Accelerometers-measured Physical Activity and Neuroimaging-Driven Brain Age (IMAGE)
Caption
A brain age prediction model is constructed by leveraging LightGBM algorithm training on 1425 image-derived phenotypes (IDPs) from T1-weighted brain MRI and chronological age. Features initially undergo tree-based feature importance ranking, where top 50 important features are picked out. Next, supervised distance between each feature is calculated then underwent hierarchy clustering to identify redundant feature groups. After removing redundancy, we visually interpret the final selected subset of features using SHAP technique. To deal with bias, predicted brain age was corrected by linear method. B We first investigate correlations between objectively measured PA and BAG using both nonlinear and linear models. Next, to gain insight into PA and brain structures, we investigate correlations between PA and 1425 IDPs using both nonlinear and linear models. C To verify whether PA and brain health was mediated by BAG, we conducted mediation analysis. Cognitive function and brain disorders were selected as brain health outcomes of interest.
Credit
Chen Han., et al, School of Public Health, Hangzhou Normal University
Usage Restrictions
Credit must be given to the creator.
License
CC BY