Brain anatomy prior modeling to forecast clinical progression of cognitive impairment with structural MRI

Lintao Zhang, Jinjian Wu, Lihong Wang, Li Wang, David C. Steffens, Shijun Qiu, Guy G. Potter, Mingxia Liu

Abstract: Highlights•A BAPM framework uses T1-MRIs to predict cognitive decline, with MRI reconstruction and tissue segmentation as auxiliary tasks to model brain anatomy priors.•Source MRIs need no category labels for auxiliary tasks and target MRIs require labels for fine-tuning, allowing different label distributions and boosting BAPM utility.•BAPM pretext model, trained on over 9,000 ADNI T1-MRIs, aids image reconstruction and tissue segmentation in other applications.•This is an early attempt to train an encoder-decoder on large data via auxilinary tasks for prediction on small target data.