This is the authors' abstract. We don't add key points for this paper.
Background Epileptic seizure prediction from scalp electroencephalography (EEG) may provide useful warning time. However, EEG noise and non-stationarity complicate the identification of preictal patterns, motivating effective signal preprocessing and feature learning for patient-specific prediction. Methods A patient-specific seizure prediction framework was evaluated using 19 retained cases from the CHB-MIT Scalp EEG Database. Models were trained and evaluated separately for each case using stratified five-fold cross-validation within that case. EEG signals were filtered, decomposed using empirical mode decomposition (EMD), and adaptively reconstructed using singular value decomposition (SVD). Short-time Fourier transform (STFT) spectrograms were encoded using an ImageNet-pretrained ResNeXt101 network, and the resulting feature vectors were classified using an attention-based bidirectional long short-term memory (BiLSTM) network. Performance was compared with conventional filtering and fixed-IMF reconstruction under the same within-case evaluation protocol. Results The proposed framework achieved mean accuracy, precision, recall (sensitivity), and area under the receiver operating characteristic curve (AUC) of 96.35%, 95.98%, 94.59%, and 95.93%, respectively. Ablation experiments showed that EMD-SVD reconstruction provided more stable training behavior and better overall predictive performance than the two alternative preprocessing strategies.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: