Spectral clustering of resting state fMRI reveals default mode network with specifically reduced network homogeneity in major depression

Although resting state fMRI seems an ideal tool for investigating clinical populations, especially in case of reduced cooperation or tolerance, unbiased methods with high sensitivity for disease relevant pathologies remain to be identified. In this paper, we perform spectral clustering on the mean time series of automated anatomical labeling regions of interest for comparing the resting state networks in healthy volunteers and major depression disorder (MDD) patients. A new network homogeneity measure is suggested as a criterion for evaluating the level of homogeneity in a network. We found reduced network homogeneity specifically within the default mode network in MDD subjects compared to age-matched controls. In contrast to previously proposed methods investigating network homogeneity, we fully relied on data-driven definition of clusters of interest to fill an important gap between ROI based network analyses and those using ICA.