Matching Functional Connectivity Patterns for Spatial Correspondence Detection in fMRI Registration

Abstract

A novel method is proposed to match functional connectivity patterns represented by graphs for spatial registration of fMRI data. Different from existing functional connectivity pattern based registration methods that detect corresponding functional units across different subjects by minimizing their difference in functional connectivity strength, our method adopts a graph representation to characterize functional connectivity information among all voxels in fMRI data of each subject, then detects spatial correspondence between subjects using graph matching. To integrate information of both functional connectivity strength and spatial relations, the graph representation of functional connectivity information of fMRI data models each voxel as one graph node and connects each pair of graph nodes with an edge weighted by their functional connectivity strength measure, estimated as correlation coefficient between their functional signals. To make the graph matching computationally feasible, an iterative matching strategy with stochastic resampling is proposed to match graphs of spatially distributed local functional connectivity patterns and subsequently to drive the image registration iteratively. The proposed method has been validated by registering resting state fMRI data of 20 healthy subjects. The validation experiment results have demonstrated that our method can achieve improved inter-subject functional consistency. A comparison experiment result has further indicated that the proposed method can achieve better performance than existing methods.