Description:The present invention relates to the domain of biomedical signal analysis and real-time medical event detection, particularly focused on seizure detection from electroencephalogram (EEG) data. It specifically pertains to an advanced deep learning framework that integrates spatiotemporal modeling, causal attention, Bayesian filtering, and context-aware meta-learning to provide accurate, low-latency, and patient-adaptive seizure detection for clinical and wearable neuro-monitoring systems
BACKGROUND OF THE INVENTION
The following description of related art is intended to provide background information pertaining to the field of the disclosure. This section may include certain aspects of the art that may be related to various features of the present disclosure. However, it should be appreciated that this section be used only to enhance the understanding of the reader with respect to the present disclosure, and not as admissions of prior art.

In the field of clinical neurology, seizure detection plays a pivotal role in diagnosing and managing epilepsy and related neurological disorders. Continuous EEG monitoring is a widely accepted diagnostic tool; however, its interpretation remains time-consuming and prone to human error, especially during long-term monitoring. Real-time seizure detection systems are thus essential for responsive neurostimulation and timely medical interventions.

Existing automatic detection systems often employ machine learning models such as Support Vector Machines, CNNs, and LSTMs. While these approaches have shown promise in controlled conditions, their performance significantly degrades in real-world noisy environments due to the inability to model the complex temporal-spatial patterns present in EEG signals.

Moreover, current systems lack adaptability to patient-specific characteristics, resulting in high false alarm rates and poor generalization across diverse patient profiles. This is particularly problematic when deploying such systems in ambulatory or real-time clinical settings where inter-patient variability is high.

Another critical limitation is that most detection algorithms are trained offline and are not equipped for online learning or recursive adaptation. This static nature prevents them from incorporating new information or contextual changes such as medication effects, circadian rhythms, or sleep cycles.

Additionally, while attention mechanisms have been recently explored, they often ignore causality and frequency-specific features, which are essential for neurological event modeling. The lack of causal inference leads to unreliable decision-making, especially in edge-based or wearable devices where resource constraints are critical.

There is, therefore, a significant need for a comprehensive, lightweight, and contextually adaptive seizure detection framework that overcomes these limitations through integrated modeling, real-time inference, and personalized adaptability.

OBJECTIVE OF THE INVENTION

Some of the objects of the present disclosure, which at least one embodiment herein satisfies are listed herein below.

The primary objective of the present invention is to develop a comprehensive real-time seizure detection framework that is capable of accurately detecting seizure events across a wide range of patients while adapting to individual variability and contextual states.

Another objective is to construct a Temporo-Spatial Transformer Graph Network (TSTG-Net) to simultaneously model spatial dependencies among EEG channels and the temporal evolution of seizure patterns using attention-based and graph-based mechanisms.

A further objective is to enhance the discriminative power of latent features by integrating a Contrastive Adversarial Embedding Module (CAEM) that uses patient-specific interictal data to strengthen the separation between seizure and non-seizure classes.

Yet another objective is to enable real-time, low-latency inference through an Online Recursive Neuro-Bayesian Filter (ORNBF) that continually updates seizure likelihood in a probabilistic manner using lightweight neural approximators.

An additional objective is to incorporate a Multi-Resolution Causal Attention Network (MCAN) that captures features across multiple EEG frequency bands while enforcing temporal causality to reduce decision errors and improve robustness.

It is also an objective to adapt decision-making thresholds in real time using Task-Modulated Learning with Patient-Specific Context (TML-PSC) that leverages external metadata such as medication status, alertness level, and circadian rhythms for personalized adaptation.

Overall, the invention seeks to provide a clinically deployable solution that offers high accuracy, low false alarm rates, and ultra-low latency, making it suitable for continuous patient monitoring, both in hospital environments and portable applications.

SUMMARY OF THE INVENTION
This section is provided to introduce certain objects and aspects of the present disclosure in a simplified form that are further described below in the detailed description. This summary is not intended to identify the key features or the scope of the claimed subject matter.

The present invention provides a novel hybrid AI-based seizure detection architecture named FUSED-SEIZURE, which integrates transformer-based spatiotemporal modeling, contrastive adversarial embeddings, Bayesian recursive filtering, multi-resolution causal attention, and online meta-learning. Each module is designed to address a specific challenge in EEG-based seizure detection, such as inter-patient variability, class overlap, signal noise, or contextual sensitivity.

By combining these modules in a lightweight and real-time computational pipeline, the proposed framework achieves superior performance on benchmark datasets with high sensitivity, specificity, and low false alarm rates. The invention’s modular and adaptive structure enables effective deployment on wearable devices and edge processors, making it a breakthrough in personalized, real-time seizure monitoring systems.

BRIEF DESCRIPTION OF DRAWINGS
The accompanying drawings, which are incorporated herein, and constitute a part of this invention, illustrate exemplary embodiments of the disclosed methods and systems in which like reference numerals refer to the same parts throughout the different drawings. Components in the drawings are not necessarily to scale, emphasis instead being placed upon clearly illustrating the principles of the present invention. Some drawings may indicate the components using block diagrams and may not represent the internal circuitry of each component. It will be appreciated by those skilled in the art that invention of such drawings includes the invention of electrical components, electronic components or circuitry commonly used to implement such components.

FIG. 1 illustrates an exemplary real-time seizure detection system, in accordance with an embodiment of the present disclosure.

DETAILED DESCRIPTION OF THE INVENTION

In the following description, for the purposes of explanation, various specific details are set forth in order to provide a thorough understanding of embodiments of the present disclosure. It will be apparent, however, that embodiments of the present disclosure may be practiced without these specific details. Several features described hereafter can each be used independently of one another or with any combination of other features. An individual feature may not address all of the problems discussed above or might address only some of the problems discussed above. Some of the problems discussed above might not be fully addressed by any of the features described herein.

The ensuing description provides exemplary embodiments only and is not intended to limit the scope, applicability, or configuration of the disclosure. Rather, the ensuing description of the exemplary embodiments will provide those skilled in the art with an enabling description for implementing an exemplary embodiment. It should be understood that various changes may be made in the function and arrangement of elements without departing from the spirit and scope of the disclosure as set forth.

Specific details are given in the following description to provide a thorough understanding of the embodiments. However, it will be understood by one of ordinary skill in the art that the embodiments may be practiced without these specific details. For example, circuits, systems, networks, processes, and other components may be shown as components in block diagram form in order not to obscure the embodiments in unnecessary detail. In other instances, well-known circuits, processes, algorithms, structures, and techniques may be shown without unnecessary detail to avoid obscuring the embodiments.

Also, it is noted that individual embodiments may be described as a process that is depicted as a flowchart, a flow diagram, a data flow diagram, a structure diagram, or a block diagram. Although a flowchart may describe the operations as a sequential process, many of the operations can be performed in parallel or concurrently. In addition, the order of the operations may be re-arranged. A process is terminated when its operations are completed but could have additional steps not included in a figure. A process may correspond to a method, a function, a procedure, a subroutine, a subprogram, etc. When a process corresponds to a function, its termination can correspond to a return of the function to the calling function or the main function.

The word “exemplary” and/or “demonstrative” is used herein to mean serving as an example, instance, or illustration. For the avoidance of doubt, the subject matter disclosed herein is not limited by such examples. In addition, any aspect or design described herein as “exemplary” and/or “demonstrative” is not necessarily to be construed as preferred or advantageous over other aspects or designs, nor is it meant to preclude equivalent exemplary structures and techniques known to those of ordinary skill in the art. Furthermore, to the extent that the terms “includes,” “has,” “contains,” and other similar words are used in either the detailed description or the claims, such terms are intended to be inclusive in a manner similar to the term “comprising” as an open transition word without precluding any additional or other elements.

Reference throughout this specification to “one embodiment” or “an embodiment” or “an instance” or “one instance” means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present disclosure. Thus, the appearances of the phrases “in one embodiment” or “in an embodiment” in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. As used herein, the singular forms “a”, “an”, and “the” are intended to include the plural forms as well, unless the context indicates otherwise. It will be further understood that the terms “comprises” and/or “comprising,” when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. As used herein, the term “and/or” includes any and all combinations of one or more of the associated listed items.

The proposed invention, referred to as FUSED-SEIZURE, is a multi-component framework designed for real-time seizure detection from electroencephalogram (EEG) signals. It is architected to overcome the deficiencies in conventional seizure detection methods by integrating transformer-based graph modeling, causal attention mechanisms, recursive Bayesian filtering, and context-aware meta-learning adaptation. The system is optimized for low-latency, high-accuracy detection under dynamic clinical conditions.

The first key component is the Temporo-Spatial Transformer Graph Network (TSTG-Net), which converts EEG signals into structured temporal-spatial representations. EEG channels are modeled as graph nodes, and temporal sequences are processed via transformer encoders that capture long-range dependencies. The graph-based architecture encodes spatial relationships, while the attention layers extract temporal features. Together, they allow the system to learn complex seizure dynamics across time and space.

Following this, the Contrastive Adversarial Embedding Module (CAEM) is introduced to enhance inter-class separability in the feature space. By generating adversarial negative samples from interictal patient-specific EEG segments, this module constructs a contrastive loss that forces the embedding space to clearly distinguish seizure (ictal) and non-seizure (interictal) patterns. The learned embeddings improve classification confidence and reduce false positives.

To enable real-time processing and adaptation, the system employs an Online Recursive Neuro-Bayesian Filter (ORNBF). This module uses lightweight neural network approximators to recursively compute posterior seizure probabilities from streaming EEG inputs. The ORNBF continuously refines its internal state as new EEG evidence arrives, allowing for low-latency and robust seizure prediction under noisy conditions.

To enhance temporal reliability and frequency-specific responsiveness, a Multi-Resolution Causal Attention Network (MCAN) processes the EEG signals in parallel across different frequency bands (delta, theta, alpha, beta, gamma). Each frequency-specific stream employs causal attention mechanisms that restrict computation to past and current time steps only. The aggregated attention outputs reinforce decision accuracy and prevent future signal leakage, which is critical in real-time systems.

Finally, Task-Modulated Learning with Patient-Specific Context (TML-PSC) is incorporated to dynamically adjust decision thresholds based on contextual metadata. This module considers non-EEG inputs like medication level, circadian rhythm, and alertness to modulate the classifier’s threshold using a gradient-based meta-learning strategy. It ensures the system remains sensitive to patient-specific physiological and behavioral patterns, enabling personalized and adaptive seizure detection.

In one embodiment, the FUSED-SEIZURE architecture is implemented and evaluated using the CHB-MIT Scalp EEG dataset. Raw EEG signals from 22 pediatric subjects are preprocessed using bandpass filtering and normalized across sessions. TSTG-Net is trained using a graph topology constructed based on inter-electrode distances and functional connectivity. CAEM is applied by selecting interictal samples 30 minutes prior to seizures as negative anchors.

The ORNBF module is initialized with prior seizure likelihoods and updated recursively during inference using incoming EEG windows (2-second duration). MCAN processes inputs at five frequency resolutions, each with independent causal attention streams, which are finally fused using a soft-attention aggregation scheme. TML-PSC uses metadata such as time-of-day and sleep state (obtained from manual annotations) to adjust decision thresholds on-the-fly.

This embodiment yields a detection accuracy of 93.1%, sensitivity of 91.8%, specificity of 95.4%, and a false alarm rate of 0.06 per minute. The average latency between seizure onset and detection is measured at 0.41 seconds, confirming the system’s suitability for clinical deployment.

In another embodiment, the invention is used to monitor a patient undergoing antiepileptic drug (AED) titration, where medication levels vary significantly over time. EEG recordings and medication logs are collected simultaneously from a wearable headband EEG system. The FUSED-SEIZURE model is pre-trained on baseline data and incrementally updated using TML-PSC during monitoring.

TML-PSC receives contextual inputs such as medication dosage (mg/day), blood concentration levels (via API from hospital database), and circadian time segment. Based on this information, the meta-learning module adjusts the seizure probability threshold by adapting the gradient in the final softmax layer. This context-aware mechanism helps the model maintain high specificity even during periods of pharmacological fluctuation.

The embodiment demonstrates that during daytime monitoring when AED concentration is lower, the model increases sensitivity to detect low-intensity seizures. At night, when drug levels peak, the system becomes more conservative, thus reducing false alarms. This embodiment validates the importance of adaptive thresholding in real-world applications.

In a third embodiment, the FUSED-SEIZURE system is integrated into a wearable EEG monitoring device designed for ambulatory seizure detection. The TSTG-Net and ORNBF components are quantized and deployed using TensorFlow Lite on an ARM Cortex-M7 microcontroller. The CAEM module is pre-trained and implemented as a lookup table, while MCAN is optimized using grouped 1D causal convolutions.

The wearable device streams EEG data in real time and performs inference every 0.5 seconds. A mobile app displays predictions and provides alert notifications when a seizure is detected. Contextual data for TML-PSC is collected via a companion smartwatch and user input, including sleep status, activity level, and medication reminders.

Performance testing in a home-based trial with five epilepsy patients reveals the system can detect seizures with over 90% accuracy and under 0.5-second latency. The edge-deployed model consumes less than 150mW, making it highly energy-efficient and suitable for long-term continuous use.

While considerable emphasis has been placed herein on the preferred embodiments, it will be appreciated that many embodiments can be made and that many changes can be made in the preferred embodiments without departing from the principles of the invention. These and other changes in the preferred embodiments of the invention will be apparent to those skilled in the art from the disclosure herein, whereby it is to be distinctly understood that the foregoing descriptive matter to be implemented merely as illustrative of the invention and not as limitation.
, Claims:1. A real-time seizure detection system comprising:
o a Temporo-Spatial Transformer Graph Network (TSTG-Net) configured to model spatial and temporal features of EEG signals;
o a Contrastive Adversarial Embedding Module (CAEM) configured to generate and utilize patient-specific interictal negative samples to improve class separability;
o an Online Recursive Neuro-Bayesian Filter (ORNBF) configured to iteratively update seizure probabilities using lightweight neural approximators;
o a Multi-Resolution Causal Attention Network (MCAN) configured to aggregate causal attention outputs across EEG frequency bands;
o and a Task-Modulated Learning with Patient-Specific Context (TML-PSC) module configured to adapt detection thresholds based on contextual metadata through an online meta-learning framework;
o wherein the system performs low-latency seizure detection with reduced false alarm rate in a real-time monitoring environment.
2. The system of claim 1, wherein said TSTG-Net includes graph attention layers to capture inter-channel spatial dependencies in EEG data.
3. The system of claim 1, wherein said CAEM employs contrastive loss functions with adversarial interictal samples for robust embedding generation.
4. The system of claim 1, wherein said ORNBF performs recursive Bayesian updates using posterior distributions generated from real-time EEG segments.
5. The system of claim 1, wherein said MCAN processes EEG inputs across delta, theta, alpha, beta, and gamma frequency bands using causal convolution.
6. The system of claim 1, wherein said TML-PSC receives contextual data including time of day, medication levels, and patient state.
7. The system of claim 1, wherein the modules are optimized for deployment on edge devices including wearable EEG monitors.
8. The system of claim 1, wherein the system latency from EEG input to seizure alert is less than 0.5 seconds.
9. The system of claim 1, wherein seizure detection performance metrics exceed 93% accuracy and 0.07/min false alarm rate on benchmark datasets.