Quantifying Functional Connectivity Network Through Synchronization and Graph Theory Approaches for Seizure Prediction

Epilepsy is a brain network disorder, which may be due to unusual neuron firing in the brain region causes seizures that lead to a loss of consciousness. Patients with epilepsy could benefit considerably from the capacity to precisely predict seizures. The ability to predict seizure activity is dependent on the proper identification of seizure activity precursors from electroencephalography recordings. Considering the fact that a huge list of characteristics has been offered, none of them can independently describe brain states. This method is to study the EEG signal through coherence and phase synchronization measures. Phase Locking Value is computed to measure the phase synchronization. Channel interconnection is accessed through functional connectivity index (FCI) features. The bispectrum measure is proposed to account for both the signal's amplitude and the degree of phase coupling between two frequencies. Thus, this work focuses on implementing a novel FCI based feature that is Bi-Spectral Phase Concurrence Index which exhibits a 3-dimensional frequency triplet mapping of the EEG signal interaction levels. FCI features have been measured through mean magnitude, normalized bi-spectral entropy and normalized squared entropy of the bi-spectrum along with graph theory approach. The bispectrum's quantitative properties were extracted and put through statistical tests to see if there were any significant changes between preictal and interictal recordings. Using all bispectrum-extracted features, the results revealed statistically significant differences (p < 0.05) between preictal and interictal states. The graph theory is also used to analyze the brain connectivity pattern for seizure and on-seizure state. Further, this study analyzes the efficiency and transferability of the proposed model for multi-class tasks with great clinical significance using different electroencephalogram datasets. The average classification accuracy, Specificity and Sensitivity of 99.79%, 100% and 98.6% were respectively are achieved.