Multi-Branch Mutual-Distillation Transformer for EEG-Based Seizure Subtype Classification.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Accurate seizure subtype classification using electroencephalogram (EEG) is crucial for epilepsy diagnosis and treatment.
• Deep learning models show promise for EEG analysis but typically require substantial labeled data, which is often scarce in clinical settings.
• Existing methods struggle with cross-subject generalization and limited data scenarios.

Purpose of the Study:

• To develop a novel deep learning approach for effective cross-subject EEG-based seizure subtype classification using limited labeled data.
• To introduce the Multi-Branch Mutual-Distillation (MBMD) Transformer architecture.
• To investigate the efficacy of knowledge distillation for enhancing EEG seizure classification.

Main Methods:

• Proposed the Multi-Branch Mutual-Distillation (MBMD) Transformer, modifying the Vision Transformer by incorporating multi-branch encoder blocks.
• Implemented a mutual-distillation strategy to transfer knowledge between raw EEG signals and their wavelet transformations across different frequency bands.
• Evaluated the model on two publicly available EEG datasets for cross-subject seizure subtype classification.

Main Results:

• The MBMD Transformer achieved superior performance compared to traditional machine learning and state-of-the-art deep learning methods.
• Demonstrated effective seizure subtype classification even with small amounts of labeled training data.
• The proposed mutual-distillation strategy enhanced the model's ability to generalize across subjects.

Conclusions:

• The MBMD Transformer offers a powerful and data-efficient solution for cross-subject EEG-based seizure subtype classification.
• Knowledge distillation is a viable and effective technique for improving deep learning models in the context of limited EEG data.
• This work represents a significant advancement in applying deep learning for precise epilepsy diagnostics.