This is the authors' abstract. We don't add key points for this paper.
Epilepsy is a common chronic neurological disorder, and timely seizure detection is essential for effective clinical management. Although deep learning models such as bidirectional long short-term memory (BiLSTM) networks can capture temporal dependencies in EEG signals, their direct application to raw EEG remains challenging due to the nonstationary, noisy, and multiscale nature of brain activity. This study proposes a hybrid framework that integrates multivariate variational mode decomposition (MVMD) with BiLSTM networks for automated epileptic seizure detection. MVMD decomposes multichannel EEG into frequency-aligned intrinsic mode functions while preserving inter-channel dependencies, and BiLSTM learns temporal representations from each component. Evaluated on a balanced binary subset of the public Epileptic Seizure Recognition Dataset, the proposed model achieved 99.78% accuracy on both validation and test sets, compared with 98.70% for a baseline BiLSTM on raw EEG. Ablation analysis confirmed that these gains stemmed from MVMD preprocessing. The findings suggest that combining multivariate signal decomposition with bidirectional temporal learning can enhance EEG-based seizure classification, although further validation on independent, multi-center datasets and event-level assessment is needed before clinical use.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Hosseini et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: