A DWT-Band power-SVM Based Architecture for Neurological Brain Disorders Diagnosis Using EEG Signals

Abstract

Electroencephalogram (EEG)-based signal processing techniques are important clinical tools for diagnosing and monitoring neurological brain disorders such as autism and epilepsy. In this paper, different methods for diagnosing autism and epilepsy by using discrete wavelet transform (DWT) and support vector machines (SVM), are investigated. For features extraction, DWT is combined with standard deviation, kurtosis, and logarithmic band power (LBP). The aim of this investigation is to recommend a combination approach that achieves best results. The proposed methods are tested using two types of datasets. The epilepsy dataset provided by MIT includes 23 subjects while autism dataset provided by King Abdulaziz Hospital includes 19 subjects. The simulation results indicate that the combination of DWT+LBP+SVM provides the best results with average classification accuracies of 98% and 96.5% for epilepsy and autism diagnosis, respectively.