Standardized and Automatic Framework for Functional Connectivity Analysis: Functional Correlation Matrix and Sorted Index Curve

Introduction: Functional connectivity refers to the spatially coherent patterns of low frequency (< 0.1 Hz) and spontaneous fluctuations in the fMRI signal [1]. These patterns, which are evident at a resting state, appear to represent the internal dynamics of the same brain systems that are engaged by various tasks (e.g. sensorimotor, language, emotion, memory among others) [1]. In contrast to the conventional fMRI studies that rely on task-evoked BOLD signals, the resting-state functional connectivity studies do not require the subjects to perform any task, and are thus particularly well-suited for patients who cannot readily perform the various required tasks. This advantage may help resting-state connectivity investigations to find broad applications in translational medicine. Indeed, recent studies showed that the resting-state functional connectivity provides the information reflecting the neurological disease progression, the treatment efficacy, and even the visual cortex functional plasticity in the early blind [2]. Several post-processing approaches (such as correlation analysis and independent component analysis (ICA)) are currently available for functional connectivity analysis. Since these existing methods were originally designed to map the functionally connected brain regions, they are limited in several ways in identifying brain regions of very low connectivity (i.e. functionally independent or functional deficient due to neurological diseases). For example, in correlation analysis, the cross correlation coefficients between a pre-defined seed and other voxels are calculated, and the whole-brain connectivity to the pre-defined seed region is then constructed. This approach, however, requires a prior knowledge or assumption for selecting the seed region. In ICA, the lowpass-filtered time course profiles from all voxels can be simultaneously examined without a prior assumption, and multiple connectivity components may be identified. However, the ICA method cannot quantify the lowconnectivity regions that are not part of the identified connectivity components. To address the limitations in existing connectivity analysis methods, we design an automatic post-processing pipeline that can reliably identify functionally correlated regions in resting-state fMRI data, without a prior assumption. Experimental results show that the proposed method provides a consistent and reliable assessment of the functional connectivity, and may potentially be applicable for an early detection of functionally deficient regions in an individual.