Cross wavelet transform as a new prototype for classification of EEG signals

Abstract

The main objective of this paper is to develop a computerised method that could be used to classify electroencephalogram (EEG) signals automatically and potentially help doctors, researchers and other medical personnel to detect epileptic signals accurately from a subject’s EEG recordings. In this paper, a Cross-Wavelet Transform (XWT) based feature extraction algorithm coupled with a few learning based classification techniques, like the Probabilistic Neural Network (PNN), the Least-Square Support Vector Machine (LS-SVM) and the Learning Vector Quantization (LVQ) is proposed to classify the EEG signals and compare the accuracy of the identification of epileptic activities. Benchmark EEG signals from the Bonn University are utilised to classify the EEG signals into the binary classes viz. Normal and Epileptic subjects. Also, a ternary classification model with categories being signals from healthy volunteers with their eyes open and eyes closed, signals from epileptic subjects during the seizure-free interval measured from within and outside the seizure generating zone of the brain and signals from epileptic subjects experiencing seizures has been put forward. The performance of the above-mentioned three supervised classification algorithms is compared by using the same training and testing datasets during stimulation. The accuracy of classification is obtained to be approximately 99%, 97.5%, 98.5% and 98.2%, 96.4%, 94% for binary and multiclass classification, respectively, using the PNN, LS-SVM and LVQ based classifier.