Biological brain age prediction using machine learning on structural neuroimaging data: Multi-cohort validation against biomarkers of Alzheimer’s disease and neurodegeneration

Abstract

Background:

Brain-age can be inferred from structural neuroimaging and compared to chronological age (brain-age delta), as a marker of accelerated/decelerated biological brain aging. Accelerated biological aging has been found in Alzheimer’s disease (AD), but validation against biomarkers of AD and neurodegeneration is lacking. We studied the association between brain-age delta vs biomarkers and risk factors for AD, neurodegeneration, and cerebrovascular disease in non-demented individuals. Furthermore, between-sex differences in the brain areas that better predicted age were sought.

Method:

We trained XGBoost regressor models to predict brain-age separately for females and males using volumes and cortical thickness in regions of the Desikan-Kiliany atlas (obtained with Freesurfer 6.0) from the UKBioBank cohort (n=22,661). Using this trained model, we estimated brain-age delta in cognitively unimpaired (CU) and mild cognitive impaired (MCI) individuals four independent cohorts: ALFA+ (nCU=380), ADNI (nCU=253, nMCI=498), EPAD (nCU=653, nMCI=155) and OASIS (nCU=407). Chronological age, sex, MMSE and APOE categories were available for all subjects. ALFA+, ADNI and EPAD cohorts included data for AD CSF biomarkers (Aß42 and p-tau) and amyloid-b/tau (AT) staging was performed using pre-established cut-off values, whereas for OASIS amyloid-b was determined by PET. White Matter Hyperintensities (WMH) were available as a marker of small vessel disease and plasma (ALFA+ and ADNI) neurofilament light (NfL) as of neurodegeneration. Linear regression models, including chronological age and sex as covariates were used to identify associations between brain-age delta and biomarkers. We identified the individuals at the 10th and 90th deciles to select those with higher (accelerated) and lower (decelerated) brain-age delta and tested for interactions between age and all the variables on brain-age delta.

Result:

Between-sex differences were found in the most predictive brain regions (Figure 1). Brain-age delta was positively associated with abnormal amyloid-ß status, advanced AT stages and APOE-e4 carriership. Furthermore, brain-age delta was positively associated with plasma NfL in MCI patients and an interaction between age and plasma NfL was found on brain-age delta of CU individuals (Figure 2).

Conclusion:

Biological brain-age can be estimated from structural neuroimaging and is associated with biomarkers and risk factors of AD pathology and neurodegeneration in non-demented individuals.