Multi-perspective characterization of seizure prediction based on microstate analysis

Epilepsy is an irregular and recurrent cerebral dysfunction that significantly impacts the affected individual's social functionality and quality of life. This study aims to integrate cognitive dynamic attributes of the brain into seizure prediction, evaluating the effectiveness of various characterization perspectives for seizure prediction, while delving into the impact of varying fragment lengths on the performance of each characterization. We adopted microstate analysis to extract the dynamic properties of cognitive states, calculated the EEG-based and microstate-based features to characterize nonlinear attributes, and assessed the power values across different frequency bands to represent the spectral information of the EEG. Based on the aforementioned characteristics, the predictor achieved a sensitivity of 93.82% on the private FH-ZJU seizure dataset and 93.22% on the Siena Scalp EEG dataset. The study outperforms state-of-the-art works in terms of sensitivity metrics in seizure prediction, indicating that it is crucial to incorporate cognitive dynamic attributes of the brain in seizure prediction.