> I'm not sure the answer here.

Has anyone started making lots (like tens of thousands) high-resolution, 3D, video (i.e., not still snapshots, but over time and ideally during episodic moments) MRI and CAT scans of diagnosed psychiatric patients, pointed a deep learning system at the scans, and see if patterns can be detected? If the deep learning systems can pick up patterns we didn't notice before, then that could be a valuable empirical tool to add.