Seizure Detection of Epileptic EEG Based on Multiple Phase-Amplitude Coupling Methods

Abstract

For epileptic electroencephalography (EEG) analysis, features extraction is crucial in seizure detection. In this paper, five methods for phase-amplitude coupling (PAC) were employed to analyze epileptic EEG to verify that PAC can be used as a biomarker to detect seizures. Specifically, five algorithms of evaluating PAC were used to compute PAC of seizure activity and seizure-free intervals at nine frequency band combinations. Then PAC of the EEG in a public dataset computed was classified by support vector machine (SVM), where the classification performance was assessed by calculating mean area under curve (AUC) based on receiver operating characteristic (ROC) with k-fold cross-validation (CV). Moreover, phase-amplitude comodulogram was applied to the same dataset to confirm intuitively classification accuracy. Results showed that the classification accuracy at band combination $\theta -\gamma$ was up to 0.96 and 0.99 for identifying seizure-free and seizure intervals both within epileptogenic zone and for classifying seizure-free interval EEG not within epileptogenic zone and seizure EEG within epileptogenic zone separately. Classification results of five different PAC methods were similar to each other. Furthermore, it was shown that there existed significant coupling at band combination $\theta -\gamma$ for EEG of seizure activities by observing from the comodulograms, which were consistent with the classification results.