Improving Epilepsy Diagnosis Outcomes in Diverse Populations Through an End-to-End Explainable EEG-Based AI System
Year: 2026
Type: conference-paper
Abstract: The issue of the diagnosis of epilepsy is an acute subject of contemporary medical care due to the growing number of cases of neurological diseases in the world, the ambiguity of the interpretation of electroencephalogram (EEG) signals, and the lack of qualified experts worldwide. The traditional forms of epilepsy detection typically require human examination of EEG signals through human neurologists or simple automated techniques and this might lead to late diagnosis, false classification, and failure to treat patients. To address these deficiencies, this paper suggests a fully automated, end-to-end, and transparent AI-based system to detect epilepsy in real-time based on EEG signals. The system to be proposed is the combination of powerful preprocessing methods, such as adaptive filtering, discrete wavelet transform, and independent component analysis to remove noise and artifacts, retain the diagnostic characteristics. The two complementary methods are used to analyze the processed signals: a hybrid machine learning classifier based on the combination of Random Forest and Gradient Boosting to achieve higher accuracy and a lightweight 1D convolutional neural network with multi-head attention mechanism to extract all the features autonomously. Bio-inspired particle swarm optimization is used to optimally select features, without performance loss. The system uses SHapley Additive Explanations to give it global and local interpretability, so that clinical practitioners can learn how each prediction was made. The proposed framework can be deployed on low-latency web and mobile platforms, thus it can be used in real-time clinical applications in a variety of populations and resource contexts. The architecture introduces an affordable, scalable, and open approach to diagnosis of precision epilepsy, which can be used in the context of sustainable healthcare practices and enhanced patient outcomes.
Institution: Dr. Babasaheb Ambedkar Technological University
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: gold
OpenAlex ID: W7224829635