Automatic Epileptic Tendency Screening using Statistical Features of MEG Data and SVM

In this study, a novel Magnetoencephalography (MEG) signal classification technique based on statistical features for classifying multi-channel signals into epileptic and healthy subjects is proposed. The method is composed of two phases: training phase and diagnosis phase. The multi-channel MEG signals are segmented into 1-minute non-overlapping segments; eight statistical features are extracted from each segment of each brain region to form the feature vector. The features are max, min, standard deviation, skewness, kurtosis, mean, median, and interquartile range. Using the feature vectors, a support vector machine (SVM) classifier was trained. The trained classifier then employed in the diagnosis phase. A four-fold cross-validation method used to evaluate the proposed technique. The proposed method is evaluated using real MEG data which includes 32 healthy subjects and 32 epileptic patients. The proposed method achieved a sensitivity of 99.35%, a specificity of 95.47%, and an accuracy of 97.41%. The obtained results show good promise of the proposed method as a screening tool for epilepsy diagnosis.