There has been a lot of work on artificial intelligence, particularly in the setting of medical care and radiology. Studies have shown that in some settings and narrow applications, AI (artificial intelligence) can do as good a job as a trained physician and radiologist.
There is motivation to develop AI to have more accurate interpretations of imaging studies, lower costs, provide more coverage especially during off hours and rural areas, and to relieve some of the burden on radiologists. At this time, there is little use in many practices. It currently serves as a complement and aid to the radiologist. AI is no where near interpreting imaging studies independently.
In fact, I am of the opinion that AI will not interpret studies independently during the lifetime of anyone reading this article. For one, who will take on the liability? Additionally, radiology studies are very complex. There is variability in patients anatomy, all types of imaging artifacts which degrade images or mimic disease, enormous variety of pathology and what it looks like on imaging studies.
Additionally, the radiologist must often use judgement and make a call based on many factors on the imaging itself and outside of it like clinical information and patient characteristics. That is why a radiologist needs strong medical training prior to becoming a radiologist. There are also many studies with findings that don’t look like any one disease and need a list of possibilities based on probabilities.
Radiologists also interact with technologists on various issues that come up while scanning patients. Radiologists interact with referring physicians and even patients at times. Radiologists perform image guided procedures of all kinds. Treat allergic reactions to contrast. The radiologist is not going anywhere and will likely supervise and interpret imaging for the foreseeable future.
I think that AI will assist the radiologist and perhaps make him more efficient but I do not see the radiologist being entirely replaced by AI. There is too much liability and complexity involved in imaging. There is too much that the radiologist does which falls outside of image interpretation and generating reports.
While the motivation is strong to create software that will independently interpret imaging, it is unlikely to happen any time soon. There will always have to be a radiologist to oversee imaging. It may reduce the demand for radiologist services because of efficiency, but I highly doubt we will see final interpretations and reports generated by AI.
I highly doubt most physicians will feel comfortable acting on information provided by software. For example, a surgeon rushing a patient to surgery based on an AI printout. I doubt AI will be able to answer questions or interact with technologists, doctors and patients. Who do you blame if something goes wrong? I think everyone should feel safe knowing that radiologists will always sign off on radiology reports for the foreseeable future.