Epileptic Activity Detection using Mean Value, RMS, Sample Entropy, and Permutation Entropy Methods

Abstract

In this study, linear and non-linear signal analysis methods are implemented for epilepsy seizure detection using CHB-MIT EEG data taken from Boston children's hospital. In linear signal analysis, EEG signals are considered linear, although they are not linear. In linear signal analysis methods, root mean square (RMS) and mean of the EEG signals are analyzed. It is detected that the RMS value increased and the mean value moved away from zero in the positive and negative directions during the seizure period. Seizure periods in EEG signals are determined with RMS and mean values with 75 % and 58.4 % accuracy, respectively. Since EEG signals are not linear, the linear analysis is assumed insufficient and so entropy is preferred to linear signal analysis methods. Sample entropy (SmpE) and permutation entropy (PE) are preferred among entropy types. While an increase is observed in the sample entropy values at the beginning of the seizure, a decrease is observed in the permutation entropy values at the same time. When the entropy methods are examined separately, the onset of a seizure is determined with an accuracy of 66.6 % for both methods. However, when the entropy methods are examined together with the increase in the sample entropy value or the decrease in the permutation entropy, the accuracy rate increases to 79.2 % The resultant accuracy rates show that when one entropy method fails to catch the onset of a seizure the other can.