Considerations on brain age predictions from repeatedly sampled data across time

Abstract

Variability in brain age predictions complicate the metric's clinical usage, for example, as a (pre-) diagnostic tool. We presented small correlations between age and brain age when repeatedly sampling T1-weighted MRI data from the same individual in a short period of time (1-3 years). Reasons might lay in the absence of maturation effects for the age range in the presented sample, brain age model-bias (including a bimodal or trimodal age training distribution) and model error. We also find evidence for the influence of field strength and scan quality on brain age. Individual differences and the processing of such in the brain age model, might lead to variability in associations between brain age and QC metrics. The presented testing of an established brain age model on multiple single-subject short-timespan retesting data is a stricter test than the usual use-case and does not invalidate MRI group differences. However, intra-individual differences contributing to brain age require further attention in order to advance brain age as a clinical tool.