Please explain; because this machine learning is the only reason I decided against pursuing diagnostic radiology. There is a small but very real possibility that radiologists will either lose their jobs or decrease their compensation. I see no way which the their work can be benefitted. Even if it is 20 years in the future, that means if I do pursue radiology, at age 43 I would be obsolete. It is simply too risky to base an entire career around it.
This sentiment is shared by many leading experts in AI.
Besides that, I see it as a perfect specialty so it is very unfortunate.
I’m wary of making this post for fear I steer this thread to AI’s impact on medicine and radiology. There is enough literature on it.
Well the question originally asked in this thread regards the fastest evolving medical specialty, and AI is likely going to be part of that equation, so I don't think the discussion is entirely off-topic; however, I hope this thread doesn't steer too far into that territory either.
I'm not sure what the consensus is regarding the timeline of radiologists becoming obsolete amongst experts in AI, but from what I've seen on this forum at least, I think many people see it as being more like 30-40 years away. I think that there are still many technical and regulatory barriers for AI to overcome in order to become better than a human radiologist, and I think until that day does come (if it does), it has the potential to improve the radiologist's workflow, increasing the number of scans they can read and actually making them more productive. Moreover IMO, there is so much focus on AI/machine learning eventually taking over the clinician's role and not enough focus on how it can make the radiologist even more valuable to patient care. For example, one could imagine that AI/machine learning may be able to detect patterns based on imaging features and combine that with clinical/genomic data to make prognostic determinations and predictive recommendations that could not have otherwise been made (I believe this is one of the goal's of the field of radiogenomics). In that sense, I think AI/machine learning has the potential to make the radiologist even more valuable to patient care than they already are, which would certainly be a plus for going into the specialty.
My rationale for considering pathology is similar. Currently, pathologists serve a crucial role as consultants to clinicians in providing diagnostic information that helps drive patient care in a major way. Now, there are recent papers showing that for certain use cases, prognostication using AI/machine learning analysis of H&E images can be more accurate than traditional pathological grading/staging. One could imagine that adding genomics and clinical data as features into these algorithms could improve the accuracy of these models even more. In this scenario, not only can the pathologist provide the diagnosis, but may be able to provide additional prognostic information to the treatment team that was previously unavailable before AI. Certain imaging and genomics features may even allow us to predict which patients will respond best to which treatment. I don't think anyone is better positioned than pathologists to provide this information, considering they have access to the imaging and genetic signatures of patient's samples. Obviously there is more to pathology than tumors, but when it comes to cancer, I could see a world (or at least hope to) in which the pathologist's role is almost like that of a data scientist, integrating different tumor markers, imaging features, and patient characteristics to help guide prognostication and treatment selection (which is essentially what they already do, just without the help of AI analyzing massive data sets and the insights that may come from doing that).