Description:FIELD OF DISCLOSURE
[0001] The present disclosure generally relates to the field of health monitoring system, and more specifically, relates to a wearable-based seizure forecasting and classification system and a method thereof.
BACKGROUND OF THE DISCLOSURE
[0002] Epileptic seizures impair a considerable percentage of the global population, and the early detection and control of seizure episodes are in direct relation to quality of life and patient safety. Seizure activity monitoring continuously facilitates early intervention, rational clinical decision-making, and enhanced patient outcomes. Wearable monitors have been proposed as an innovative method for offering real-time, unobtrusive monitoring and can alert patients and caregivers instantaneously upon the onset of seizures.
[0003] Existing seizure monitoring in the traditional setting is based mostly on in-clinic electroencephalography (EEG) recordings, periodic observation, or patient seizure diaries. Although useful for diagnosis, these are not practical for continuous, real-time monitoring in everyday life. Also, some wearable technologies for seizure monitoring target single physiological signals, e.g., heart rate or activity, but miss the subtle interplay of neural, physiological, and behavioral signals involved in seizure initiation.
[0004] Existing solutions have issues such as low accuracy of detection in ambulatory settings, delayed alerts to caregivers, excessive reliance on manual input, and non-adaptation to individual patient-specific seizure patterns. These issues make early intervention less effective and restrict patient independence and safety.
[0005] The current invention solves these challenges by offering an improved, real-time, and customized wearable seizure monitoring system intended to monitor relevant physiological signals continuously, enhance detection reliability under various conditions, and deliver timely warnings to caregivers or healthcare professionals. By facilitating personalized accommodation of the patient's unique profile, the system is envisioned to increase safety, enable proactive clinical management, and enhance overall patient quality of life.
[0006] Thus, in light of the above-stated discussion, there exists a need for a wearable-based seizure forecasting and classification system and a method thereof.
SUMMARY OF THE DISCLOSURE
[0007] The following is a summary description of illustrative embodiments of the invention. It is provided as a preface to assist those skilled in the art to more rapidly assimilate the detailed design discussion which ensues and is not intended in any way to limit the scope of the claims which are appended hereto in order to particularly point out the invention.
[0008] According to illustrative embodiments, the present disclosure focuses on a wearable-based seizure forecasting and classification system and a method thereof which overcomes the above-mentioned disadvantages or provide the users with a useful or commercial choice.
[0009] An objective of the present invention is to provide a system capable of real-time monitoring of neurological activity to support timely detection of abnormal events in a non-intrusive, user-friendly manner.
[0010] Another objective of the present invention is to enable personalized monitoring through adaptive processing that accounts for individual-specific characteristics and historical patterns.
[0011] Another objective of the present invention is to improve accuracy and reliability of event detection by analyzing signals over multiple dimensions and time scales, thereby reducing false positives and false negatives.
[0012] Yet another objective of the present invention is to facilitate continuous and long-term monitoring, allowing secure storage, retrieval, and trend analysis of patient-specific data to support preventive or responsive interventions.
[0013] Yet another objective of the present invention is to provide a system that can communicate alerts and insights efficiently to relevant devices or caregivers, ensuring timely response without disrupting patient comfort or daily activities.
[0014] Yet another objective of the present invention is to allow scalable and flexible integration with various sensor types and processing architectures, enhancing robustness under different operating conditions.
[0015] Yet another objective of the present invention is to improve interpretability and transparency of monitoring outcomes, enabling verification, validation, and informed decision-making by healthcare providers or caregivers.
[0016] In light of the above, in one aspect of the present disclosure, a wearable-based seizure forecasting and classification system is disclosed herein. The system comprises a user device configured to store patient data including but not limited to medical history, seizure logs, and prescribed medications. The system also includes an electroencephalography (EEG) acquisition unit comprising a plurality of electrodes and an amplifier, the EEG acquisition unit configured to be placed on the patient’s scalp and to capture real-time multichannel EEG signals. The system also includes a processing unit operatively connected to the user device and the EEG acquisition unit, the processing unit comprising a data input module configured to receive the captured EEG signals from the EEG acquisition unit and the patient data from the user device; a preprocessing module configured to filter and normalize the received signals from the data input module; a segmentation module configured to divide the EEG signals into time windows suitable for real-time analysis; a feature extraction module configured to extract signal features from the segmented EEG signals; a temporal analysis module configured to analyze temporal dependencies across the extracted features from multiple EEG channels and time windows; a classification module configured to analyse the temporally processed features to detect onset of a seizure event; an explainability module configured to generate interpretable reports of the classification results, including influential EEG regions, intervals, and frequency bands; a notification module configured to transmit real-time message pop-ups, alerts, and reports generated by the processing unit to the user device; and an output module configured to provide seizure detection results and analysis summary to the user device. The system also includes a communication unit configured to enable secure data exchange between the user device, the EEG acquisition unit and the processing unit.
[0017] In one embodiment, the system further comprises a cloud database operatively connected to the user device via the communication unit, the cloud database configured to store and retrieve long-term patient data including multichannel EEG recordings, seizure occurrence logs, prescribed medications, and clinician feedback for enabling continuous patient-specific model adaptation.
[0018] In one embodiment, the processing unit is further configured to adapt classification thresholds dynamically based on patient data including medical history, seizure frequency patterns, and prescribed medications received from the user device, thereby enabling personalized seizure detection.
[0019] In one embodiment, the preprocessing module is configured to remove artifacts caused by eye blinks, muscle activity, or powerline interference using adaptive filtering and wavelet-based denoising techniques, thereby improving accuracy of seizure onset detection.
[0020] In one embodiment, the feature extraction module is configured to perform time–frequency transformation to generate spectral representations of EEG signals using at least one of a short-time Fourier transform (STFT) or a continuous wavelet transform (CWT), thereby enabling robust seizure onset feature extraction across time and frequency domains.
[0021] In one embodiment, the temporal analysis module comprises a multi-head self-attention mechanism configured to concurrently capture long-range temporal dependencies and inter-channel correlations across EEG signals, thereby improving robustness of seizure onset prediction.
[0022] In one embodiment, the classification module employs spiking neural network models with leaky integrate-and-fire neurons configured to encode temporal patterns of EEG signals into sparse spike trains for energy-efficient real-time computation.
[0023] In one embodiment, the classification module is configured to detect and classify multiple seizure states including preictal, ictal, postictal, and interictal phases, based on temporal and spectral features extracted from the EEG signals, thereby enabling multiclass seizure cycle prediction for personalized patient monitoring and intervention.
[0024] In one embodiment, the explainability module is configured to generate clinician-readable reports indicating influential EEG channels, critical time windows, and discriminative frequency bands contributing to seizure prediction, and to provide interpretable outputs including channel-wise importance scores and temporal heatmaps for verification of seizure onset detection.
[0025] In light of the above, in another aspect of the present disclosure, a method for a wearable-based seizure forecasting and classification system is disclosed herein. The method comprises storing patient data, including medical history, seizure logs, and prescribed medications, on a user device. The method also includes capturing multichannel electroencephalography (EEG) signals via an EEG acquisition unit comprising plurality of electrodes and an amplifier positioned on a patient’s scalp. The method also includes receiving the EEG signals and patient data at a processing unit via a data input module. The method also includes preprocessing the EEG signals via a preprocessing module to filter, normalize, and remove artifacts caused by eye blinks, muscle activity, or powerline interference. The method also includes segmenting the preprocessed EEG signals into time windows suitable for real-time analysis via a segmentation module. The method also includes extracting signal features from the segmented EEG signals using a feature extraction module, including spectral representations via at least one of a short-time Fourier transform (STFT) or a continuous wavelet transform (CWT). The method also includes analyzing temporal dependencies across the extracted features via a temporal analysis module comprising a multi-head self-attention mechanism. The method also includes classifying the temporally processed features via a classification module to detect seizure onset and classify multiple seizure states including preictal, ictal, postictal, and interictal phases. The method also includes generating interpretable reports of the classification results via an explainability module, including channel-wise importance scores, temporal heatmaps, and influential EEG regions, intervals, and frequency bands. The method also includes transmitting real-time alerts and notifications via a notification module to the user device, including wearable vibration, audio notifications, and automatic emergency call initiation upon detection of an ictal state. The method also includes exchanging data securely between the user device, the EEG acquisition unit and the processing unit via a communication unit.
[0026] These and other advantages will be apparent from the present application of the embodiments described herein.
[0027] The preceding is a simplified summary to provide an understanding of some embodiments of the present invention. This summary is neither an extensive nor exhaustive overview of the present invention and its various embodiments. The summary presents selected concepts of the embodiments of the present invention in a simplified form as an introduction to the more detailed description presented below. As will be appreciated, other embodiments of the present invention are possible utilizing, alone or in combination, one or more of the features set forth above or described in detail below.
[0028] These elements, together with the other aspects of the present disclosure and various features are pointed out with particularity in the claims annexed hereto and form a part of the present disclosure. For a better understanding of the present disclosure, its operating advantages, and the specified object attained by its uses, reference should be made to the accompanying drawings and descriptive matter in which there are illustrated exemplary embodiments of the present disclosure.
BRIEF DESCRIPTION OF THE DRAWINGS
[0029] To describe the technical solutions in the embodiments of the present disclosure or in the prior art more clearly, the following briefly describes the accompanying drawings required for describing the embodiments or the prior art. Apparently, the accompanying drawings in the following description merely show some embodiments of the present disclosure, and a person of ordinary skill in the art can derive other implementations from these accompanying drawings without creative efforts. All of the embodiments or the implementations shall fall within the protection scope of the present disclosure.
[0030] The advantages and features of the present disclosure will become better understood with reference to the following detailed description taken in conjunction with the accompanying drawing, in which:
[0031] FIG. 1 illustrates a block diagram of a system for a wearable-based seizure forecasting and classification system, in accordance with an exemplary embodiment of the present disclosure;
[0032] FIG. 2 illustrates a flowchart of a method, outlining the sequential steps for a wearable-based seizure forecasting and classification system, in accordance with an exemplary embodiment of the present disclosure.
[0033] FIG. 3 illustrates a block diagram 300 of the seizure forecastion and classification system, depicting the work flow, in accordance with an exemplary embodiment of the present disclosure.
[0034] Like reference, numerals refer to like parts throughout the description of several views of the drawing.
[0035] A wearable-based seizure forecasting and classification system and a method thereof is illustrated in the accompanying drawings, which like reference letters indicate corresponding parts in the various figures. It should be noted that the accompanying figure is intended to present illustrations of exemplary embodiments of the present disclosure. This figure is not intended to limit the scope of the present disclosure. It should also be noted that the accompanying figure is not necessarily drawn to scale.
DETAILED DESCRIPTION OF THE DISCLOSURE
[0036] The following is a detailed description of embodiments of the disclosure depicted in the accompanying drawings. The embodiments are in such detail as to communicate the disclosure. However, the amount of detail offered is not intended to limit the anticipated variations of embodiments; on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present disclosure.
[0037] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the present disclosure. It may be apparent to one skilled in the art that embodiments of the present disclosure may be practiced without some of these specific details.
[0038] Various terms as used herein are shown below. To the extent a term is used, it should be given the broadest definition persons in the pertinent art have given that term as reflected in printed publications and issued patents at the time of filing.
[0039] The terms “a” and “an” herein do not denote a limitation of quantity, but rather denote the presence of at least one of the referenced items.
[0040] The terms “having”, “comprising”, “including”, and variations thereof signify the presence of a component.
[0041] Referring now to FIG. 1 to FIG. 3 to describe various exemplary embodiments of the present disclosure. FIG. 1 illustrates a block diagram of a wearable-based seizure forecasting and classification system, in accordance with an exemplary embodiment of the present disclosure.
[0042] The system 100 may include a user device 102, an electroencephalography (EEG) acquisition unit 104, a processing unit 110, a communication unit 130 and a cloud database 132.
[0043] The user device 102 configured to store patient data including but not limited to medical history, seizure logs, and prescribed medications. Such information is important for generating individualized seizure risk profiles, correlating detected patterns with prior episodes, and aligning alerts with prescribed therapeutic regimens. The inclusion of medical history supports long-term trend analysis, seizure logs assist in identifying preictal triggers and frequency, and medication records enable context-aware recommendations and physician review, thereby enhancing both accuracy of detection and continuity of care.
[0044] In a preferred embodiment, the user device 102 may be implemented as a smartphone, tablet, or a dedicated handheld unit, incorporating secure local storage and encryption modules to protect sensitive patient data. The user device 102 may employ database indexing to organize medical history, seizure logs, and medication schedules in a retrievable manner, enabling rapid access for real-time analysis and clinician review.
[0045] In another embodiment, the user device 102 may include a graphical user interface (GUI) for patients or caregivers to manually input seizure occurrences, medication adherence, or symptom notes. This manual entry, when correlated with automatically detected events, provides a more comprehensive seizure diary that supports accurate classification and treatment adjustments.
[0046] In a further embodiment, the user device 102 may synchronize with cloud storage or electronic health record (EHR) platforms through secure communication protocols, allowing physicians to remotely monitor seizure frequency, medication compliance, and treatment efficacy. This ensures continuity of care and enables timely medical interventions.
[0047] The electroencephalography (EEG) acquisition unit 104 comprising a plurality of electrodes 106 and an amplifier 108, the EEG acquisition unit 104 configured to be placed on the patient’s scalp and to capture real-time multichannel EEG signals. The multichannel EEG signals refers to electrical recordings obtained simultaneously from multiple electrodes positioned across different cortical regions, each channel representing localized neural activity over a specific scalp location. The multichannel design allows the system to capture both spatial and temporal variations in brain activity, enabling identification of seizure onset zones and propagation patterns with higher accuracy.
[0048] The electroencephalography (EEG) acquisition unit 104 comprises a plurality of electrodes 106 and an amplifier 108.
[0049] The plurality of electrodes 106 can comprise wet, dry, or semi-dry types of electrodes, designed for long-term wear and comfort of the patient. Wet electrodes have high signal fidelity, while dry or semi-dry electrodes minimize preparation time and enhance the convenience of use in ambulatory or home settings.
[0050] The amplifier 108 preserves signal quality through the amplification of low-amplitude brain activity and rejection of noise and motion artifacts, which is essential for accurate seizure classification.
[0051] In another embodiment, the EEG acquisition unit 104 can include adaptive gain control within amplifier 108 to dynamically adjust signal amplification in response to patient activity levels or electrode impedance changes. This eliminates false positives due to non-seizure-related artifacts such as blinking, jaw activity, or head movement.
[0052] In another embodiment, the EEG acquisition unit 104 can be configured as a headband, cap, or inconspicuous patch-type accessory, such that long-term wear may be carried out in clinical and home environments. Such form-factor versatility enables patient age groups, medical settings, and comfort factors, thus augmenting compliance with long-term seizure monitoring.
[0053] The processing unit 110 operatively connected to the user device 102 and the EEG acquisition unit 104. The processing unit 110 is configured to receive the EEG signals and patient data, analyze the signals, and produce seizure detection results in real time. The processing unit 110 facilitates combining physiological inputs with patient history information to enable accurate detection and timely notification.
[0054] In one embodiment, the processing unit 110 can have a modular design with input handling, signal conditioning, analysis, and output reporting submodules. This modularity enables the system 100 to be adjusted for various patient profiles and usage modes while ensuring real-time responses.
[0055] In one preferred embodiment, the processing unit 110 can further provide explainable outputs to better enable clinicians to comprehend the decision process so as to enhance trust and clinical approval of the resultant outputs.
[0056] In one embodiment, the EEG acquisition unit 104 and the processing unit 110 are integrated within a wearable head-mounted device, thereby enabling mobile monitoring without tethering to external equipment.
[0057] The processing unit 110 is further configured to adapt classification thresholds dynamically based on patient data including medical history, seizure frequency patterns, and prescribed medications received from the user device 102, thereby enabling personalized seizure detection. Such dynamic adaptation ensures that the detection model is not restricted to fixed population-level thresholds, but rather calibrated for individual variability. For example, a patient with high baseline neural activity or frequent false-positive episodes may have thresholds adjusted upward, while a patient with infrequent but severe seizures may have more sensitive thresholds. This personalization increases reliability, reduces false alarms, and aligns detection performance with the patient’s clinical profile.
[0058] In a preferred embodiment, the processing unit 110 is implemented as an edge deployment unit configured to execute the analysis in real time with reduced latency and power consumption, thereby enabling integration into wearable or implantable systems. In alternative embodiments, the processing unit may be realized as a local computing device or a remote server connected via a communication interface.
[0059] The processing unit 110 having several specialized modules including a data input module 112, a preprocessing unit 114, a segmentation module 116, a feature extraction module 118, a temporal analysis module 120, a classification module 122, an explainability module 124, a notification module 126 and an output module 128.
[0060] The data input module 112 configured to receive the captured EEG signals from the EEG acquisition unit 104 and the patient data from the user device 102. The data input module 112 synchronizes heterogeneous inputs by correlating real-time multichannel EEG signals with relevant contextual patient information like medical records, seizure history, and medications. Such correlation allows the following modules to relate signal anomalies with patient-specific clinical context, hence enhancing detection quality.
[0061] In a further embodiment, the data input module 112 could comprise buffering and timestamping facilities to handle sampling rate fluctuations or delays in transmission between the EEG acquisition unit 104 and the user device 102. Synchronization in this way preserves temporal continuity upon fusing biosignal streams with stored patient information.
[0062] In a preferred embodiment, the data input module 112 may perform secure data parsing and validation procedures to guarantee data integrity of arriving data packets, rejecting corrupted or incomplete signals before directing them to the preprocessing module. This eliminates noise at the initial processing stage and maintains computational efficiency.
[0063] In another form, the data input module 112 can embed an encryption and authentication layer in interfacing with the user device 102, protecting sensitive patient information from being transmitted and accessed by unauthorized users.
[0064] The preprocessing module 114 configured to filter and normalize the received signals from the data input module 112. The preprocessing unit 114 guarantees that raw multichannel EEG signals are converted into a stable and noise-minimized version that can be processed reliably by downstream analytical modules. Normalization is used to ensure comparability in amplitude ranges between channels and sessions, allowing for real-time and long-term analysis of comparable signals.
[0065] The preprocessing module 114 is configured to remove artifacts caused by eye blinks, muscle activity, or powerline interference using adaptive filtering and wavelet-based denoising techniques, thereby improving accuracy of seizure onset detection. These preprocessing steps address the common challenge of EEG contamination by non-neural sources. Eye blinks and facial muscle contractions introduce low-frequency transients, muscle activity contributes broadband noise, and powerline interference produces narrowband 50/60 Hz noise. Artifact removal preserves seizure-relevant neural patterns and strengthens downstream feature extraction and classification. Wavelet denoising selectively suppresses such artifacts while retaining seizure-related transients, with wavelet thresholding enabling multiresolution decomposition to separate short artifacts from rhythmic brain activity.
[0066] In one embodiment, the preprocessing module 114 may utilize band-pass filters such as 0.5–40 Hz to remove baseline drift and high-frequency noise and retain the dominant frequency bands of seizures.
[0067] In another embodiment, the application of a notch filter (50/60 Hz) to specifically eliminate powerline interference, an omnipresent artifact in ambulatory and clinical EEG recordings.
[0068] In a preferred embodiment, the preprocessing module 114 can apply adaptive filtering algorithms like least-mean-square (LMS) or recursive least squares (RLS) filters, adapting filter weights in real-time based on continuous input to cancel time-varying artifacts like muscle contractions or electrode impedance changes.
[0069] The segmentation module 116 configured to divide the EEG signals into time windows suitable for real-time analysis. Adequate segmentation guarantees that seizure-related transient events are observed within computationally tractable frames without compromising temporal resolution. By dividing the streaming EEG into overlapping or constant-length windows, the system is able to balance sensitivity to seizure onset with computational efficiency, allowing timely processing.
[0070] In a preferred embodiment, the segmentation module 116 can use sliding window methods with tunable overlap ratios (e.g., 25–50%) in order to maintain continuity of temporal dynamics between consecutive segments. Overlapping avoids loss of important ictal onset characteristics that might otherwise straddle window boundaries.
[0071] In still another embodiment, the segmentation module 116 can be made adaptive to window length in accordance with patient-specific seizure dynamics or signal features in order to optimize the latency-robustness trade-off. Short windows can capture fleeting preictal transitions, while longer windows enable robust frequency-domain feature extraction.
[0072] In another embodiment, segmentation can be synchronized with other biosignal indicators (such as heart rate variability or accelerometer signals) in order to give multimodal context to the detection of abnormal events. This cross-modal synchronization enhances temporal alignment between EEG activity and physiological counterparts, thus enhancing seizure onset prediction.
[0073] The feature extraction module 118 configured to extract signal features from the segmented EEG signals.
[0074] The feature extraction module 118 is configured to perform time–frequency transformation to generate spectral representations of EEG signals using at least one of a short-time Fourier transform (STFT) or a continuous wavelet transform (CWT), thereby enabling robust seizure onset feature extraction across time and frequency domains. The STFT method partitions each window of EEG into small subframes and takes a Fourier transform, capturing localized frequency changes corresponding to ongoing seizure activity. This allows detection of brief ictal spikes and rhythmic discharges. The continuous wavelet transform (CWT) may be applied to break down EEG signals into multiple scales to facilitate multiresolution analysis of abrupt transients and oscillations with long durations. The wavelet basis functions offer flexibility to capture the non-stationarity of EEG and enhance robustness to noise and artifacts.
[0075] In yet another embodiment, the feature extraction module 118 can compute additional statistical and nonlinear descriptors like entropy, fractal dimension, and energy distribution over frequency bands. The features act as complementary markers to spectral representations so that better discrimination of preictal, ictal, postictal, and interictal states can be made.
[0076] In one embodiment, the feature extraction module 118 can integrate spectral, temporal, and statistical features into a common representation, allowing downstream temporal analysis and classification modules to perform with greater sensitivity and specificity.
[0077] The temporal analysis module 120 configured to analyze temporal dependencies across the extracted features from multiple EEG channels and time windows. Temporal modeling is necessary in seizure prediction because epileptic discharges tend to appear as developing patterns over time rather than as unique signal aberrations.
[0078] The temporal analysis module 120 comprises a multi-head self-attention mechanism configured to concurrently capture long-range temporal dependencies and inter-channel correlations across EEG signals, thereby improving robustness of seizure onset prediction. The self-attention mechanism enables the model to assign weights from remote time steps and multiple channels in parallel, such that essential dependencies are not lost during analysis. The attention-based deep learning approach that captures sequential dependencies of EEG features at multiple scales, allowing it to identify subtle preictal transitions leading up to clinical seizures.
[0079] In another embodiment, the multi-head attention mechanism is designed to learn independent attention weights for different EEG channels and time windows and hence retain local signal dynamics (short-term ictal bursts) and global dynamics (long-term preictal buildup). This adds phase-wise seizure classification.
[0080] In one additional embodiment, the temporal analysis module 120 can be coupled with recurrent frameworks like gated recurrent units (GRU) or long short-term memory (LSTM) networks to supplement attention processes. Such a hybrid modeling enhances stability when processing variable-length EEG recordings and ensures continuity of tracking temporal relationships.
[0081] In one preferred embodiment, the temporal analysis module 120 adjusts attention weights dynamically according to patient-specific EEG profiles, drawing on stored seizure history data in the system to emphasize signal regions that exhibited greater predictive significance. Personalization enhances reliability across patients with heterogeneous seizure phenomenology.
[0082] The classification module 122 configured to analyse the temporally processed features to detect onset of a seizure event.
[0083] The classification module 122 employs spiking neural network models with leaky integrate-and-fire neurons configured to encode temporal patterns of EEG signals into sparse spike trains for energy-efficient real-time computation. This event-driven representation reduces redundant computations, offering energy-efficient real-time processing suitable for wearable and embedded platforms.
[0084] The classification module 122 is configured to detect and classify multiple seizure states including preictal, ictal, postictal, and interictal phases, based on temporal and spectral features extracted from the EEG signals, thereby enabling multiclass seizure cycle prediction for personalized patient monitoring and intervention. Such multiclass seizure cycle prediction allows personalized patient monitoring and timely intervention beyond binary seizure/no-seizure detection.
[0085] In another embodiment, the classification module 122 can be trained using patient-specific datasets synchronized with seizure logs and medication history from the user device 102, thereby enabling adaptive recalibration of the decision boundaries over time to account for individual variability in seizure patterns.
[0086] In a further embodiment, the classification module 122 may employ hybrid architectures, wherein the SNN layer is combined with attention-based or convolutional layers, allowing extraction of both fine-grained temporal spikes and spatial EEG features. This enhances robustness against noise while maintaining energy efficiency.
[0087] The explainability module 124 configured to generate interpretable reports of the classification results, including influential EEG regions, intervals, and frequency bands. These reports emphasize the spatial brain regions most associated with seizure onset, the critical time segments where abnormal activity emerges, and the dominant spectral components that distinguish seizure states, thus offering clinicians an understandable rationale for the system’s predictions.
[0088] The explainability module 124 is configured to generate clinician-readable reports indicating influential EEG channels, critical time windows, and discriminative frequency bands contributing to seizure prediction, and to provide interpretable outputs including channel-wise importance scores and temporal heatmaps for verification of seizure onset detection. Such reports highlight which electrodes most strongly contributed to the prediction, when in the EEG sequence abnormal dynamics were most relevant, and which spectral ranges differentiated seizure activity, thereby enabling clinicians to validate and trust the automated system’s decisions
[0089] In a preferred embodiment, interpretability is obtained through Transformer attention weight analysis, where attention scores identify EEG channels and temporal windows most indicative of preictal or ictal states. In another mode, gradient-weighted class activation mapping (Grad-CAM) is used for convolutional layers of the model to yield heatmaps correlating spectral features with seizure activity. These multimodal explainabilities allow clinicians to verify predictions, build greater trust in automated results, and improve integration with clinical workflows.
[0090] The notification module 126 configured to transmit real-time message pop-ups, alerts, and reports generated by the processing unit 110 to the user device 102.
[0091] In one embodiment, the notification unit 126 sends alerts in multimodal manner, such as vibration on a wearable device, audio cues on a caregiver device, and automatic calling of emergency numbers of predefined contacts upon the detection of an ictal state. In yet another embodiment, the module has filtering by seizure severity or prediction confidence, so important ictal events cause interventions to be triggered immediately, while less dangerous preictal conditions create warnings. In yet another embodiment, the notification module 126 is integrated with secure mobile applications or hospital information systems so that it can communicate quickly with caregivers and clinicians, thus facilitating quick medical response.
[0092] The output module 128 configured to provide seizure detection results and analysis summary to the user device 102.
[0093] In a preferred embodiment, the output module 128 displays clear detection results, i.e., seizure onset flags, seizure phase classification, and prediction confidence scores, in patient-, caregiver-, or clinician-friendly format. In an alternative embodiment, the module supplies summary analyses such as trend reports, distributions of seizure frequencies, and patient-specific indicators of progression to aid long-term monitoring and adjustment of treatment. In an additional embodiment, the output module 128 can be configured to vary presentation levels for instance, simpler alerts for patients and more comprehensive analytics for healthcare providers thus balancing usability and clinical applicability.
[0094] The communication unit 130 configured to enable secure data exchange between the user device 102, the EEG acquisition unit 104 and the processing unit 110. The communication unit 130 facilitates smooth bidirectional transmission of EEG signals, patient records, and classification results while maintaining confidentiality and integrity of sensitive medical data. In a preferred embodiment, the unit could utilize wireless protocols like Bluetooth Low Energy (BLE) or Wi-Fi Direct for low-latency transmission of EEG signals from the acquisition unit 104 to the processing unit 110 to eliminate tethering and enhance patient mobility. In another embodiment, encryption modules based on AES or TLS protocols could be used in the communication unit 130 to ensure protection of patient data against unauthorized access during transmission. In another embodiment, the communication unit 130 can provide hybrid connectivity, where short-range wireless connections are employed for near-area communication and secure Internet protocols facilitate synchronization with cloud databases or remote clinician dashboards. Such integration allows both real-time monitoring and long-term data access and retention while maintaining medical-grade privacy and compliance regulations.
[0095] The system 100 further comprises a cloud database 132 operatively connected to the user device 102 via the communication unit 130, the cloud database 132 configured to store and retrieve long-term patient data including multichannel EEG recordings, seizure occurrence logs, prescribed medications, and clinician feedback for enabling continuous patient-specific model adaptation. In a particular implementation, the cloud database 132 might be realized on a HIPAA-compliant, secure server platform that accommodates encrypted storage and role-based access controls to protect sensitive health data. In another implementation, the cloud database 132 can store indexed datasets of EEG recordings synchronized over long periods, allowing clinicians to monitor trends of seizure frequency, treatment effectiveness, and medication compliance. In a further embodiment, the cloud database 132 can allow integration with electronic health record (EHR) systems, thus enabling smooth sharing of annotated EEG data sets and clinical documentation with approved healthcare professionals. Having longitudinal data stored in the cloud database 132 enables ongoing machine learning model adjustments, wherein the latest seizure patterns and drug responses are included to optimize detection thresholds and prediction values for specific patients.
[0096] The method 200 may include storing patient data, including medical history, seizure logs, and prescribed medications, on a user device 102. The method 200 may also include capturing multichannel electroencephalography (EEG) signals via an EEG acquisition unit 104 comprising plurality of electrodes 106 and an amplifier 108 positioned on a patient’s scalp. The method 200 may also include receiving the EEG signals and patient data at a processing unit 110 via a data input module 112. The method 200 may also include preprocessing the EEG signals via a preprocessing module 114 to filter, normalize, and remove artifacts caused by eye blinks, muscle activity, or powerline interference. The method 200 may also include segmenting the preprocessed EEG signals into time windows suitable for real-time analysis via a segmentation module 116. The method 200 may also include extracting signal features from the segmented EEG signals using a feature extraction module 118, including spectral representations via at least one of a short-time Fourier transform (STFT) or a continuous wavelet transform (CWT). The method 200 may also include analyzing temporal dependencies across the extracted features via a temporal analysis module 120 comprising a multi-head self-attention mechanism. The method 200 may also include classifying the temporally processed features via a classification module 122 to detect seizure onset and classify multiple seizure states including preictal, ictal, postictal, and interictal phases. The method 200 may also include generating interpretable reports of the classification results via an explainability module 124, including channel-wise importance scores, temporal heatmaps, and influential EEG regions, intervals, and frequency bands. The method 200 may also include transmitting real-time alerts and notifications via a notification module 126 to the user device 102, including wearable vibration, audio notifications, and automatic emergency call initiation upon detection of an ictal state. The method 200 may also include exchanging data securely between the user device 102, the EEG acquisition unit 104 and the processing unit 110 via a communication unit 130.
[0097] FIG. 2 illustrates a flowchart of a method 200, outlining the sequential steps for a wearable-based seizure forecasting and classification system, 100 in accordance with an exemplary embodiment of the present disclosure.
[0098] At step 202, store patient data, including medical history, seizure logs, and prescribed medications, on a user device 102.
[0099] At step 204, capture multichannel electroencephalography (EEG) signals via an EEG acquisition unit 104 comprising electrodes 106 and an amplifier 108 positioned on a patient’s scalp.
[0100] At step 206, receive the EEG signals and patient data at a processing unit 110 via a data input module 112.
[0101] At step 208, preprocess the EEG signals via a preprocessing module 114 to filter, normalize, and remove artifacts caused by eye blinks, muscle activity, or powerline interference.
[0102] At step 210, segment the preprocessed EEG signals into time windows suitable for real-time analysis via a segmentation module 116.
[0103] At step 212, extract signal features from the segmented EEG signals using a feature extraction module 118, including spectral representations via at least one of a short-time fourier transform (STFT) or a continuous wavelet transform (CWT).
[0104] At step 214, analyze temporal dependencies across the extracted features via a temporal analysis module 120 comprising a multi-head self-attention mechanism.
[0105] At step 216, classify the temporally processed features via a classification module 122 to detect seizure onset and classify multiple seizure states including preictal, ictal, postictal, and interictal phases.
[0106] At step 218, generate interpretable reports of the classification results via an explainability module 124, including channel-wise importance scores, temporal heatmaps, and influential EEG regions, intervals, and frequency bands.
[0107] At step 220, transmit real-time alerts and notifications via a notification module 126 to the user device 102, including wearable vibration, audio notifications, and automatic emergency call initiation upon detection of an ictal state.
[0108] At step 222, exchange data securely between the EEG acquisition unit 104, the processing unit 110, and the user device 102 via a communication unit 130.
[0109] FIG. 3 illustrates a block diagram 300 of the seizure forecastion and classification system, depicting the work flow, in accordance with an exemplary embodiment of the present disclosure.
[0110] At 302, a plurality of EEG electrodes, preferably arranged according to the international 10–20 system, are positioned on the patient’s scalp to capture electrical activity from different cortical regions. The electrodes are configured to measure multichannel EEG signals reflecting underlying brain dynamics.
[0111] At 304, the EEG acquisition unit receives the raw electrical signals from the electrodes and amplifies them to levels suitable for digital processing. The unit ensures low-noise signal conditioning and maintains fidelity of multichannel recordings.
[0112] At 306, a signal preprocessing module performs filtering and segmentation of the EEG signals. The module removes common artifacts such as eye blinks, muscle noise, and powerline interference, while dividing the continuous data into suitable time windows for real-time analysis.
[0113] At 308, a transformer-based self-attention module analyzes temporal dependencies across EEG segments, capturing long-range patterns and inter-channel correlations to improve robustness of seizure onset prediction.
[0114] At 310, a spiking convolutional neural network (CNN) module encodes temporal patterns of EEG signals into sparse spike trains using leaky integrate-and-fire neurons. This design reduces computational overhead and enhances energy-efficient real-time processing.
[0115] At 312, a classification layer integrates outputs from both the spiking CNN and transformer modules to detect seizure onset and classify seizure states, including preictal, ictal, postictal, and interictal phases, thereby enabling multiclass seizure cycle prediction.
[0116] At 314, an explainability module (XAI) generates interpretable reports of the classification results, highlighting influential EEG channels, time intervals, and discriminative frequency bands. The module provides channel-wise importance scores and temporal heatmaps to assist clinicians in verifying automated predictions.
[0117] At 316, real-time alerts and user interface outputs are transmitted to patient or caregiver devices, such as dashboards, smartphone applications, or wearable interfaces. Notifications may include visual alerts, vibration feedback, audio alarms, or automatic emergency call initiation, ensuring timely intervention during seizure onset.
[0118] In best mode of operation, the system 100 operates by initially positioning the electroencephalography (EEG) acquisition unit 104, consisting of a plurality of electrodes 106 and an amplifier 108, on the scalp of the patient to acquire real-time multichannel EEG signals from different cortical areas. These inputs, along with patient information like medical history, seizure history, and medications dispensed onto the user device 102, are input to the processing unit 110 by the data input module 112. The preprocessing module 114 normalizes and filters the signals, eliminating artifacts like muscle activity, eye blinks, and powerline interference, after which the segmentation module 116 segments the purged signals into time windows for real-time analysis. The feature extraction module 118 thereafter extracts signal features, such as spectral representations in time and frequency domains, which are input into the temporal analysis module 120 using a multi-head self-attention mechanism to identify dependencies between channels and windows. The classification module 122 processes these temporally processed features using spiking neural networks and affiliated models to identify seizure onset and identify seizure phases such as preictal, ictal, postictal, and interictal states. After classification, the explainability module 124 produces understandable reports presenting influential EEG areas, key time periods, and discriminative frequency bands, with outputs like channel-wise importance scores and temporal heatmaps for clinician validation. The notification module 126 sends real-time notifications, such as wearable vibration, audio alarm, or emergency call initiation, while the output module 128 displays analysis summaries and detection results to the user device 102. Any data transfer between the EEG acquisition unit 104, the processing unit 110, and the user device 102 is protected by the communication unit 130, and long-term EEG recordings, seizure logs, and clinician responses are maintained in the cloud database 132 to allow ongoing patient-specific adaptation of classification thresholds, hence guaranteeing robust, personalized, and clinically meaningful seizure monitoring in real-world applications.
[0119] While the invention has been described in connection with what is presently considered to be the most practical and various embodiments, it will be understood that the invention is not to be limited to the disclosed embodiments, but on the contrary, is intended to cover various modifications and equivalent arrangements included within the scope of the appended claims.
[0120] A person of ordinary skill in the art may be aware that, in combination with the examples described in the embodiments disclosed in this specification, units and algorithm steps may be implemented by electronic hardware, computer software, or a combination thereof.
[0121] The foregoing descriptions of specific embodiments of the present disclosure have been presented for purposes of illustration and description. They are not intended to be exhaustive or to limit the present disclosure to the precise forms disclosed, and many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described to best explain the principles of the present disclosure and its practical application, and to thereby enable others skilled in the art to best utilize the present disclosure and various embodiments with various modifications as are suited to the particular use contemplated. It is understood that various omissions and substitutions of equivalents are contemplated as circumstances may suggest or render expedient, but such omissions and substitutions are intended to cover the application or implementation without departing from the scope of the present disclosure.
[0122] Disjunctive language such as the phrase “at least one of X, Y, Z,” unless specifically stated otherwise, is otherwise understood with the context as used in general to present that an item, term, etc., may be either X, Y, or Z, or any combination thereof (e.g., X, Y, and/or Z). Thus, such disjunctive language is not generally intended to, and should not, imply that certain embodiments require at least one of X, at least one of Y, or at least one of Z to each be present.
[0123] In a case that no conflict occurs, the embodiments in the present disclosure and the features in the embodiments may be mutually combined. The foregoing descriptions are merely specific implementations of the present disclosure, but are not intended to limit the protection scope of the present disclosure. Any variation or replacement readily figured out by a person skilled in the art within the technical scope disclosed in the present disclosure shall fall within the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure shall be subject to the protection scope of the claims.
, Claims:I/We Claim:
1. A wearable-based seizure forecasting and classification system (100), the system (100) comprising:
a user device (102) configured to store patient data including but not limited to medical history, seizure logs, and prescribed medications;
an electroencephalography (EEG) acquisition unit (104) comprising a plurality of electrodes (106) and an amplifier (108), the EEG acquisition unit (104) configured to be placed on the patient’s scalp and to capture real-time multichannel EEG signals;
a processing unit (110) operatively connected to the user device (102) and the EEG acquisition unit (104), the processing unit (110) comprising:
a data input module (112) configured to receive the captured EEG signals from the EEG acquisition unit (104) and the patient data from the user device (102);
a preprocessing module (114) configured to filter and normalize the received signals from the data input module (112);
a segmentation module (116) configured to divide the EEG signals into time windows suitable for real-time analysis;
a feature extraction module (118) configured to extract signal features from the segmented EEG signals;
a temporal analysis module (120) configured to analyze temporal dependencies across the extracted features from multiple EEG channels and time windows;
a classification module (122) configured to analyse the temporally processed features to detect onset of a seizure event;
an explainability module (124) configured to generate interpretable reports of the classification results, including influential EEG regions, intervals, and frequency bands;
a notification module (126) configured to transmit real-time message pop-ups, alerts, and reports generated by the processing unit (110) to the user device (102); and
an output module (128) configured to provide seizure detection results and analysis summary to the user device (102);
a communication unit (130) configured to enable secure data exchange between the user device (102), the EEG acquisition unit (104) and the processing unit (110).
2. The system (100) as claimed in claim 1, wherein the system (100) further comprises a cloud database (132) operatively connected to the user device (102) via the communication unit (130), the cloud database (132) configured to store and retrieve long-term patient data including multichannel EEG recordings, seizure occurrence logs, prescribed medications, and clinician feedback for enabling continuous patient-specific model adaptation.
3. The system (100) as claimed in claim 1, wherein the processing unit (110) is further configured to adapt classification thresholds dynamically based on patient data including medical history, seizure frequency patterns, and prescribed medications received from the user device (102), thereby enabling personalized seizure detection.
4. The system (100) as claimed in claim 1, wherein the preprocessing module (114) is configured to remove artifacts caused by eye blinks, muscle activity, or powerline interference using adaptive filtering and wavelet-based denoising techniques, thereby improving accuracy of seizure onset detection.
5. The system (100) as claimed in claim 1, wherein the feature extraction module (118) is configured to perform time–frequency transformation to generate spectral representations of EEG signals using at least one of a short-time Fourier transform (STFT) or a continuous wavelet transform (CWT), thereby enabling robust seizure onset feature extraction across time and frequency domains.
6. The system (100) as claimed in claim 1, wherein the temporal analysis module (120) comprises a multi-head self-attention mechanism configured to concurrently capture long-range temporal dependencies and inter-channel correlations across EEG signals, thereby improving robustness of seizure onset prediction.
7. The system (100) as claimed in claim 1, wherein the classification module (122) employs spiking neural network models with leaky integrate-and-fire neurons configured to encode temporal patterns of EEG signals into sparse spike trains for energy-efficient real-time computation.
8. The system (100) as claimed in claim 1, wherein the classification module (122) is configured to detect and classify multiple seizure states including preictal, ictal, postictal, and interictal phases, based on temporal and spectral features extracted from the EEG signals, thereby enabling multiclass seizure cycle prediction for personalized patient monitoring and intervention.
9. The system (100) as claimed in claim 1, wherein the explainability module (124) is configured to generate clinician-readable reports indicating influential EEG channels, critical time windows, and discriminative frequency bands contributing to seizure prediction, and to provide interpretable outputs including channel-wise importance scores and temporal heatmaps for verification of seizure onset detection.
10. A method (200) for wearable-based seizure forecasting and classification system (100), the method (200) comprising:
storing patient data, including medical history, seizure logs, and prescribed medications, on a user device (102);
capturing multichannel electroencephalography (EEG) signals via an EEG acquisition unit (104) comprising plurality of electrodes (106) and an amplifier (108) positioned on a patient’s scalp;
receiving the EEG signals and patient data at a processing unit (110) via a data input module (112);
preprocessing the EEG signals via a preprocessing module (114) to filter, normalize, and remove artifacts caused by eye blinks, muscle activity, or powerline interference;
segmenting the preprocessed EEG signals into time windows suitable for real-time analysis via a segmentation module (116);
extracting signal features from the segmented EEG signals using a feature extraction module (118), including spectral representations via at least one of a short-time Fourier transform (STFT) or a continuous wavelet transform (CWT);
analyzing temporal dependencies across the extracted features via a temporal analysis module (120) comprising a multi-head self-attention mechanism;
classifying the temporally processed features via a classification module (122) to detect seizure onset and classify multiple seizure states including preictal, ictal, postictal, and interictal phases;
generating interpretable reports of the classification results via an explainability module (124), including channel-wise importance scores, temporal heatmaps, and influential EEG regions, intervals, and frequency bands;
transmitting real-time alerts and notifications via a notification module (126) to the user device (102), including wearable vibration, audio notifications, and automatic emergency call initiation upon detection of an ictal state; and
exchanging data securely between the user device (102), the EEG acquisition unit (104) and the processing unit (110) via a communication unit (130).