Automatic Detection of Epileptic EEG Signals Based on Channel Attention and Bidirectional Temporal Modeling

Abstract

Epilepsy is a common chronic neurological disease. As an important tool for epilepsy diagnosis and disease assessment, electroencephalogram (EEG) can reflect the abnormal discharge activity of brain neurons. However, traditional EEG interpretation relies heavily on expert manual analysis, which has problems such as time-consuming, strong subjectivity and low efficiency. To improve the automation level of seizure detection, this paper proposes a pyramid-type one-dimensional convolutional neural network model (SE-P1D-LSTM) that integrates a channel attention mechanism and a Bidirectional Long Short-Term Memory (BiLSTM) network. The method first preprocesses the EEG signal through overlapping sliding window and Z-Score standardization, and combines Gaussian noise disturbance and random amplitude scaling for data augmentation; Then use the pyramidal one-dimensional convolutional structure to extract local temporal features, and adaptively strengthen the key channel information through SEBlock; Finally, a BiLSTM is introduced to model the long-range temporal dependencies in EEG signals to realize the automatic classification of normal and seizure EEG signals. The experiment was carried out based on the public epilepsy EEG dataset of the University of Bonn in Germany, and the performance of the model was evaluated using 10-fold cross-verification. The results show that the method in this paper has achieved an accuracy rate of 99.87%, a sensitivity of 99.87% and a specificity of 99.87% in the two-classification task, which is better than a variety of comparative models. The research results show that SE-P1D-LSTM can effectively extract the discriminative characteristics of epilepsy EEG signals, and has good classification performance and application potential in the automatic detection task of seizures.