Analysis of Seizure Prediction Horizon on Scalp EEG Using MDWP Approach

Abstract

This study proposes a statistical approach to examine the pre-ictal period before the onset of seizures. The study employs the multidepth wavelet packet (MDWP) approach by excavating through the wavelet packet tree to the eighth level of decomposition. Numerous statistical measures were chosen to extract features over raw signal and the retained wavelet packets from the MDWP approach. This extensive process extracted more than twelve thousand features from every five-minute window taken two hours before to five minutes before the seizure onset. Ranking the features extracted from each five-minute window separately revealed the feature of mode computed on the 11th packet of the 4th level of decomposition, 6th packet of the 3rd level of decomposition and 3rd packet of the 2nd level of decomposition among the top three features during the pre-ictal duration. Moreover, the rank of these features shows a drooping nature around 70 minutes before seizure onset. This indicates the sign of prediction horizon to be close to 70 minutes before seizure onset for patient-1 of the CHB-MIT scalp EEG dataset. MATLAB installed on Workstation with 24 cores was used to process the enormous data involved in this study.