Interpretable end to end Epileptic Seizure Detection via Linear and Nonlinear Filtering Networks.

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Epilepsy is a neurological disorder characterized by recurrent seizures.
• Electroencephalogram (EEG) analysis is crucial for identifying abnormal brain activity.
• Current EEG seizure detection methods struggle with complex signal dynamics and lack interpretability.

Purpose of the Study:

• To develop a novel, interpretable seizure detection framework for EEG signals.
• To address the limitations of single-modal feature analysis in existing methods.
• To enhance the clinical utility of deep learning models in epilepsy diagnostics.

Main Methods:

• Proposed a contrastive learning framework with linear and nonlinear filtering networks (CL-LNFNet).
• Employed recursive residual decomposition and dual-branch decoupling networks for feature extraction.
• Utilized adaptive filtering networks with feature selection gating and a multi-scale convolutional module.
• Implemented a hybrid learning strategy combining supervised and self-supervised contrastive learning.

Main Results:

• CL-LNFNet achieved over 95% accuracy in seizure detection on scalp and intracranial EEG datasets.
• Demonstrated superior performance compared to existing state-of-the-art methods.
• Showcased enhanced explainability by tracing the decision-making pathway from raw EEG to outcomes.

Conclusions:

• CL-LNFNet offers a robust and interpretable solution for EEG-based epilepsy seizure detection.
• The framework effectively disentangles complex linear and nonlinear EEG signal dynamics.
• The model bridges the gap between deep learning's 'black-box' nature and clinical transparency requirements.