Patient-independent EEG seizure detection using channel attention and dual-branch time–frequency features

Abstract

We developed a patient-independent EEG-based seizure detection model that generalizes across subjects despite inter-subject variability in spatial seizure expression, electrode characteristics, and artifacts. Scalp EEG recordings from the CHB–MIT (18 channels) and the TUSZ v2.0.3 (20 channels) datasets were evaluated under strict subject-disjoint protocols, employing leave-one-subject-out (LOSO) cross-validation for CHB–MIT and the official subject-disjoint train/dev/eval split for TUSZ. A channel-wise attention front end adaptively reweights EEG channels prior to feature extraction. The reweighted signals are processed through dual-branch time–frequency representations based on empirical mode decomposition (EMD) and discrete wavelet transform (DWT). Lightweight attention-augmented convolutional neural networks model each branch, and their embeddings are fused for seizure/non-seizure classification. Episode-level decisions are obtained by averaging window-level probabilities from short, overlapping segments. The proposed model demonstrated robust cross-subject generalization under strict subject-disjoint evaluation on both datasets. Ablation studies showed that the EMD and DWT branches provide complementary information and that the complete configuration, combining front-end channel attention with intra-CNN attention, achieved the highest accuracy, AUROC, F1-score, and recall on both datasets. These findings indicate that complementary EMD–DWT representations combined with channel reweighting enhance generalization for patient-independent seizure detection. The proposed framework may support scalable and robust seizure detection across unseen patients and heterogeneous EEG acquisition settings.