SVM based Automated EEG Seizure Detection using 'Coiflets' Wavelet Packets

Manual classification of ictal and non-ictal activities continues to be very perplexing even for any experienced neurophysiologist. Mostly due of the presence of considerable heterogeneity in the seizure patterns. Extensive research efforts have gone in solving this issue. But, the shortcomings and complexity of the deployed methods till date have been noteworthy to realize their practical applications. Present study showcased an expert system design for automated classification of ictal activities in electroencephalogram signals. The development used 'coiflets' wavelet packets for decomposition of signals to extract energy, standard deviation and Shannon entropy as features. Followed by support vector machine classifier with feds of various feature sets combinations. In the presented scheme, standard deviation feature set proved to be the best input features. It showed mean classification accuracy = 99.46 %, sensitivity = 99.40 % and specificity = 99.48 % with computation time = 5.60e-04 s. These outcomes demonstrated an improvement over the existing expert systems and also shed light on using different features. Proposed scheme hold promises for deployment in clinics and also for improvement in existing expert system designs.