Research Square · Oct 22, 2026
Translating Imaging-based Brain Age Models for Predicting Accelerated Brain Ageing in African Populations: A Pilot Study on a Nigerian Cohort
By Anazodo U, Olusuyi T, Cakmak J et al.
Background: Brain age is an emerging imaging biomarker for early detection of neurodegenerative disorders, based on the estimation of the brain’s biological age. Existing predictive models are predominantly trained on Global North (GN) neuroimaging data, limiting their sensitivity when applied to underrepresented populations due to geographic, imaging scanner, and biological heterogeneity. We investigated this bias by adapting the standard Gaussian Process Regression (GPR) method to a population-specific (PS) brain age model. Methods: Using a multi-institutional T1-weighted brain MRI dataset from 309 Nigerian adults, we implemented a robust low-compute image analysis pipeline which includes repurposing clinical scans and high-resolution cortical parcellation with multi-stage quality control. The PS-GPR model was trained on regional brain volumetric features from the Nigerian MRI scans and its performance compared to four common GN pre-trained models. Results: The proposed PS-GPR model showed a significantly lower prediction error (mean absolute error = 9.82 years) without age-bias correction and few seconds inference time compared to GN-trained machine learning models (mean absolute error = 10.00 – 16.70 years). SPARROW, a generalizable deep learning framework achieved an MAE of 7.87 but failed on 6% of the test set. This early work demonstrates the importance of representative data to build resource-efficient clinically viable models. Conclusion: This population-specific model is a step towards sensitive low-compute African brain age models. With the expansion of imaging data beyond Nigeria through the Africa Neuroimaging Archive (AfNiA), we will further develop the PS model to identify individuals with accelerated brain ageing for early intervention in at-risk adults.