Detection of Epileptic Electroencephalogram Signals Employing Visibility Graph Motifs

In this chapter, we propose a novel technique for the detection of epileptic electroencephalogram (EEG) signals using sequential visibility graph (VG) motifs. VG is a technique, which converts any time-domain signal to an undirected and binary graph while retaining the signals’ temporal characteristics. However, one limitation of a VG-based approach is that it captures the variations present in a time series on a global scale instead of providing insight into the local fluctuations present in a signal. Considering the aforesaid issue, a novel method to analyze the topological variations of the EEG signals on a local scale using sequential visibility graph motifs is proposed in this chapter. For this contribution, we constructed a VG of the EEG time series and extracted different sequential VG motif profiles for a fixed number of sample points. The process was repeated by translating the window along the entire path of the graph (length of the signal), and finally the frequency of occurrence of different sequential motifs for a particular EEG signal was computed. In this way, the frequency of occurrence of different sequential motifs for healthy, inter-ictal, and seizure EEG signals was computed. For this chapter, we computed the frequency of occurrence of different sequential motifs using both conventional visibility and horizontal visibility graph methods for different classes of EEG signals and further examined their discriminative capabilities between the respective classes of EEG signals using a one-way analysis of variance (ANOVA) test. Finally, using the frequency of occurrence of different sequential motifs as the input features, EEG signal discrimination was carried out using several benchmark classifiers. We investigated the performance of our proposed methodology by procuring EEG signals from an online available benchmark database and observed that promisingly high detection accuracy had been obtained in classifying different categories of EEG signals, indicating the practicability of our proposed framework for the automated detection of epilepsy.