Optimising EEG Analysis for Enhanced Epilepsy Detection using DRCNN

Accurate diagnosis of epilepsy from electroencephalography (EEG) signals is a pressing concern in neurology, prompting the adoption of advanced deep learning techniques for automated seizure detection. However, the lack of interpretability in these models poses a significant challenge. To tackle this issue, recent research has integrated graph representations of EEG signals with graph neural network (GNN) architectures. However, existing methods encounter obstacles, including the high time and space complexity of graph construction and the inability of GNNs to effectively utilize information from both time and frequency domain graphs. In response, this study introduces a novel approach leveraging the Weighted Neighbourhood Field Graph (WNFG) representation, enhancing both efficiency and information retention compared to dense counterparts. Furthermore, we propose a comprehensive pipeline encompassing data preprocessing steps such as filtering and Independent Component Analysis (ICA), followed by the WNFG-based graph representation. The core of our method lies in the integration of the Deep Recurrent Convolutional Neural Network (DRCNN) architecture, which facilitates feature extraction and classification of seizures.