Low-Rank Learning of Functional Connectivity Reveals Neural Traits of Individual Differences

Abstract

The intrinsic functional connectivity not only follows underlying anatomical connectivity architecture but also varies across individuals. However, how this connectivity variability underpins the diversity of cognitive and behavioral ability has not been well established. In this chapter, we sought to build a mapping between the individual behavioral differences and the inter-subject variability of functional anatomy in an “individual traits space” obtained by a background-subtraction technique. Applying a novel low-rank matrix recovery method, the functional connectivity was dissociated into a background skeleton that delineates functional characteristics common across the population and individual traits that were expected to account for individual behavioral differences. Subsequently, a sparse dictionary learning algorithm was performed on the extracted individual connectivity traits, identifying multiple individual difference units (IDUs) that were then used to map the inter-subject variability of cognitive behavior. The identified functional background skeleton was demonstrated to have high similarity to anatomical connectivity substrate. More importantly, the IDUs dissociated multiple sources of connectivity variability and consisted of an over-completed space representing individual behavior. Our findings suggest that steady anatomical substrates and multiple-dimensional functional traits shape an individual’s connectome and well support individual cognitive behaviors, which may advance our understanding of the relationship among anatomy, function, and behavior.