The Seizure Detection Based on a Novel Electroencephalogram Segmentation

Due to the non-stationary nature of electroencephalogram (EEG) signals, it is imperative to partition the EEG into quasi-stationary intervals with similar statistical characteristics. This study introduces an innovative method leveraging the relative stability of synchronous brain electrical activity for the automatic segmentation of EEG signals, eliminating the need for human intervention. The experiments illustrate that EEG signals segmented using this method demonstrate a certain level of stability across different frequency ranges and frequency power distribution. This consistency ensures the EEG signals remain quasi-stationary to a significant degree. For validation, this method was applied to seizure detection. The evaluation, utilizing EEG data from the department of pediatrics of Chinese PLA General Hospital, reveals that our method outperforms existing techniques. Notably, the approach achieved a seizure detection accuracy of 92.26%, specificity of 91.43%, and sensitivity of 92.93%.