A Clustering Strategy for Quantitative Assessment of Functional Connectivity in Resting-state fMRI Data

S. J. Kisner, T-Q. Li, Y. Wang, T. Talavage, V. Mathews, W. Kronenberger, D. Dunn Department of Electrical and Computer Engineering, Purdue University, West Lafayette, IN, United States, Department of Radiology, Indiana University School of Medicine, Indianapolis, IN, United States, Department of Psychiatry, Indiana University School of Medicine, Indianapolis, IN, United States Introduction Fluctuations in resting-state BOLD contrast images have been used to identify functional connectivity in the human brain. While many previous studies relied on the specification of “seed” location, the detection of functionally connected regions may be performed automatically through a clustering process. This approach may greatly reduce the subjective character of seed-based methods, and also allow a broader class of analyses that consider multiple regions across the brain. Two issues need to be addressed in characterizing functional connectivity: 1) Which brain regions are correlated with a given region of interest (ROI)? 2) How strongly is the ROI correlated with other areas? When a “seed” location must be specified for the analysis, the result is not only dependent on the location but also the size of the seed, because the average time course over a seed area is used as the reference for cross correlation computation. With the clustering approach this arbitrariness may be reduced by considering each voxel time course in turn as the reference and identifying voxel groups with strong inter-correlations. In this study, we further extend the clustering approach into the context of a comparative study between two groups of subjects. Two characteristics of interest are the spatial extent and the “strength” of the clusters. Spatial extent may be measured by voxel count and the number of spatially isolated components comprising each cluster, while cluster strength may be measured by the average correlation coefficient over all pairs comprising a cluster.