Transfer Learning Models on Brain Age Prediction

The early diagnosis and intervention for neurological diseases such as Alzheimer's disease has been becoming more and more important in medical treatment. Some state-of-the-art studies have shown that people with an older brain-age, i.e., degenerating faster, than control people of the same age, may have a larger risk for cognitive decline. In this paper, we propose a method to build a model to predict brain age by end-to-end training of preprocessed MRI scans, of which the backbone of this model is a 3D-CNN. We firstly train on a large-scale dataset: UkBiobank, and conduct transfer learning on specific small datasets. We illustrate the performance on small datasets such as NKI and Cambridge that this transfer learning model can achieve better performance than the baseline methods reported in literature, with a MAE of 4.20 years at the top model.