Quantitying Nonlinear Dynamic Complexity of Epileptic EEG by Conditional Entropy Based on Different Entropy Measures

摘要

Brain is a typical nonlinear complex system, influenced by different factors. We employ CondEn (conditional entropy) based on linear, kernel and k-nearest-neighbor estimators to quantify nonlinear dynamic complex of epileptic brain electric activities from Bonn database. The three entropy measures all have promising results, among which kernel estimator shows optimal performance with feature of insensitivity to tolerance. CondEn of seizure EEG is the highest 3.2bit approximately while the seizure-free brain activities have lowest 1.5bit, and the entropy value of EEGs of the normal subjects is 1.9bit. CondEn is an effective parameter to measure nonlinear dynamic complexity of EEG, and EEG during seizure have the highest entropy, the normal EEG signal followed, and the seizure-free state was the lowest.