Multi-modal brain age estimation: a comparative study confirms the importance of microstructure

Brain age inferred from neuroimaging data could reveal important information about the evolution of structural and functional cerebral features across the life span. This has important implications for understanding healthy aging and for identifying Imaging-Derived Phenotypes (IDPs) that characterise age-related neurodegenerative illnesses, such as Alzheimer’s and Parkinson’s disease. The so-called brain age delta refers to the difference between image-derived brain age and chronological age. Accelerated aging (positive delta) or resilience to aging (negative delta) have been found to be useful correlates of factors such as disease and cognitive decline. Multiple studies have proposed prediction models using brain IDPs as predictor variables, mostly relying on simple linear regression. However, methodological and population heterogeneity in these studies precludes definitive conclusions regarding the most informative modelling methodologies or predictor IDPs. To provide first hints in this respect, in this paper we propose to address three questions. First, four different state-of-the-art models are ranked based on well-known performance indices (e.g., mean absolute error) using the UK Biobank brain MRI data in different single/multi-modal settings. Second, for the best model, the association with individual IDPs are calculated to identify those that could play a prominent role in the aging process. Third, associations with non-brain variables are assessed as a first step towards a holistic approach. Our findings demonstrate a prominent role for dMRI IDPs in reducing the mean absolute error and rank high in the association study, dominating the first ten positions and being preceded only by three structural measures that are known to be related to the aging process. This provides evidence of the potential of dMRI IDPs as biomarkers of aging in health and disease.