Inferring Individual-Level Variations in the Functional Parcellation of the Cerebral Cortex.

Abstract

Functional parcellation of the cerebral cortex is variable across different subjects or between cognitive states. Ignoring individual-or state-dependent variations in the functional parcellation may lead to inaccurate representations of individual functional connectivity, limiting the precision of interpretations of differences in individual connectivity profiles. However, it is difficult to infer the individual-level variations due to the relatively low robustness of methods for parcellation of individual subjects. We propose a method called "joint K-means" to robustly parcellate the cerebral cortex using functional magnetic resonance imaging (fMRI) data for contrasts between two states or subjects that intended to characterize variance in individual functional parcellations. The key idea of the proposed method is to jointly infer parcellations in contrasted datasets by iterative descent, while constraining the similarity of the two pathways in searches for local minima to reduce spurious variations. Parcellations of resting-state fMRI datasets from the Human Connectome Project show that the similarity of parcellations for an individual subject studied on two sessions is greater than that between different subjects. Differences in parcellations between subjects are nonuniformly distributed across the cerebral cortex, with clusters of higher variance in the prefrontal, lateral temporal, and occipito-parietal cortices. This pattern is reproducible across sessions, between groups, and using different numbers of parcels. The individual-level variations inferred by the proposed method are plausible and consistent with the previously reported functional connectivity variability. The proposed method is a promising tool for investigating relationships between the cerebral functional organization and behavioral differences.