Adaptive multi-scale phase-aware fusion network for EEG seizure recognition.

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Epilepsy detection relies on electroencephalogram (EEG) analysis.
• Current deep learning methods struggle with frequency adaptability, multi-scale feature integration, and phase alignment for accurate seizure detection.
• Manual feature extraction in traditional methods is time-consuming and less effective.

Purpose of the Study:

• To introduce the Adaptive Multi-Scale Phase-Aware Fusion Network (AMS-PAFN) for enhanced EEG-based epilepsy seizure detection.
• To overcome limitations in frequency adaptability, multi-scale feature integration, and phase alignment in existing deep learning models.
• To improve the accuracy and reliability of automated seizure detection systems.

Main Methods:

• The proposed AMS-PAFN integrates three novel modules: Dynamic Frequency Selection (DFS) for adaptive spectral filtering, Multi-Scale Feature Extraction (MCFE) for capturing diverse EEG patterns, and Multi-Scale Phase-Aware Fusion (MCPA) for temporal feature alignment.
• DFS utilizes Gumbel-SoftMax for optimizing seizure-related frequency bands.
• MCFE employs hierarchical downsampling and attention mechanisms, while MCPA uses phase-sensitive weighting for cross-scale synchronization.

Main Results:

• The AMS-PAFN achieved state-of-the-art performance on the CHB-MIT dataset, reaching 98.97% accuracy, 99.53% sensitivity, and 95.21% specificity.
• It demonstrated a significant improvement over STFTormer, with a 1.58% increase in accuracy and a 2.66% increase in specificity.
• Ablation studies confirmed the individual contributions of DFS (improving specificity by 6.87%) and MCPA (enhancing cross-scale synchronization by 5.54%).

Conclusions:

• The AMS-PAFN shows significant potential for clinical application in epilepsy seizure recognition.
• Its adaptability to spectral variability and spatiotemporal dynamics makes it suitable for real-time epilepsy monitoring and alert systems.
• The network offers a robust solution for improving diagnostic accuracy and patient care in epilepsy management.