Federated deep learning model for epilepsy seizure detection using electroencephalogram (EEG) signal

Area of Science:

• Neurology
• Artificial Intelligence
• Signal Processing

Background:

• Epilepsy is a chronic neurological disorder impacting quality of life.
• Traditional seizure detection methods struggle with data privacy and capturing complex EEG signal relationships.

Purpose of the Study:

• To propose a Federated Learning Enabled Unified Transformer model for improved seizure detection.
• To address limitations in data privacy, security, and feature extraction in Electroencephalography (EEG) analysis.

Main Methods:

• Utilized Federated Learning with Paillier Homomorphic Encryption for privacy-preserving collaborative training.
• Employed adaptive noise filtering, Independent Component Analysis, and Multi-Scale Wavelet Coefficient for signal preprocessing and feature extraction.
• Integrated a Hybrid Graph-Based Attention Framework with Edge-Enhanced Graph Convolutional Networks and Spectral Graph Attention for spatial and frequency feature analysis.
• Applied a Unified Transformer model for accurate seizure classification, capturing temporal and spatial EEG dependencies.

Main Results:

• The model achieved high performance metrics: 98.91% accuracy, 98.93% security, and 98.82% precision on three datasets.
• Demonstrated superior performance compared to existing seizure detection models.

Conclusions:

• The Federated Learning Enabled Unified Transformer model offers a significant advancement in epilepsy seizure detection.
• The integration of federated learning and deep learning enhances suitability for healthcare applications.