Journal of Neuroscience Methods 199 (2011) 108–118 Contents lists available at ScienceDirect Journal of Neuroscience Methods j o ur nal homep age: www.elsevier.com/locate/jneumeth Detecting overlapped functional clusters in resting state fMRI with Connected Iterative Scan: A graph theory based clustering algorithm Xiaodan Yan a,b,∗ , Stephen Kelley c , Mark Goldberg c , Bharat B. Biswal d a Cognitive Science Department, Rensselaer Polytechnic Institute, Troy, NY 12180, USA b Center for Neural Science, New York University, New York, NY, 10003, USA c Computer Science Department, Rensselaer Polytechnic Institute, Troy, NY 12180, USA d Department of Radiology, UMDNJ-New Jersey Medical School, Newark, NJ 07103, USA a r t i c l e i n f o Article history: Received 19 November 2010 Received in revised form 30 November 2010 Accepted 1 December 2010 Keywords: Clustering Resting state fMRI Graph theory Brain network Hub a b s t r a c t The brain is a complex neural network with interleaving functional connectivity among anatomical regions. However, current functional parcellation algorithms usually emphasize independence or orthog- onality between the spatial components, with the interleaving nature underrepresented. This study investigates a method, Connected Iterative Scan (CIS), for identifying functionally overlapped anatomi- cal groups with resting state fMRI. CIS iteratively optimizes a grouping of vertexes in a weighted graph, using a density metric computed based on the input and output weights of a candidate cluster. In this study, CIS is able to detect the overlapped clusters in a simulated dataset. CIS also detects that the default mode network and the task positive network, which were known as two anti-correlated networks, are overlapped at the posterior cingulate cortex and the lateral parietal cortex. CIS also detects the conven- tional functional clusters in the whole brain neural network (e.g., the visual cluster, the motor cluster, the frontal cluster, etc.), as well as meaningful overlaps, and also revealed the possible existence of an emotional memory functional cluster. CIS was able to identify several hub regions actively participating in many clusters. With the ability to reveal overlapping functional clusters, CIS is potentially useful in revealing the delicate architecture of the brain neural network. © 2011 Elsevier B.V. All rights reserved. 1. Introduction In resting state fMRI, while subjects are not performing any prescribed cognitive tasks, functionally related brain regions have been observed with significant temporal correlations originating from the synchronization of low frequency spontaneous oscilla- tion (Biswal et al., 1995, 1997, 2010). The concept of functional connectivity has been applied to resting state fMRI to capture this temporal synchronization (Friston, 1994) in the resting state network (RSN). Many methods have been developed to analyze the functional connectivity, such as hypothesis-based analysis or data-driven methods. Smith and colleagues have published a com- prehensive summation on these methods (Cole et al., 2010). A common approach for these analyses has been to select an appropriate seed voxel or seed region based on a research interest, correlate the average signal from the seed with all other voxels in the brain, and determine the statistical significance of the correla- tion (Biswal et al., 1995; Cordes et al., 2000a; Li et al., 2000; Stein ∗ Corresponding author at: Cognitive Science Department, Rensselaer Polytechnic Institute, Troy, NY 12180, USA. Tel.: +1 518 892 0064. E-mail address: xiaodan.yan@yahoo.com (X. Yan). et al., 2000; Margulies et al., 2007). This approach, although simple to implement and popular in usage, is subject to researcher bias in the selection of region of interest (ROI) (Friston et al., 2006). Data-driven methods have also been developed as an effort to capture the intrinsic connectivity in RSN without researcher bias. These methods include clustering methods (Goutte et al., 1999; Cordes et al., 2000a, 2002; Stanberry et al., 2003; Golland et al., 2008; Morgan et al., 2008), principle component analysis (PCA) and independent component analysis (ICA) (McKeown et al., 1998; Oja and Hyvarinen, 2000; Calhoun et al., 2003; Smith et al., 2009). Recently there are some attempts to detect the communities among the neural network in the brain (Fortunato, 2009; Lancichinetti and Fortunato, 2009; Lancichinetti et al., 2010). All of these methods are based on the assumption that the neural network in the brain consists of several orthogonal or statistically independent com- ponents and aim to dissociate these components. Specifically, the assumption is made that functional components are disjoint from one another. However, this assumption may not be strictly justi- fied. Unlike anatomical brain regions, the boundaries between the functional clusters in the brain network may not be clearly demar- cated. Many brain regions have been found with multiple functions and should be considered to participate in many functional clusters. Therefore, it is meaningful and worthwhile to investigate functional 0165-0270/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.jneumeth.2011.05.001