Automatic seizure detection in ECoG by DB4 wavelets and windowed variance: A comparison

Abstract

Automatic detection of seizure in a continuous multichannel recording of EEG and ECoG has remained a challenging task even after more than three decades of research. Here we report that differential operator significantly accentuates the seizure part of depth electrode recordings (ECoG) relative to the non-seizure part. The success rate of detection by windowed variance method goes up considerably if the signal is treated with differential operator beforehand. In order to keep the false positive rate at the minimum a number of statistical checks have been introduced. Altogether they take only linear time and therefore well suited for real time applications. Detection on the same data with the same windowed variance method has also been performed using DB4 wavelet filtering instead of the differential operator (DB4 has been chosen from among Haar, DB1, DB2, DB3, DB4, DB5 and Morlet based on comparative study). It showed almost equal success but with higher time complexity.