This is the authors' abstract. We don't add key points for this paper.
Epilepsy is a common neurological disorder that demands reliable seizure detection. Although deep learning has improved electroencephalography (EEG) analysis, existing methods often struggle to fully capture nonlinear interchannel dependencies, multiscale temporal dynamics, and computational efficiency. We propose TMI‐SD, a framework that integrates established techniques including mutual information neural estimation (MINE), its smoothed variant (SMILE), and factor graph inference, while introducing two novel components: tensor mutual information (TMI) and a multibranch lightweight CNN. TMI extends conventional mutual information to the tensor domain to preserve the spatiotemporal structure within each EEG channel, better capturing seizure‐related synchronization. The multibranch CNN uses four temporal windows (4, 8, 16, and 32 s) to extract multiscale features with only 37.50‐K parameters. Factor graph inference further refines predictions via first‐order Markov smoothing. Extensive experiments on the CHB‐MIT and TUSZ datasets under three evaluation strategies demonstrate that TMI‐SD consistently outperforms state‐of‐the‐art methods, achieving high AUC‐ROC, AUC‐PR, F1 score, precision, recall, and accuracy, while exhibiting strong cross‐dataset generalization and suitability for real‐time deployment.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: