In the analysis of resting state functional connectivity (RS-FC), functional magnetic resonance imaging (fMRI) data can be modeled as a graph of nodes  and edges representing brain regions or image voxels, to capture their existing interrelationship due to functional activities [1]. There are three key points in the analysis of the functional connectivity graphs that significantly influence inter-subject classification (to be explored in this work): 1) how to define the nodes; 2) how to define the  dependency measure between time series; and 3) how to define the graph theoretical features. Traditional graph theoretical analyses of functional connectivity using RS fMRI are based on measuring the correlation between time series of predefined nodes or random voxels in the human brain. However, higher-order interactions are not captured by correlation. Moreover, in most available research works, the functional connectivity graphs were defined by their binary adjacency matrix, whereas the functional connectivity graphs can be fully characterized by undirected weighted adjacency matrix. In this study, we developed a novel analysis technique for RS-FC technique in the following steps: 1) extracting the fMRI time series for 264 nodes, selected based on neurological principals; 2) calculating normalized mutual information to capture the complex and higher order interaction between two time series; and 3) constructing the weighted graphs called mutual information weighted graphs (MIWG) for each dataset.. We utilized the proposed method to demonstrate alterations in functional communication of patients with multiple sclerosis (MS) and in comparison with healthy controls. Results were compared with the classic Pearson correlation graphs employing SVM classifier.