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

Area of Science:

• Neurology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Epilepsy is a prevalent neurological disorder significantly affecting patient quality of life.
• Electroencephalography (EEG) is crucial for epilepsy diagnosis and management, but current detection methods face challenges in generalizability and computational cost.
• There is a critical need for rapid, accurate, and non-invasive seizure detection techniques.

Purpose of the Study:

• To propose a lightweight, end-to-end, attention-based deep learning network for automatic seizure detection using raw multichannel EEG signals.
• To develop an efficient model that overcomes the limitations of existing seizure detection methods.

Main Methods:

• The study designed a novel deep learning architecture incorporating residual depthwise separable convolutional (RDSC) blocks for spatial feature extraction.
• A channel-wise attention mechanism was employed to highlight salient EEG information.
• Temporal dependencies were captured using a gated recurrent unit (GRU) layer, followed by a classification head.

Main Results:

• The model was evaluated on the CHB-MIT EEG dataset using leave-one-patient-out cross-validation (LOPOCV).
• Achieved high performance metrics: 91.08% average accuracy, 91.92% precision, 90.36% sensitivity, 91.86% specificity, and 90.86% F1-score.
• Demonstrated the model's effectiveness for patient-independent epileptic seizure detection.

Conclusions:

• The proposed lightweight deep learning network demonstrates significant effectiveness for automatic epileptic seizure detection.
• The model's high accuracy and efficiency show potential for improving the clinical management of epilepsy.
• This approach offers a viable solution for non-invasive, patient-independent seizure detection using EEG data.