Model-free characterization of brain functional networks for motor sequence learning using fMRI

Abstract

Neuroimaging experiments have identified several brain regions that appear to play roles in motor learning. Here we apply a novel multivariate analytical approach to explore the dynamic interactions of brain activation regions as spatio-temporally coherent functional networks. We acquired BOLD fMRI signal during explicit motor sequence learning task to characterize the adaptive functional changes in the early phase of motor learning. Subjects practiced a 10-digit, visually cued, fixed motor sequence during 15 consecutive 30 s practice blocks interleaved with similarly cued random sequence blocks. Tensor Independent Component Analysis (TICA) decomposed the data into statistically independent spatio-temporal processes. Two components were identified that represented task-related activations. The first component showed decreasing activity of a fronto-parieto-cerebellar network during task conditions. The other exclusively related to sequence learning blocks showed activation in a network including the posterior parietal and premotor cortices. Variation in expression of this component across individual subjects correlated with differences in behavior. Relative deactivations also were found in patterns similar to those described previously as “resting state” networks. Some of these deactivation components also showed task- and time-related modulations and were related to the behavioral improvement. The spatio-temporal coherence within these networks suggests that their elements are functionally integrated. Their anatomical plausibility and correlation with behavioral measures also suggest that this approach allows characterization of the interactions of functional networks relevant to the task. Particular value for multi-variant, model-free methods such as TICA lies in the potential for generating hypotheses regarding functional anatomical networks underlying specific behaviors.