Why AI will not replace radiologists

In my opinion, although this wave of enthusiasm and optimism has successfully brought radiology artificial intelligence to the forefront of people’s imaginations, and immense amounts of funding with it, it has also done untold harm by over-inflating the expectations of policy and decision makers, and is having tangible knock-on effects on recruitment as disillusioned junior doctors start believing that machines are indeed replacing humans and so they shouldn’t bother applying to become radiologists..This is starkly counter-balanced by sensationalist headlines along the lines of “machine beats radiologist” which only serve to further misinform the general public on the real state of AI currently, misleading them into thinking radiologists’ days are numbered.However infatuated or convinced you are about the possibilities of AI and automation, it is simply not realistic to expect it to entirely replace human radiologists in the near future, if at all. My estimate is 10 years until we see AI in routine NHS practice — and my opinion here is now a matter of parliamentary record!.I know this may be controversial given the amount of hope and hype currently, and maybe even surprising from someone like me who has essentially dedicated their career to AI in radiology, but I believe it is absolutely crucial to have sensible discussions on the future of the profession, rather than listen solely to silicon valley evangelists and the media who, let’s admit it, haven’t a clue what it is radiologists actually do, and just love to overplay the power of what they are peddling.In this article I’m going to attempt to break down the three main reasons why diagnostic radiologists are safe (as long as they transform alongside technology), and even argue why we need to train even more.If there is one thing that I would like to scream at anyone who says AI will replace radiologists, it is this — radiologists do not just look at pictures! All of the media hype about AI in radiology pertains to image perception only and, as clearly visualised in my diagram below, image perception is not the totality of what a human radiologist does in their day job..(There’s even the separate profession of interventional radiology, more akin to surgery than image perception, also a profession suffering a workforce crisis, that is less likely to benefit from AI systems)..Simplified schematic of the diagnostic radiology workflow, with examples of where AI systems can be implemented..Image copyright @drhughharvey.The diagnostic radiology workflow can be simplified into its component steps as visualised above: from patient presentation and history which leads to decision-making on whether or not to image, and what type of imaging to perform, to scheduling the imaging, and automating or standardising image acquisition..And this is only for starters…While artificial intelligence can absolutely play a part in each of the steps in this diagnostic workflow, and even replace a human in some of them (like scheduling) it simply cannot replace a radiologist entirely..This to me is a pipe dream, especially given the current state-of-the-art in AI systems which are only just barely making it into clinical workflows at present, none of which are anywhere close to replacing radiologists’ image perception work in any significant sense.In 2017 not one single human being died in a commercial airplane accident..If we build systems that massively improve radiology workflow and diagnostic turnaround, we will almost certainly see a massive increase in demand for medical imaging.I’ve seen this with my own eyes — when I was a trainee, our department started a new initiative to try and reduce waiting times for ultrasound lists..My point here is that in radiology, if you offer a doctor a slot to scan a patient, they will find a patient to fill that slot!As AI becomes the new normal in radiology, as scan times and waiting lists reduce, and as radiology reports become more accurate and useful, we will continue to see an increase in demand for our services..It won’t quite be like Minority Report, but if you want to imagine yourself as Tom Cruise swiping and gesturing away at a screen of futuristic malleable real-time data, then go right ahead.Where radiology artificial intelligence is heading towards is digital augmentation of radiologists, to the point at which their job becomes to monitor and assess machine outputs, rather than manually go through every possible mundane finding as they do now.. More details

Leave a Reply