AI ML: neuroimaging transfer learning

Goal: learn advanced machine learning models from a large sample size and transfer the model to a smaller sample size

Data:

Large sample size: Human Connectome Project, UK Biobank, ABIDE (autism), Open neuro. These open-access databases have samples from thousands (or more) subjects.

Most neuroimaging studies recruit about tens of participants. Training advanced machine learning models using such a small sample size is prone to over-fit.

Goal

Learn models from a database with a larger sample size (n>1,000). Then transfer part of the learned model to adapt to a smaller sample size (n~20).

The model can be targeted for estimating individual's age based on neuroimaging data.