Brain Age Prediction from Structural Magnetic Resonance Imaging using Deep Learning

Abstract

Brain-age delta, a biomarker based on the brain-age prediction from neuroimaging data, has garnered significant attention in recent years. In this project, a deep learning model for brain age prediction is going to be proposed, to enhance the validity of this biomarker. MRI data from 16772 subjects from the UKBiobank database was used for the model?s development. Additionally, an independent cohort, ALFA+, was employed to evaluate the model?s generalization ability. While the predictions on the UKBiobank dataset indicated an exceptionally low MAE error, this extreme precision did not translate nearly as well to the ALFA+ cohort. This report includes a literature review on brain-age delta biomarker research, along a presentation of multiple machine learning, brain-age prediction models. In addition, a methodology to deal with large databases and 3D images is also be illustrated. Overall, the study demonstrates that high brain-age prediction accuracy can be archived with deep learning models, but the network?s generalization capacity to other cohorts determines its convenience and validity.