FWIW my wife’s a researcher at Stanford applying ML/DL to MRI to speed up acquisition times and working closely with clinicians (I think they’re actually rolling it out in the hospitals) and the results have all been positive.