Research on epilepsy detection and recognition based on the combination of time frequency transform and deep learning...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Epileptic electroencephalogram (EEG) signals present non-stationary characteristics, challenging accurate detection.
• Existing methods require improvement in performance and robustness for clinical applications.

Purpose of the Study:

• To enhance the detection performance of epileptic EEG signals.
• To compare the efficacy of continuous wavelet transform (CWT) and short-time Fourier transform (STFT) for feature extraction.
• To evaluate optimized neural network models including EEGNet, AlexNet, and Shallow ConvNet.

Main Methods:

• Feature extraction using CWT and STFT.
• Implementation and optimization of EEGNet, AlexNet, and Shallow ConvNet models.
• Integration of Focal Loss, dynamic data augmentation, and early stopping for enhanced robustness.
• Optimization of EEGNet with Squeeze-and-Excitation (SE) attention and depthwise separable convolution.
• Enhancement of Shallow ConvNet with layered convolution and average pooling.

Main Results:

• CWT-based feature extraction demonstrated superior performance over STFT.
• The CWT+Shallow ConvNet combination achieved optimal overall performance.
• The CWT+EEGNet combination showed excellent precision, closely following the top performer.
• Optimized neural network models significantly improved classification accuracy and robustness.

Conclusions:

• Combining precise time-frequency features from CWT with optimized neural networks offers a reliable approach for epileptic EEG signal detection.
• The study provides a robust technical foundation for clinical diagnosis of epilepsy using EEG data.
• Optimized deep learning models coupled with advanced signal processing techniques are crucial for advancing neurological disorder detection.