Permutation Fuzzy Entropy—An Index for the Analysis of Epileptic Electroencephalogram

Authors:
  • YNYan Niu
    Yan Niu
  • RCRui Cao
    Rui Cao
  • HWHuiyun Wang
    Huiyun Wang
  • CLConggai Li
    Conggai Li
  • MZMengni Zhou
    Mengni Zhou
  • YGYuxiang Guo
    Yuxiang Guo
  • BWBin Wang
    Bin Wang
  • PYPengfei Yan
    Pengfei Yan
  • JXJie Xiang
    Jie Xiang

Abstract
Entropy is a nonlinear index reflecting the degree of chaos within a system. As an improvement of sample entropy (SampEn), fuzzy entropy (FuzzyEn) has achieved relatively ideal classification results in the research of epilepsy detection over the sample entropy. However, because of its sensitivity to noise, FuzzyEn's antinoise ability is relatively weak, while the EEG signals are quite susceptible to noise during collection. Therefore, this work proposed permutation fuzzy entropy (PFEN), which improved the anti-noise ability of FuzzyEn by using permutation-based signifying time series. The performances of anti-noise and epilepsy detection for the proposed PFEN were analyzed. Compared with FuzzyEn and permutation entropy (PE), PFEN had the best antinoise performance based on simulated EEG data generated from the neural mass model. In addition, based on the EEG data from epileptic supplied by the CHB_MIT dataset and the Freiburg dataset, PFEN (CHB_MIT AUC (Area under Curve) = 0.988, Freiburg AUC = 0.958) had better classification performance than that of FuzzyEn (CHB_MIT AUC = 0.976, Freiburg AUC = 0.938) and PE (CHB_MIT AUC = 0.972, Freiburg AUC = 0.911). The experimental results demonstrate that PFEN is a novel index suitable for analyzing epileptic EEG.