Results of CHB-MIT single patient experiments.

Epilepsy is a neurological disorder characterized by recurrent seizures caused by abnormal brain activity, which can severely affects people’s normal lives. To improve the lives of these patients, it is necessary to develop accurate methods to predict seizures. Electroencephalography (EEG), as a non-invasive and real-time technique, is crucial for the early diagnosis of epileptic seizures by monitoring abnormal brain activity associated with seizures. Deep learning EEG-based detection methods have made significant progress, but still face challenges such as the underutilization of spatial relationships, inter-individual physiological variability, and sequence intricacies. To tackle these challenges, we introduce the Dual-Branch Deepwalk-Transformer Spatiotemporal Fusion Network (Deepwalk-TS), which effectively integrates spatiotemporal information from EEG signals to enable accurate and reliable epilepsy diagnosis. Specifically, the Spatio-branch introduces an adaptive multi-channel deepwalk-based graph framework for capturing intricate relationships within EEG channels. Furthermore, we develop a Guided-CNN Transformer branch to optimize the utilization of temporal sequence features. The novel dual-branch networks co-optimize features and achieve mutual gains through fusion strategies. The results of extensive experiments demonstrate that our method achieves the state-of-the-art performance in multiple datasets, such as achieving 99.54% accuracy, 99.07% sensitivity and 98.87% specificity. This shows that the Deepwalk-TS model achieved accurate epilepsy detection while analyzing the spatiotemporal relationship between EEG and seizures. The method further offers an optimized solution for addressing health issues related to seizure diagnosis.