Radiologists Are Being Sold a Lie

If you are a radiologist, how many times have you heard the phrase, "AI won't replace radiologist. Radiologist who use AI will replace those who don't." Now, this quote gets repeated every day on LinkedIn in the radiology and AI community. And it's true. AI will make radiologist more efficient. But ask yourself, who will benefit from this increased productivity? That's what we will explore today. And by the end of this video, I will tell you whether I would recommend a radiologist career to medical students or even my own children. But first, let me show you why this famous radiology quote might reassure you more than it should, and what happens to regular workers when a technology makes them more productive. Later in this video, we sit down with the CEO of a radiology AI company whose team analyzed 3,500 open radiology jobs across the United States. We found that one type of practice pays $150,000 more per year, and those same practices are adopting AI the fastest. For most radiologist, however, that might not be good news. He drew an interesting parallel between radiology and software engineering. But before we go there, let's first talk about why you keep hearing the same sunny story about AI everywhere. People have talked about computers analyzing medical images for decades, but nobody really panicked until 2016. That year, in 2016, Geoffrey Hinton, the computer scientist often called the godfather of AI, made an infamous prediction. I think if you work as a radiologist, you're like the coyote that's already over the edge of the cliff but hasn't yet looked down, so it doesn't realize there's no ground underneath him. Um people should stop training radiologists now. It's just completely obvious that within 5 years um deep learning is going to do better than radiologists. >> Hinton's quote made radiologist panic. In one survey, 44% of US medical students said AI made radiology less appealing as a career. Big radiology had to come up with a response. So, 1 year later at RSNA 2017, one of the world's top authorities in radiology and AI made an intervention. Dr. Curtis Langlotz called AI an autopilot for radiologist. Then he announced a powerful new slogan for the field, "AI will not replace radiologist. Radiologist who use AI will replace those who don't." The panic began to fade, and by 2019, US residency applications for radiology bounced back to a 9-year high. Since then, the dominant story has been that AI will not take your job as a radiologist. Instead, it will make you more efficient. That sounds great, but here's the important question that nobody's asking. What happens to your income when your productivity doubles as a result of AI? Who benefits from your increased efficiency? Spoiler alert, it's not you. We can learn from the history books, from what happened to weavers, typists, and graphics designers when technology and automation made them more productive. This pattern was studied by the Nobel Prize-winning economist Daron Acemoglu. He concluded that automation increases the size of the pie, but labor gets a smaller slice. That means that automation typically leads to more growth. In radiology, there will be more scans, imaging for everybody, for anything at any time, with overall more money spent on imaging for a given population. But in the past, automation often went hand in hand with less money for the individual worker. So, the promised increase in productivity is real, but its benefits are not meant for you, the employed radiologist. The true beneficiaries will be AI companies, private equity firms, and investors. And maybe department chiefs who additionally benefit from clinical advisory roles, early investment into the exact AI companies they later employ in their own networks. Next time you see anybody repeating the famous quote that AI will not replace radiologist, but those who don't use it, check their LinkedIn profile, and I bet you will find some AI affiliation somewhere. Now, let's see what the CEO of a radiology AI company has to say about the future of radiology. And later, I'll show you what you can do to position yourself for exactly this scenario. CT packs and voice recognition all made radiologist more productive in the past, but imaging volumes grew just as fast. So, the field always needed more radiologist. AI is the first technology that could break this pattern. A Swedish trial showed that AI could replace one of two readers in breast cancer screening, cutting workload by 44%. And study after study is showing the same thing. Imagine you are a radiologist on a fixed salary. Let's say AI doubles your output. Now, you have to look at twice the number of images every day, double the risk for the same amount of money. Somebody has to pay the AI company, and it's you. If you have a productivity-based salary component, you will have an increased productivity due to the AI tools, but don't assume you will make double the money. There will not be a linear scaling of productivity and salaries. The AI company needs to get paid, the IT gets a share, your head of department still wants to fly business class, and insurance companies will reduce reimbursement fees because everything is now more efficient. The trend will be this, and it's already in place in many countries with or without AI. You as a regular employed radiologist will be paid less and less per study. So, the actual value of our work, the image interpretation, gets devalued over time. Now, you've probably heard the counterargument that there is a shortage of radiologist, so demand outstrips supply, and therefore salaries stay high. Well, okay, that is maybe true now in some places, but definitely not in Switzerland. But a shortage is historically why jobs get automated in the first place. As a company, you do not invest millions in automation when labor is cheap. You invest when it's expensive and scarce. We spoke to someone who sees this close up. Kirill Lopatin is the CEO of XAID. His company builds foundation models for CT reporting, and his team just analyzed 3,500 open radiology jobs in the US. >> PE-backed practices offers $150,000 more per year than those of academic ones or privately held, for example. In terms of sign-on bonus, for example, they offer 100K versus 35K. So, the difference is quite quite huge, and there are many radiologists that are seeking more income, and they're like trying to attend these PE-backed radiology groups, teleradiology groups. That's the reason why they able to scale so fast because the main resource right now in the market is a humans, human radiologists, because without them you cannot scale. And for PE-backed practices, it's much easier to attract new radiologists because they provide more money. They also provide better vision, better visibility in terms of the job boards and all that kind of things. PE-backed practices are more into getting more money, so they're more into optimization, and I hear more and more they're making the conditions under which radiologists are working, not in terms of the money, but in terms of the number of RVUs, the number of exams that they're doing daily, uh much strict. And also, I think that's uh they're more willing to employ new AI things because they need to. So, the CEOs, for example, of PE-backed practices, they have KPIs of improving margins, improving speed. So, they're eager to employ as many AI as possible and then try new things. >> These staffing decisions are not being made by radiologist. They are made by department heads, practice owners, and private equity firms. And once AI can match a radiologist's reporting accuracy, they will reduce their largest expense, salaries, for even greater profit. Okay, they will not put all radiologist onto the street at the same time, but if three radiologist get the same amount of work done as previously seven radiologist, you no longer need seven guys. You can get rid of four of them. At some point, AI will be so good that radiologist will not even have to change anything in these AI preliminary reports anymore, and you just need one radiologist to sign them all off as the legal scapegoat. That person will still be called a radiologist and will probably have a decent salary, but these other three former colleagues, I don't know, maybe are retired or transitioning into other fields or enjoy universal basic income. So, what can you actually do about it? Your future value as a radiologist will not depend on how good you are, unfortunately, or how much you know about AI. It will depend on two things. Are you an owner or an employee? Do you own part of the practice or not? Are you paid a salary, and how fast or productive are you? We asked Kirill where this goes next. >> Big big PE-backed companies are buying technical teams inside them. So, they are not trying to develop AI they're not trying to buy AI from external vendors, rather they are trying to develop their own because it is now clear that companies that employ both technical and medical things, like knowledge, will win finally because they cannot be separate, as it was like previously. It seems to me that private practice, privately backed radiology companies, will be first to find this kind of AGI for radiology, and it will lead them that they will start selling they will start providing their services at much lower prices than their competitors without this AGI or like whatever VLMs name it have how you want. These independently held practices without this technology will need to somehow buy this technology from them or just like put their radiologists inside that companies because they will be just like acquired because without this technology, definitely, it will be impossible to compete. >> Independent practices will either have to buy AI from private equity-backed companies or get bought themselves. So, if you own some part of a radiology center and sell it, at least you get, hopefully, a good payday. Or if you decide to stay in the game, you can apply AI yourself and get some form of a dividend. At least you have some form of control, or at least a sense of control, over this whole thing. If you are just employed, you cannot prevent your employer's practice or hospital from being bought up by private equity, and then they check the numbers, kick the radiologists with the lowest productivity out, and implementing AI to further enhance the efficiency of the top radiologists. Remember this, the slow radiologists will get sacked first. If you're fast, you can stay in the game a little longer. And even if you keep your job, the work itself will change. We asked Kirill what the radiologists role will look like in 5 to 10 years. It will just like put radiologists in kind of administration role because there will be many many different like radiology agents like we do like we see right now in coding. Actually, it is it is happening happening right now in coding. Radiology agents that are lower level, for them it will be much much harder to find job because it is happening right now in the coding, you know, like it doesn't worth to hire junior because you can like hire one senior and give him course or cloud whatever and he will be doing the same amount of job. So, in radiology it will be the same, I think. >> The future of radiology looks like what is already happening in software engineering. Senior people supervise algorithms as junior roles become less and less needed. My advice is, if you can get equity in a practice, do it now because that window is closing. If equity is not realistic, move towards work that AI cannot do yet like interventions, ultrasound procedures. And make sure to think about secondary income streams and work on your reporting speed. Would I recommend radiology to my own children? No. I love radiology as it is now. I will get a couple of more good years in before a major shift suddenly disrupts everything. But not just in radiology, but that's for another video. What I know is that radiology will not be the same in 20 years when my kids would start considering it as a career. My kids are still young though. If you are a medical student deciding on a career in the next few years, I can't fully heartedly recommend radiology either. I'm sorry, but that's just my honest opinion. I don't personally benefit if AI takes over. I have no AI affiliations, no advisory role, no stock in an AI company. The opposite even. If AI takes over, who needs the virtual MSK fellowship? [music] So, in a way I hope I'm wrong. Just because everybody is repeating the mantra that AI will not replace radiologists doesn't make it true. Tell me in the comments, do you own part of your practice? And if not, what's your plan? AI's impact in radiology is foreseeable and inevitable. The only uncertainty is the timeline. The real quote should be, AI won't replace all radiologists, private equity will replace the rest. But what does it look like when it plays out? What happens when AI signs 2,391 studies overnight and only a handful of radiologists are left to approve what the algorithm already decided? In the next video, we jump to the year 2030 and follow one of those remaining radiologists through a normal day so you can see what your role becomes if this trajectory continues.