Multimodal MRI prediction of cognitive functioning across the lifespan: separating between-person differences from within-person changes

Abstract

Brain MRI shows promise for predicting cognitive functioning, but its utility depends on its capacity to capture stable between-person differences (e.g., patient stratification), longitudinal within-person changes (e.g., prognosis, treatment monitoring), or both. Using longitudinal data from 450 adults (aged 21-90; up to three waves, five years apart) in the Dallas Lifespan Brain Study, we benchmarked five modalities, task fMRI, functional connectivity (FC), structural MRI (sMRI), diffusion-weighted imaging (DWI), and arterial spin labeling (ASL), across 37 phenotypes and their combination. Stacking all MRI modalities into one marker predicted cognitive functioning with the highest accuracy (R{superscript 2}=.51), followed by DWI and FC. Variance decomposition showed MRI markers explained substantial between-person variance (up to 60.3%) but modest within-person changes (up to 17.2%) in cognitive functioning. Commonality analysis revealed most markers, except ASL, overlapped with age-related variance in cognitive functioning. These findings clarify the strengths and limitations of MRI markers for stratifying and monitoring cognitive aging.