A new algorithm for classification of ictal and pre-ictal epilepsy ECoG using MI and SVM

Abstract

Electrocorticogram (ECoG) is an effective way for Epilepsy research, as well as automatic seizure detection. This study proposes a method for feature extraction and classification of pre-ictal and ictal ECoGs, based upon mutual information (MI) and support vector machine (SVM) which has not only high accuracy but also fast speed. First, the mutual information among 76 channels is computed and converted into a 76×76 matrix, and then statistical significance of splicing mutual information between pre-ictal and ictal ECoGs is tested, and the coefficients of variation and fluctuation indexes of MI corresponding to the selected channels which exhibit the most significant differences are selected as features. SVM is then used as the classifier for identifying ictal ECoGs. In addition, two methods based on empirical mode decomposition (EMD) and wavelet, are applied on data as a control. The results of this study show that the MI of the selected channels from pre-ictal ECoGs is higher than that from ictal ECoGs, and the classification accuracy by combining coefficients of variation and fluctuation indexes as feature vectors is up to 100% and faster than the other methods.