines, programs, objects, components, and data structures of the data types required to perform tasks and implement the techniques described above. The functionality described herein can be performed at least in part by one or more hardware logic components.

[0102] To provide supplementary background on its various aspects, Figure 9The following description is intended to provide a brief overview of a computing system 900 that can implement the various aspects. While the above description is made in the general context of computer-executable instructions that can run on one or more computers, those skilled in the art will recognize that novel implementations can also be achieved in combination with other program modules and / or in a combination of hardware and software. The computing system 900 or computer system for implementing the various aspects includes a processing unit 908 having one or more processors (also referred to as microprocessors), a computer-readable storage medium (wherein the medium is any physical device or material that can store and retrieve data electronically and / or optically) such as a data storage unit 910 (computer-readable storage media / mediums also include disks, optical disks, solid-state drives, external memory systems, and flash drives), and a system bus 912. The system bus 912 can provide interfaces to the processing unit 908 for system components (including, but not limited to, system memory 914). This system bus 912 can be any of several types of bus architectures, which can be further interconnected to memory buses (with or without controllers) and peripheral buses (e.g., peripheral component interconnect (PCI), peripheral component interconnect high speed (PCIe), accelerated graphics port (AGP), low pin count (LPC), etc.), using any of a variety of commercially available bus architectures.

[0103] Figure 9 An example configuration of a typical computer is shown, which can be other commercially available microprocessors, such as single-processor, multi-processor, single-core, and multi-core processing and / or storage circuits. Furthermore, those skilled in the art will understand that novel systems and methods can be implemented using other computer system configurations, including minicomputers, mainframe computers, and personal computers (e.g., desktops, laptops, tablets, etc.), handheld computing devices, microprocessor-based or programmable consumer electronics, each of which can be cooperatively coupled to one or more associated devices.

[0104] In some aspects, computing system 900 may be one of multiple computers and / or computing resources (hardware and / or software) used in a data center to support cloud computing services for portable and / or mobile computing systems (such as wireless communication devices, cellular phones, and other mobile devices). Cloud computing services include, but are not limited to, Infrastructure as a Service (IaaS), Platform as a Service (PaaS), Software as a Service (SaaS), Storage as a Service (SaaS), Desktop as a Service (DAS), Data as a Service (DAS), Security as a Service (SAS), and API (Application Programming Interface) as a Service (API) as a Service. In some instances, system memory 914 may include computer-readable storage (physical storage) media such as volatile memory (e.g., random access memory (RAM) 916) and non-volatile memory (e.g., read-only memory (ROM) 918). The Basic Input / Output System (BIOS) may be stored in non-volatile memory, which includes basic routines for data and signal communication between components within computing system 900, such as during startup. Volatile memory also includes high-speed RAM for caching data, such as static RAM.

[0105] For example (but not limited to), system memory 914 may also include program module 904, which may include client applications, web browsers, middleware applications, relational database management systems (RDBMS), program data 906, and operating system 902. For example, operating system 902 may include various versions of Microsoft Windows®, Apple Macintosh®, and / or Linux operating systems, various commercially available UNIX® or UNIX-like operating systems (OS) (including but not limited to various Gnu's Not Unix (GNU) / Linux operating systems, Google Chrome OS, etc.) and / or mobile operating systems such as iOS, Windows® Phone, Android OS, BlackBerry® OS, and Palm® OS. All or part of operating system 902, program module 904, and / or program data 906 may also be cached in memory, such as volatile memory and / or non-volatile memory (e.g., RAM 916 or ROM 918). It should be understood that the disclosed architecture can be implemented using various commercially available operating systems or combinations of operating systems (e.g., virtual machines).

[0106] In some other examples, the computing system 900 may have other features or functions. For example, the computing system 900 may also include additional data storage devices (removable and / or non-removable), such as, for example, a hard disk, optical disk, or magnetic tape. Computer-readable media may include at least two types of computer-readable media: computer storage media and communication media. Computer storage media may include volatile and non-volatile, removable and non-removable media, which may be implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data.

[0107] System memory 914 and data storage 910 (including removable and non-removable storage) are examples of computer storage media. In addition to RAM 916 and ROM 918, computer storage media include, but are not limited to: electrically erasable programmable read-only memory (EEPROM), flash memory or other storage technologies, optical disc (CD)-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape cassettes, magnetic tape, disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store target information and is accessible by computing system 900. Furthermore, computer-readable media may also include computer-executable instructions that, when executed by processing unit 908, can perform the various functions and / or operations described herein. Communication media may embody computer-readable instructions, data structures, program modules, or other data in modulated data signals (e.g., carrier waves or other transmission mechanisms).

[0108] The computing system 900 may also include one or more input devices 920, such as a keyboard, mouse, pen, voice input device, touch input device, etc. It may also include one or more output devices 922, such as a display, speaker, printer, etc. These devices are well known in the art and will not be described further herein. The computing system 900 may also include one or more network interfaces 924 for establishing communication, enabling the computing system 900 to communicate with other systems or devices, for example, via a network. These networks may include wired networks and wireless networks. Here, the computing system 900 is an example of a suitable device or system and is not intended to impose any limitation on the scope or functionality of the various embodiments described.

[0109] Other well-known computer systems, environments, and / or configurations suitable for use with the implementation scheme include, but are not limited to, personal computers, server computers, handheld or laptop devices, multiprocessor systems, microprocessor-based systems, set-top boxes, game consoles, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the aforementioned systems or devices. For example, some or all components of computing system 900 may be implemented in a cloud computing environment, thereby providing resources and / or services via computer networks for selective use by user devices.

[0110] Some embodiments of this disclosure include a system comprising one or more data processors. In some embodiments, the system includes a non-transitory computer-readable storage medium comprising instructions that, when executed on the one or more data processors, cause the one or more data processors to perform some or all of the one or more methods and / or some or all of the one or more processes disclosed herein. Some embodiments of this disclosure include a computer program product tangibly embodied in a non-transitory machine-readable storage medium, the computer program product including instructions configured to cause the one or more data processors to perform some or all of the one or more methods and / or some or all of the one or more processes disclosed herein.

[0111] The terminology and expressions used are for descriptive purposes only and not for limitation, and their use is not intended to exclude any equivalents of the shown and described features or portions thereof. However, it should be understood that various modifications can be made within the scope of the claims of this invention. Therefore, it should be understood that although the claims of this invention have been specifically disclosed through embodiments and optional features, those skilled in the art can modify and alter the concepts disclosed herein, and such modifications and alterations should be considered within the scope of the invention as defined by the appended claims.

[0112] This specification provides only preferred embodiments and is not intended to limit the scope, applicability, or configuration of this disclosure. Rather, the description of preferred embodiments is intended to provide those skilled in the art with an illustration of the feasibility of implementing various embodiments. It should be understood that various changes can be made to the function and arrangement of the elements without departing from the spirit and scope set forth in the appended claims.

[0113] The following description provides specific details to provide a comprehensive understanding of the implementation scheme. However, it should be understood that the implementation scheme can be implemented without these specific details. For example, circuits, systems, networks, processes, and other components may be shown in block diagram form to avoid unnecessary details that could obscure the implementation scheme. In other cases, to avoid making the implementation scheme difficult to understand, some well-known circuits, processes, algorithms, structures, and techniques may omit unnecessary details.

Claims

1. A computer-implemented method, comprising: Access to physiological data of a subject collected by a physiological data acquisition component over a period of time, wherein the physiological data acquisition component includes a sensing device and one or more electrode clusters, and wherein each of the one or more electrode clusters includes at least an active electrode; Sleep structure is generated by analyzing the physiological data; At least in part, it is based on determining one or more biomarkers by using the sleep structure, wherein the one or more biomarkers include indications of apnea, apnea risk, sleep changes, or transhemispheric changes; The subject's stroke risk score is predicted using a predictive model based on one or more biomarkers; At least in part based on the fact that the stroke risk score determination criteria are met; and One or more actions are triggered based on the determination that the conditions are met, wherein the one or more actions include alerting the subject, alerting a caregiver or clinician, or outputting results that provide a basis for recommendations or include recommendations to perform assessments or interventions to reduce stroke risk scores.

2. The computer-implemented method according to claim 1, wherein generating the sleep structure comprises: For each of the multiple time intervals within the stated time period, a feature set is extracted based on a portion of the physiological data; Predict the state of each in the feature set corresponding to each time interval, wherein the state corresponds to any one or more sleep stages or wakefulness states, and wherein the one or more sleep stages include rapid eye movement (REM) stages and one or more non-REM stages; and Determined based on predicted state: Sleep mode; The relative frequency of each of the one or more sleep stages or the waking state; and The duration of each of the one or more sleep stages or the waking state.

3. The computer-implemented method according to claim 1, wherein determining one or more biomarkers further comprises: Determine a first portion of the physiological data corresponding to the first hemisphere of the subject's brain; Determine the second portion of the physiological data corresponding to the second hemisphere of the subject's brain; Calculate the coherence between the first and second portions of the physiological data; as well as Based on the apnea, the apnea risk, the sleep changes, and the coherence, determine the presence or likelihood of hemorrhagic stroke, asymptomatic stroke (SBI), transient ischemic attack (TIA), aneurysm, brain tumor, or traumatic brain injury (TBI).

4. The computer-implemented method of claim 1, wherein the one or more biomarkers further include the presence of obstructive sleep apnea (OSA), the duration of OSA, the intensity of OSA, or a reduction in the duration of slow-wave sleep (SWS), the presence of SBI, or the detection of TIA.

5. The computer-implemented method of claim 1, wherein the stroke risk score is further predicted based on risk factor data including the subject's medical information and activity indicators, wherein the medical information includes body mass index (BMI) value, atrial fibrillation diagnosis, diabetes diagnosis, blood glucose level history, blood pressure data, family history of stroke or TIA, family history of heart attack, or blood cholesterol level, wherein the activity indicators include weekly physical activity intensity.

6. The computer-implemented method of claim 1, wherein the physiological data corresponds to physiological signals collected during nighttime, rest periods, or multiple previous nighttimes or rest periods.

7. The computer-implemented method of claim 1, wherein the prediction model comprises a machine learning model, the machine learning model comprising a recurrent neural network (RNN), a transformer model, or a neural network.

8. The computer-implemented method of claim 1, wherein the one or more electrode clusters comprise electroencephalography (EEG) electrodes, electromyography (EMG) electrodes, magnetoencephalography (MEG) electrodes, or electrooculography (EOG) electrodes.

9. A system comprising: One or more data processors; as well as A non-transitory computer-readable storage medium including instructions that, when executed on one or more data processors, cause the one or more data processors to perform a set of operations, including: Access to physiological data of a subject collected by a physiological data acquisition component over a period of time, wherein the physiological data acquisition component includes a sensing device and one or more electrode clusters, and wherein each of the one or more electrode clusters includes at least an active electrode; Sleep structure is generated by analyzing physiological data; At least in part, it is based on determining one or more biomarkers by using the sleep structure, wherein the one or more biomarkers include indications of apnea, apnea risk, sleep changes, or transhemispheric changes; The subject's stroke risk score is predicted using a predictive model based on one or more biomarkers; At least in part based on the fact that the stroke risk score determination criteria are met; and One or more actions are triggered based on the determination that the conditions are met, wherein the one or more actions include alerting the subject, alerting a caregiver or clinician, or outputting results that provide a basis for recommendations or include recommendations to perform assessments or interventions to reduce stroke risk scores.

10. The system of claim 9, wherein generating the sleep structure comprises: For each of the multiple time intervals within the stated time period, a feature set is extracted based on a portion of the physiological data; Predict the state of each in the feature set corresponding to each time interval, wherein the state corresponds to any one or more sleep stages or wakefulness states, and wherein the one or more sleep stages include rapid eye movement (REM) stages and one or more non-REM stages; and Determined based on predicted state: Sleep mode; The relative frequency of each of the one or more sleep stages or the waking state; and The duration of each of the one or more sleep stages or the waking state.

11. The system of claim 9, wherein determining one or more biomarkers further comprises: Determine a first portion of the physiological data corresponding to the first hemisphere of the subject's brain; Determine the second portion of the physiological data corresponding to the second hemisphere of the subject's brain; Calculate the coherence between the first and second portions of the physiological data; and Based on the apnea, the apnea risk, the sleep changes, and the coherence, determine the presence or likelihood of hemorrhagic stroke, asymptomatic stroke (SBI), transient ischemic attack (TIA), aneurysm, brain tumor, or traumatic brain injury (TBI).

12. The system of claim 9, wherein the one or more biomarkers further include the presence of obstructive sleep apnea (OSA), the duration of OSA, the intensity of OSA, or a reduction in the duration of slow-wave sleep (SWS), the presence of SBI, or the detection of TIA.

13. The system of claim 10, wherein the stroke risk score is further predicted based on risk factor data including the subject's medical information and activity indicators, wherein the medical information includes body mass index (BMI) value, atrial fibrillation diagnosis, diabetes diagnosis, blood glucose level history, blood pressure data, family history of stroke or TIA, family history of heart attack, or blood cholesterol level, wherein the activity indicators include weekly physical activity intensity.

14. The system of claim 9, wherein the physiological data corresponds to physiological signals collected during nighttime, rest periods, or multiple previous nighttimes or rest periods.

15. The system of claim 9, wherein the prediction model comprises a machine learning model, the machine learning model comprising a recurrent neural network (RNN), a transformer model, or a neural network.

16. A computer program product tangibly embodied in a non-transitory machine-readable storage medium, comprising instructions configured to cause one or more data processors to perform a set of operations, said operations including: Access to physiological data of a subject collected by a physiological data acquisition component over a period of time, wherein the physiological data acquisition component includes a sensing device and one or more electrode clusters, and wherein each of the one or more electrode clusters includes at least an active electrode; Sleep structure is generated by analyzing physiological data; At least in part, it is based on determining one or more biomarkers by using the sleep structure, wherein the one or more biomarkers include indications of apnea, apnea risk, sleep changes, or transhemispheric changes; The subject's stroke risk score is predicted using a predictive model based on one or more biomarkers; At least in part based on the fact that the stroke risk score determination criteria are met; and One or more actions are triggered based on the determination that the conditions are met, wherein the one or more actions include alerting the subject, alerting a caregiver or clinician, or outputting results that provide a basis for recommendations or include recommendations to perform assessments or interventions to reduce stroke risk scores.

17. The computer program product of claim 16, wherein generating the sleep structure comprises: For each of the multiple time intervals within the stated time period, a feature set is extracted based on a portion of the physiological data; Predict the state of each in the feature set corresponding to each time interval, wherein the state corresponds to any one or more sleep stages or wakefulness states, and wherein the one or more sleep stages include rapid eye movement (REM) stages and one or more non-REM stages; and Determined based on predicted state: Sleep mode; The relative frequency of each of the one or more sleep stages or the waking state; and The duration of each of the one or more sleep stages or the waking state.

18. The computer program product of claim 16, wherein determining one or more biomarkers further comprises: Determine a first portion of the physiological data corresponding to the first hemisphere of the subject's brain; Determine the second portion of the physiological data corresponding to the second hemisphere of the subject's brain; Calculate the coherence between the first and second portions of the physiological data; and Based on the apnea, the apnea risk, the sleep changes, and the coherence, determine the presence or likelihood of hemorrhagic stroke, asymptomatic stroke (SBI), transient ischemic attack (TIA), aneurysm, brain tumor, or traumatic brain injury (TBI).

19. The computer program product of claim 16, wherein the one or more biomarkers further include the presence of obstructive sleep apnea (OSA), the duration of OSA, the intensity of OSA, or a reduction in the duration of slow-wave sleep (SWS), the presence of SBI, or the detection of TIA.

20. The computer program product of claim 16, wherein the stroke risk score is further predicted based on risk factor data including the subject's medical information and activity indicators, wherein the medical information includes body mass index (BMI) value, atrial fibrillation diagnosis, diabetes diagnosis, blood glucose level history, blood pressure data, family history of stroke or TIA, family history of heart attack, or blood cholesterol level, wherein the activity indicators include weekly physical activity intensity.

Images