Artificial intelligence (AI) is still front and center for the layperson when talking about the world of radiology. Family and friends always ask me why AI will not take over my day job. In one of those discussions recently, I had one of those “aha” moments. We talked about how many factors will prevent AI from taking over our jobs anytime soon. These included legal, ethical/moral, and financial reasons. As I was droning on, I realized I could argue why AI could overcome these issues. However, one reason not related to these is why we won’t see a blank screen or a computer person from India, China, or any other country for that matter replacing our presence for a very long time. And that is that medicine is local, not universal.
Medicine Is Local, Not Universal- AI Cannot Account For It!
Now, why would differing local “standards” be the most critical reason for keeping radiologists busy? Well, every country and every physician has their opinions about the best way to treat patients. Medicine is not universal. It is local. Yes, there are a few standards that are omnipresent, like the Hippocratic Oath not to harm. But, other standards like lung nodule management vary widely among physicians, counties, states, and countries. The Fleishner criteria for managing pulmonary nodules are not standard. Some folks use that criterion; Others use LI-RADS.; And even others use ELCAP.
I also know of some clinicians that modify all these criteria to fit their patient populations. Therefore, it is next to impossible to standardize standards in an AI computer algorithm when your physician wants to use a different bar from the rest. One great way to lose the radiology business is to make recommendations that run counter to your referrers!
Management Differences Between Places
Different countries have different standards of care. For example, it would not be appropriate to recommend imaging a patient with an MRI of the shoulder in Canada due to lack of availability. Over there, physicians may be more apt to order a musculoskeletal ultrasound. Likewise, a radiologist in Canada may be more likely to recommend a musculoskeletal ultrasound for a possible rotator cuff tear. Yet, in the United States, an MRI is part of a routine workup. Why? Because they have a much more significant backlog of patients waiting to get their studies done with fewer MRIs than we do in the United States.
Or, in China, clinicians may regularly recommend “cupping” for different ailments. How can AI programs account for each cultural difference among countries, states, or even counties, based on availability, preferences, and cultural norms? These obstacles would be exceedingly difficult to overcome.
Differences Between Surgical And Medical Preferences
We work for other physicians. Our role is to make their job easier to treat patients. And, each clinician has their own specific needs for caring for their population. Oncologists look at different criteria for assessment than surgeons. And, neurosurgeons have other interests than internal medicine doctors when they order a study. An AI program needs to take all these factors into account when they summarize findings and make recommendations. AI is not ready to make different individualized reports for each subspecialist clinician. It would take massive programming power for which it’s not ready!
Differences Among Individual Patients
And, finally, even among patients, culturally speaking, some patients want more aggressive workups, and others are more conservative. For instance, I may have a patient that can’t live with a small complex cyst in their breast and wants it drained. Meanwhile, another patient might be more willing to follow it. Some of these differences may be cultural or related to individual differences. How would an AI program account for that? AI is not ready to interpret the cultural and emotional status of every patient to make these decisions. Again, no one supercomputer could make these individual recommendations for patients.
A Radiologists Job Is Still Way Too Complex For AI!
Whether it is differing standards, cultural differences, physician preferences, or individual patient preferences, radiology, in particular, is not a one size fits all discipline. No program can take all of these issues into account to replace a radiologist within the foreseeable future. The amount of processing power required to figure this out for every clinician’s report would be enormous. Of course, 500 years later, maybe a program will accomplish all of these tasks to replace radiologists. But, by that time, the same computer will replace every other job on this planet. So, for those of you thinking about entering radiology, don’t let these issues dissuade you. Over your career lifetime, you will still have a job!