I agree, nevertheless it's exciting to see some progress being made (see the heart flow and iSchemaView companies cited below). Incidentally, there is a new angle to the radiologists push back - interventional radiologist are generally quite in favor of automating reads. This year IR is its own specialty, but coupled to three years of DR training. This is the speciality that I'm entering. I want to be an interventionalist while advocating for the adoption of machine learning systems on the diagnostic side.