Epileptic Seizures Detection using Fusion of Artificial Neural Network with Hybrid Deep Learning

The neurological condition epilepsy, which is characterized by recurring seizures, poses challenges in accurate and timely detection for effective management. This proposed system presents and evaluates a deep learning approach aimed at building patient-specific classifiers to use scalp EEG monitoring, a non-invasive measure of the brain's electrical activity, to identify the beginning of epileptic episodes. This endeavor is particularly difficult because of the differing nature of the brain's electrical action, which comprises various classes with covering characteristics. The EEG data is preprocessed to extract key features before training the ANN-CNN-GRU fusion model on a large dataset of labeled recordings, including seizure and non-seizure occurrences. The proposed fusion model demonstrates superior performance compared to individual ANN, CNN, and GRU models, achieving high accuracy, sensitivity, and specificity in detecting epileptic seizures. Structuring the issue into a suitable deep learning framework and determining crucial characteristics for differentiating seizure from other forms of brain activity were important stages in developing a high-performance algorithm. After being trained on two or more known seizures and evaluated on 969 hours of uninterrupted EEG data from 24 patients, the algorithm detected of 173 test seizures with a middle location delay of 3 seconds and a middle false discovery rate of two false discoveries per 24-hour period. The physio net database provided the CHB-MIT dataset used in this study.