Automatic epileptic seizure detection method based on spatio-temporal feature fusion.

This paper proposes a spatiotemporal feature fusion method for automatic epileptic seizure detection, integrating Common Spatial Pattern (CSP) and Least Squares Support Vector Machine (LSSVM). First, it reconstructs electroencephalogram (EEG) noise using Ensemble Empirical Mode Decomposition (EEMD), then decomposes the original EEG signals using improved EEMD (IEEMD). Next, features are extracted from temporal and spatial dimensions to form a feature set. The classification process adopts a novel dual-classification mode based on LSSVM ultimately achieving high-performance automatic recognition of normal, seizure, and interictal EEG signals. Validated on Bonn and CHB-MIT EEG datasets, the IEEMD algorithm achieves 99.57% ± 0.02 accuracy on Bonn and 96.43% overall accuracy on CHB-MIT. Results show IEEMD and spatiotemporal features effectively address low interictal-ictal recognition rates in existing studies, offering a reliable means for epileptic seizure prediction.