I'm entering radiology residency, and I'm very pro-automation / machine learning. There's a contentious debate in the field about whether radiologists will be replaced:
https://forums.studentdoctor.net/threads/will-ai-replace-rad...HackerNews is very developer-focused. If you guys saw what a radiologist does on a 9-5 basis you'd be amazed it hasn't already been automated. Sitting behind a computer, looking at images and writing a note takes up 90% of a radiologist's time. There are innumerable tools to help radiologists read more images in less time: Dictation software, pre-filled templates, IDE-like editors with hotkeys for navigating reports, etc. There are even programs that automate the order in which images are presented so a radiologist can read high-complexity cases early, and burn through low-complexity ones later on.
What's even more striking is that the field of radiology is standardized, in stark contrast to the EMR world. All images are stored on PACS which communicate using DICOM and HL7. The challenges to full-automation are gaining access to data, training effective models, and, most importantly, driving user adoption. If case volumes continue to rise, radiologists will be more than happy to automate additional steps of their workflow.
Edit: A lot of push back from radiologist is in regards to the feasibility of automated reads, as these have been preached for years with few coming to fruition. I like to point out that the deep learning renaissance in computer vision started in 2012 with AlexNet; this stuff is very new, more effective, and quite different than previous models.