Exploring Links between Brain Image-Derived Phenotypes and Accelerometer-Measured Physical Activity in the UK Biobank

Abstract

A broad range of neurodegenerative disorders are associated with altered functional connectivity (FC) patterns and atrophy of gray matter volume (GMV). Similarly, there are links between physical activity (PA) and a number of neurodegenerative disorders. However, studies investigating the link between brain image-derived phenotypes (IDPs) and PA remain limited. Using data from the UK Biobank, we investigated the multivariate association between two sets of brain IDPs (related to FC and GMV) and PA using canonical correlation analysis (CCA). We further quantified the importance of individual PA variables in modeling each set of IDPs using both supervised and unsupervised approaches, and assessed their predictive performance for individual brain phenotypes. Finally, we evaluated the predictive performance of brain IDPs and PA variables for diabetes, stroke, coronary heart disease (CHD), and cancer using nested logistic regression models, with their relative contributions to explained variation in disease status quantified using a coefficient of determination specifically designed for logistic regression. Our analyses identified a statistically robust but low-dimensional axis of shared variation between PA and FC (canonical correlation r = 0.50), whereas the corresponding association between PA and GMV was weaker (r = 0.19). Brain features contributing most strongly to these associations were located in motor- and attention-related networks. Across predictive models, a small set of correlated PA measures reflecting activity intensity and circadian rhythm consistently emerged as representative predictors of both FC and GMV variation. Finally, we found that PA variables demonstrated greater predictive utility than either FC or GMV alone, particularly for CHD and diabetes, as assessed by both the area under the receiver operating characteristic (ROC) curve (AUC) and the proportion of explained variation. Together, these findings indicate that objectively measured PA is strongly associated with a set of motor-related brain features and provides substantial predictive information for cardiometabolic disease risk, while cross-sectional neuroimaging measures offer more modest incremental explanatory value.