A deep hybrid CSAE-GRU framework with two-stage balancing for automatic epileptic seizure detection using EEG-derived...

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Epilepsy diagnosis relies on identifying abnormal brain activity, often challenging due to signal complexity.
• Electroencephalogram (EEG) signals are crucial for epilepsy monitoring but require sophisticated analysis for accurate seizure detection.

Purpose of the Study:

• To develop a high-performance framework for automated epileptic seizure detection using EEG signals.
• To integrate advanced feature extraction and a hybrid deep learning model for improved diagnostic accuracy.

Main Methods:

• EEG signals were preprocessed using bandpass filtering and Discrete Wavelet Transform (DWT).
• A two-stage class balancing strategy involving cluster centroid-based under-sampling and Borderline Synthetic Minority Oversampling Technique (BLSMOTE) was employed.
• A hybrid Convolutional Sparse Autoencoder (CSAE) and Gated Recurrent Unit (GRU) model was utilized, incorporating transfer learning.

Main Results:

• The framework achieved high performance on the Bonn and CHB-MIT EEG datasets.
• Key performance metrics included accuracy (up to 99.49%), sensitivity (up to 99.21%), and specificity (up to 99.77%).
• The results demonstrate the effectiveness of the integrated approach for seizure detection.

Conclusions:

• The proposed framework combines novel data balancing and a hybrid deep learning model for robust epilepsy detection.
• This approach shows significant potential for reliable and clinically applicable automated epilepsy diagnosis.