A lightweight depthwise separable convolution and channel attention based GRU network for multichannel EEG seizure detection

Epilepsy is the fourth most common neurological disorder, and seizures significantly impact quality of life of affected individuals. Electroencephalography (EEG) based seizure detection has emerged as a critical research area to improve the diagnosis and management of epilepsy. However, existing methods often suffer from poor generalisability and high computational complexity, emphasizing the need for rapid, accurate, and non-invasive detection frameworks. This study proposes a lightweight end to end attention based deep learning network for automatic seizure detection from raw multichannel EEG signals. The architecture employs multiple residual depthwise separable convolutional (RDSC) blocks for efficient spatial feature extraction, followed by a channel wise attention mechanism to emphasize salient information. Temporal dependencies are modelled using a gated recurrent unit (GRU) layer integrated with a classification head. The proposed model was evaluated on the CHB-MIT dataset using leave-one-patient-out cross-validation (LOPOCV) method, achieving an average accuracy of 91.08%, precision of 91.92%, sensitivity of 90.36%, specificity of 91.86%, and F1-score of 90.86%. These results demonstrate the effectiveness of the model and potential for patient independent epileptic seizure detection.