Neurological Disorder Diagnosis through Deep Residual Network-based EEG Signal Analysis

Quantifying anomalies in brain signals can reveal various brain conditions and pathologies. Most recent studies on neurological disorders diagnosis such as epilepsy and autism spectrum disorder (ASD) based on electroencephalogram (EEG) rely on custom feature extraction techniques. Traditional methods of feature extraction approaches are time-consuming and provide limited accuracy. So, to balance accuracy and efficiency, more ability is required in the choice of such significant features. This study introduces a feature extraction model based on a deep residual network (ResNet) capable of automatically extracting representative features from EEG signal to address this issue. This proposed method consists of three steps: signal preprocessing using a static filtering method, hidden pattern extraction from EEG signals using the ResNet model, and classification using a SoftMax layer. EEG data sets from the UBonn University database and King Abdulaziz University (KAU) are used in this study. The ResNet model’s results are evaluated using accuracy, sensitivity, and specificity. In this study, the proposed diagnostic system achieves a classification for 3 classes accuracy of 100% for Epilepsy in offline diagnosis and an accuracy of 95.5% for Autism in a three-class classification.