Adapting and applying the brain age paradigm for clinical imaging in multiple sclerosis (MS)

Brain-predicted age difference (brain-PAD; brain-predicted age – chronological age) is a potential biomarker for neurodegenerative diseases, including multiple sclerosis (MS). Previous models generally rely on T1-weighted (T1w) MRI brain scans. Here, we developed a deep-learning brain-age prediction model on FLAIR MRI. Our Inception-ResNet-V2 model was more accurate than a current state-of-the-art architecture and the FLAIR based model is comparable to a T1w MRI model. We used saliency maps, showing that areas such as the thalamus and ventricles are salient for brain-age prediction. We applied the FLAIR model to patients with MS, finding significantly higher brain-PAD compared to healthy controls.