Detection Method of Epileptic Seizures Using a Neural Network Model Based on Multimodal Dual-Stream Networks

Area of Science:

• Neurology
• Artificial Intelligence
• Signal Processing

Background:

• Epilepsy diagnosis relies heavily on electroencephalogram (EEG) signal analysis.
• Raw EEG signals often lack sufficient recognizable features for accurate detection.
• Extracting multimodal features is crucial for improving diagnostic accuracy.

Purpose of the Study:

• To develop an advanced neural network model for improved epileptic seizure detection.
• To enhance the extraction of recognizable features from EEG signals.
• To address the challenge of recognizing multimodal features in EEG data.

Main Methods:

• A multimodal dual-stream neural network was proposed.
• Extracted differential, amplitude, and phase spectrum features to form 2D vectors.
• Employed a hybrid network combining 1D convolution, 2D convolution, and LSTM for spatial and temporal feature extraction.
• Integrated a channel attention module to focus on seizure-related features.

Main Results:

• The proposed model achieved high accuracy rates of 99.69% on the Bonn dataset and 97.5% on the New Delhi dataset.
• Experimental results verified the model's superiority in epileptic seizure detection.
• The hybrid approach effectively captured both temporal and spatial signal characteristics.

Conclusions:

• The multimodal dual-stream network offers a superior approach for epileptic seizure detection.
• Feature engineering combined with advanced neural networks significantly improves diagnostic performance.
• The model demonstrates strong potential for clinical application in epilepsy diagnosis.