Detection of Focal EEG Signals Employing Weighted Visibility Graph

Focal epilepsy is a neurological disease, stem from a specific region of human brain, which is known as epileptogenic focus. Diagnosis of focal seizures is usually done using EEG analysis. In this paper, a novel technique to discriminate EEG signals into focal and nonfocal categories is proposed employing weighted visibility graph theory. Visibility graph provides a topological representation of any signal while preserving its temporal characteristics. In this present contribution, the EEG signals of focal and nonfocal categories are transformed into complex networks employing weighted visibility algorithm. From the transformed signals, several feature parameters were extracted and their statistical significance was examined using one-way analysis of variance test. Finally, the features were served as inputs to a support vector machines classifier for the purpose of classification. Two important parameters of weighted visibility graph i.e. penetrability and scale factor have been varied to investigate their effect on the performance of SVM classifier. It has been observed that both of these two previously mentioned parameters directly influence the classifier performance. In the present work, highest classification accuracy of 92.5% has been obtained in discriminating focal and nonfocal EEG signals. The proposed method can be potentially implemented for real-time detection of focal EEG signals.