Raw LLM Responses

As a student who has been working with computational imaging algorithms, most of these in development are not intended to replace a radiologist's job, but rather to help aid them with a form of feedback. It's nearly impossible for algorithms to be 100% foolproof, as there will always exist some false positive rate. Nor is it possible that an algorithm can achieve a 100% correct diagnosis rate without some amount of false positive. On top of that, with the algorithm demonstrated in the video, the output may yield competing diagnoses, many of which could be false positives. There are many other external factors that may confirm or deny a diagnosis, such as the patient's medical history or family background. While images are easy to quantify as data, external qualitative factors are very difficult to feed into a machine learning or deep learning model. Under the circumstance of a potential false positive or multiple competing diagnoses, ultimately it is up to the radiologist and their peers to make the final call. Anyway, while I'm not an expert in either radiology or computation, after working in this area for a couple of years, I would say people shouldn't jump to the conclusion of AI taking over medical jobs. Patient data is a lot more finnicky than what people think, it's not clean spreadsheets, and many assumptions and estimates have to be made (like how do you quantify family history of lung cancer, is it a yes/no answer? Do we need to know how many generations have had lung cancer? Do we need to know which side of the family has it? How severe the cancer was? etc.)