Combining Spatial Independent Component Analysis with Regression to Identify the Subcortical Components of Resting-State fMRI Functional Networks

Functional brain networks are sets of cortical, subcortical, and cerebellar regions whose neuronal activities are synchronous over multiple time scales. Spatial independent component analysis (sICA) is a widespread approach that is used to identify functional networks in the human brain from functional magnetic resonance imaging (fMRI) resting-state data, and there is now a general agreement regarding the cortical regions involved in each network. It is well known that these cortical regions are preferentially connected with specific subcortical functional territories; however, subcortical components (SC) have not been observed whether in a robust or in a reproducible manner using sICA. This article presents a new method to analyze resting-state fMRI data that enables robust and reproducible association of subcortical regions with well-known patterns of cortical regions. The approach relies on the hypothesis that the time course in subcortical regions is similar to that in cortical regions belonging to the same network. First, sICA followed by hierarchical clustering is performed on cortical time series to extract group functional cortical networks. Second, these networks are complemented with related subcortical areas based on the similarity of their time courses, using an individual general linear model and a random-effect group analysis. Two independent resting-state fMRI datasets were processed, and the SC of both datasets overlapped by 69% to 99% depending on the network, showing the reproducibility and the robustness of our approach. The relationship between SC and functional cortical networks was consistent with functional territories (sensorimotor, associative, and limbic) from an immunohistochemical atlas of the basal ganglia.