Multiscale Permutation Entropy Analysis of EEG-Based Seizure Classification: A Machine Learning Approach

Epilepsy is an abnormal brain condition characterized by the excessive synchronous firing of cortical neurons. Electroencephalogram (EEG) is a diagnostic tool to identify the underlying brain dynamics responsible for seizures. To detect seizures, neurologists must continuously evaluate the EEG recordings from the background activity during preoperative stages, which is tedious and susceptible to errors. Hence, automatic identification of seizure and seizure-free EEG signals is crucial for effective treatment and diagnosis. Entropy-based techniques are effective for detecting seizures due to the non-linear, non-stationary, and chaotic nature of EEG signals. The present study explores the efficacy of Multiscale Permutation Entropy (MSPE) for quantifying and classifying EEG signals by utilizing two public datasets. MSPE is assessed in each EEG epoch from healthy subjects and epileptic subjects; thereby, fifteen relevant features are utilized for training the two machine learning algorithms, namely the K-nearest neighbor (KNN) and Support Vector Machine (SVM). Results show that MSPE features combined with the KNN classifier can effectively discriminate seizure EEG signals from the healthy subjects and seizure-free signals with 100% and 98% accuracy, respectively. Therefore, the MSPE method is a potential feature extraction technique for detecting seizures from the EEG signals, which can assist neurologists in correctly diagnosing epilepsy.