Epileptic EEG information entropy based on different entropy estimation methods

Abstract

Abstract We use the information entropy based on linear model, K-nearest neighbour estimation and kernel estimation to study the brain, a complex nonlinear dynamic system, and distinguish the nonlinear dynamic complexity of epileptic and normal EEG signals in Bonn database. The entropy estimation method of linear model and K-nearest neighbour estimation can only distinguish the information entropy of epileptic period from that of other two cases, but it can’t distinguish the information entropy of epileptic interval and normal EEG signals. However, kernel estimation can distinguish the information entropy of EEG signals in three states well, and the threshold range is [0.1,3]. With the increase of threshold, the discrimination effect is gradually significant until stable, and the discrimination effect is obvious when the threshold is 0.5. The result of analysis indicated that EEG information entropy was the highest in the epileptic seizure period, followed by the epileptic seizure interval, and the lowest in normal human brain.