Dispersion Entropy for the automated detection of epileptic seizures

The electroencephalogram (EEG) is a powerful tool for evaluation of seizure activities in the human brain. This paper introduces Dispersion Entropy (DispEn) as a feature extraction method for the first time in the automated detection of epileptic seizures. A random forest (RF) classifier is utilized for the classification of normal, seizure-free, and seizure EEG segments. Additionally, we have also compared the performance of the proposed DispEn feature with existing entropy features such as Permutation Entropy, Fuzzy Entropy, Sample Entropy, and Approximate Entropy. The experimental results demonstrated that DispEn outperforms other traditional entropies and yielded the highest sensitivity, specificity, classification accuracy, precision and F1-score of 92.86%, 95.45%, 94%, 96.67%, and 95.05%, respectively for the classification of normal, seizure-free and seizure EEG segments using RF classifier.