Wavelet Packet Analysis for Automatic Seizure Detection and Latency Study in Scalp EEG

Epileptic seizure detection using EEG signal analysis has been an important area of research for the last two decades. In this paper, a new approach based on wavelet packet decomposition of EEG signal is introduced to detect epileptic activity in scalp EEG. The study was conducted on EEG data of 24 subjects suffering from epilepsy. Ictal as well as interictal EEG epochs were decomposed using wavelet packet decomposition and the best basis selection was performed. Eight features such as minimum, maximum, standard deviation, mean, inter quartile range, median absolute deviation, and energy and entropy were calculated for four levels of decomposition. Two features from each level were selected after ranking the features using t-test class separability criterion and the classification was done using linear classifier. Total 199 seizures were analyzed and 100 percent sensitivity was achieved with an average latency of 3.7 seconds. The results show improvement over the reported works on the same database.