Nightmare on ML Street: the dark potential of AI in radiology
Hugh Harvey

Intriguing post. But when it comes to retrospective analysis you can only use methods that were already available at the time when the exam was originally performed. The same way I cannot sue the physicians of my late grandfather for not trying thrombolysis when he had an acute stroke, it is not possible to take legal action by relying on a then nonexistent algorithm. What needs to be demonstrated is that a human observer of the day working with the then state of the art medical knowledge could have described it as abnormal. Then again this varies a lot by jurisdiction.