A method based on the granger causality and graph kernels for discriminating resting state from attentional task

Abstract

Exploring the directional connections between brain regions is of great importance in understanding the brain function. As a method of this exploration, Granger causality is defined in terms of the amount of improvement in the estimation of a signal by past samples of another signal (cause). This method produced reliable results in various applications. In current study, we use connections of directed graphs as the features for discriminating two brain states, rest and attentional cueing task, in a block design fMRI dataset. We apply a support vector machine (SVM) which is enriched by graph kernels like random walk, graphlet and sub-tree kernels on directed graphs of different brain states. Graph kernel methods are a branch of graph matching methods and have recently been proposed as a theoretically sound and promising approach to the problem of graph comparison. They measure the inexact similarity between graphs. For the first time, we apply graph kernels on graphs of brain’s effective connectivity. We achieved classification accuracy of 100% in discrimination of resting state from attentional task. We also obtain one graph for each brain state representing causal connections between brain regions. From the networks obtained for each state, we can infer that caudate is the source of information in both states and Left ventromedial prefrontal is the sink of information in the resting state.