Deep learning-based EEG seizure classification using comparative time–frequency spectral representations

This is the authors' abstract. We don't add key points for this paper.

The automated prediction and classification of seizures using electroencephalography (EEG) remains a challenging task due to the inherent complexity and significant inter-subject variability observed in seizure patterns. To address these challenges, we explored the use of two-dimensional time-frequency representations of EEG signals to enhance the effectiveness of seizure type classification. In this study, we employed EEG data corresponding to background, focal, and generalized seizures, sourced from the Temple University Hospital (TUH v2.0.1) dataset. The EEG signals were first preprocessed and segmented. Each segment was then transformed into a two-dimensional time-frequency image using three techniques: Short-Time Fourier Transform (STFT), Mel-Frequency Cepstrum (MFC), and Continuous Wavelet Transform (CWT). We developed and evaluated five deep learning architectures for seizure classification: AlexNet, VGGNet, configurable Convolutional Neural Network (cCNN), configurable Recurrent Neural Network (cRNN), and configurable Long Short-Term Memory (cLSTM). All models were assessed using 10-fold cross-validation and evaluated across standard performance metrics such as accuracy, precision, recall, and F1-score. Experimental results showed that STFT-based representations with cCNN achieved an accuracy of 83.33%, while MFC spectrograms with cCNN yielded 76.31%. The highest classification performance, an accuracy of 85.17% was obtained using CWT scalograms with the cRNN model. In conclusion, the combination of CWT-based time-frequency features with the cRNN model proved to be most effective for the classification of seizures type. This study underscores the potential of deep learning and time-frequency analysis in developing reliable EEG-based diagnostic tools for improved clinical decision-making in epilepsy care.