Feature Extraction and Comparison of EEG-Based Brain Connectivity Networks Using Graph Metrics

Abstract

The human brain with an enormous number of interconnected neurons forms a complex network. Various techniques have been introduced by neuroscientists for analyzing the functionality of the human brain using different neuroimaging techniques. The recent trend for brain functionality analysis uses the concept of brain connectivity network. Brain connectivity network represents the association between functionally independent regions while performing cognitive tasks or in diseased conditions. This study implements the Electroencephalogram (EEG) signal-based brain connectivity network (BCN) corresponding to epilepsy diseased as well as healthy subjects. The BCN constructed for diseased as well as healthy subjects is analyzed by calculating graph-based metrics. Two types of graph metrics are calculated; (i) Graph metrics dependent on numbers of nodes and (ii) Graph metrics independent of numbers of nodes. Finally, the importance score is calculated for each metric and these metrics are compared to identify the best graph-based metric for the identification of epilepsy diseased subjects. On the basis of importance score obtained using decision tree regressor, it is found that participation coefficient metric from node-dependent type of metrics assigned maximum importance score.