Key points are not available for this paper at this time.
Artificial Intelligence (AI) has advanced electroencephalography (EEG)-based epilepsy management, yet high-dimensional multi-channel EEG signals remain difficult to exploit effectively. Many existing approaches inadequately capture spatiotemporal characteristics and often fail to identify seizure-sensitive channels, with more emphasis placed on classification than prevention. To address these limitations, a multi-channel feature fusion framework is proposed. Temporal dynamics are modeled by a Temporal Convolutional Network (TCN), and spatial attention is learned by a Vision Transformer (ViT). Channel selection and attention-based reweighting are further introduced to optimize the fusion process. The proposed framework was evaluated on the CHB-MIT scalp EEG dataset and the Mayo Clinic intracranial EEG (iEEG) dataset. AUC scores of 95.6% and 90.8% were obtained, with false-positive rates of 2.7% and 8.2%, respectively. These results indicate that preictal EEG segments can be identified prior to seizure onset with improved robustness.
Gong et al. (Mon,) studied this question.