Why AI Will Not Take Over Radiology

We can often read and hear AI will have the greatest impact on radiology. That radiologists will even be replaced with AI. These were my thoughts exactly, until I dug a little deeper with this article.

Why AI Will Not Take Over Radiology

A few weeks ago it was one of the first times I was exposed to radiology in a more clinical setting. It’s one of the fastest developing medical fields and that got me thinking about its future. In issue #21 I wrote that radiology could be one of the first specialities in medicine to lose significance because of AI. But that’s not necessarily true.

Pretty much every image that’s made in healthcare today is already in a digital format. That’s why the adoption of AI and deep learning in radiology is very high. It’s one of the easiest to add AI to and start using it and it's already improving image quality, helping to interpret images etc.

Naturally, with such trends come doubts about what its future holds. AI will only grow better and will only be more integrated into it. Will radiologists lose their significance? Or will that only be the ones not using AI?

Let’s suppose that a radiologist’s job is what everyone thinks it is. Sitting in a dark room, looking at screens, scrolling and clicking through a bunch of images and making diagnoses. In that case, it’s easy to imagine radiologists becoming obsolete. Why would they even bother with it if an automated system can do that?

These were my thoughts exactly, which I also wrote. However, I left out an important detail that I now know. Radiology as we know today (or a few years ago) is dead. But that doesn’t mean radiologists will be replaced with AI. Their work will just change to involve AI.

One of the most beneficial aspects of technology entering medicine is scalability. The population is getting older and the number of doctors won’t be able to follow the trend. The methods in radiology have improved substantially to make them more precise, faster and cheaper. It inevitably follows that a combination of developing technologies and ever more patients yields more work for radiologists. This is where automation comes into play.

The innovation did not replace real pilots, it augmented their tasks. On very long flights, it is handy to turn on the autopilot, but they are useless when you need rapid judgment. - The Medical Futurist
While I am certain that radiologists’ creative work will be necessary in the future to solve complex issues and supervising diagnostic processes; AI will definitely become part of their daily routine in diagnosing simpler cases and taking over repetitive tasks. - The Medical Futurist

It turns out that radiologists, and all doctors for that matter, will start doing the high-level tasks an algorithm simply can’t do as effectively. Or it doesn’t pay off to develop algorithms or systems that a radiologist can more efficiently do. Therefore, along with the vast clinical knowledge, it’ll be crucial to also know the basics of algorithms, data science, artificial intelligence and possibly coding.

The benefits of knowledge about these fields are two-fold. On the one hand, it’ll unquestionably be crucial for clinical decision-making and diagnosing. On the other hand, the more radiologists know about AI, the less they’re afraid of it. A survey published in European Radiology concluded that more than half of participants with intermediate or advanced knowledge about AI had an open and proactive approach to it. Those with less knowledge about it were (and are) more threatened by it.

Even though AI will have a big part in radiology (and medicine), it’s hard to imagine it could replace all radiologists. For example, trends are going towards minimally invasive surgery and treatment. Therefore, interventional radiology will only become more wide-spread. And there are no reasons to think AI or machines would replace interventional radiologists. We can’t even yet make cars completely driverless, now imagine surgery. But that’s not to say AI won’t be helpful.

The last point I want to make is medical education, which once again shows how it always lags - based on my experience in Slovenia. Understandably, we have to know the basics of medicine such as biochemistry, physiology, pathology, pathophysiology etc. in great detail. But the trend is clear that we’ll also have to know the basics of informatics and technology, which is not in any way present in the medical syllabus. A quote from Eric Topol comes to mind, where he describes medicine as sclerotic.

"Medicine is remarkably conservative to the point of being properly characterised as sclerotic, even ossified." - Eric Topol in Creative Destruction of Medicine

Radiology is a clear exception and the conclusion about it is clear. There will be radiologists that will adapt and use AI and radiologists that won’t. And the latter won’t have it easy.