Efficient Seizure Detection by Complementary Integration of Convolutional Neural Network and Vision Transformer.

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Epilepsy is a common neurological disorder requiring accurate, real-time seizure detection for diagnosis and treatment.
• Existing automatic seizure detection systems struggle to analyze both local and long-range features in electroencephalogram (EEG) signals.
• Limitations in current methods hinder precise diagnosis and timely intervention for epilepsy patients.

Purpose of the Study:

• To develop an advanced, end-to-end seizure detection framework for epilepsy.
• To enhance the accuracy and real-time capabilities of automatic seizure detection systems.
• To address the challenge of capturing both local and long-range dependencies in EEG signals.

Main Methods:

• A novel CNN-ViT framework integrating Convolutional Neural Network (CNN) and Vision Transformer (ViT) was proposed.
• The CNN component captures local EEG features, while ViT analyzes long-range dependencies.
• Raw EEG signals underwent filtering, segmentation, and processing via CNN-ViT, incorporating global max-pooling and post-processing for artifact reduction.

Main Results:

• The CNN-ViT model achieved high sensitivity (99.34% segment-based, 99.70% event-based) on the CHB-MIT EEG dataset.
• On the SH-SDU dataset, the method demonstrated 99.86% segment-based sensitivity and 100% event-based sensitivity.
• Processing 1 hour of EEG data took only 3.07 seconds, indicating efficient real-time performance.

Conclusions:

• The CNN-ViT framework offers a significant advancement in automatic seizure detection for epilepsy.
• The model's ability to capture diverse EEG signal features ensures high accuracy and efficiency.
• This method shows strong potential for clinical real-time seizure detection applications.