Epilepsy seizure detection using kurtosis based VMD’s parameters selection and bandwidth features

Abstract

This paper presents an automated seizure detection method based on variational mode decomposition (VMD). In VMD, the number of decomposed modes K and the penalty coefficient α are selected empirically based on experience and observation which impacts its adaptability. To overcome this difficulty, we have proposed a novel method based on kurtosis to select K and α automatically for VMD decomposition of EEG signals. Primarily, the five sets of EEG data obtained from Bonn University database are decomposed into bandlimited intrinsic mode functions (BIMFs) using VMD with parameters K and α selected using the proposed kurtosis method, then amplitude modulation bandwidth (AMBω), frequency modulation bandwidth (FMBω) and spectral features of VMD decomposed BIMFs are evaluated. The significant features are obtained using Kruskal-Wallis test and fed to four different classifiers including Decision Tree (DT), K-Nearest Neighbor (KNN), Support Vector Machine (SVM) and Random Forest (RF) for five clinically relevant classification cases. Experimental results show that our presented work deals efficiently with all the five classification cases and accuracies greater than or equal to 98.7% have been achieved using RF classifier. Finally, in comparison with related works, our proposed seizure detection scheme performs better with higher accuracies in all the five cases.