 Here we go again. It looks like my job is on the chopping block yet again as artificial intelligence is coming for radiology. It's a sad day. Everybody, sad day. Welcome back to the channel, everybody. For those of you who are new around here, my name is Michael AKA Dr. Chalini and I'm a board certified diagnostic and interventional radiologist in New Jersey on today's video. We are going to be looking at the latest AI versus radiology article that I swear everybody in their family has sent me on Instagram and TikTok and even YouTube. This article is making its way through social media and it's obviously prompting the typical doom and gloom phrases like, if you're a med student, you better think twice about going into radiology. Looks like radiology is starting to be replaced by AI. Don't go into radiology because it's going to be replaced by computers. I wish I had a dollar for every time I heard statements like these. Anyways, let's go ahead and dive into this article and see why my job is on the chopping block. Let's go. Okay. So we'll get straight into it here. So the article we are analyzing today is from the verge and it is titled first autonomous x-ray analyzing the AI is cleared in the European Union or EU. And the subheading is the AI imaging tool reads chest x-rays without the involvement of a radiologist sounds intriguing. Let's go further. And then it shows some picture of like probably the oldest partially imaged chest x-ray I've ever seen from like the 1900s or something like out of all the chest x-rays and the medical advances we have, and we're talking about AI, they use this like, what is this chest x-ray? Is that the first chest x-ray ever? I hope so. Okay. So as you can see, this article is obviously a little bit clickbaity in that's why it grabbed my attention and everybody else who saw it. And the problem with this clickbaity title is that most people probably just read this and then immediately posted it on Instagram without actually reading the article itself, which is classic for social media. But I'm not that person. I read the article and we are going to read the article together and talk about it. So first paragraph starts off here. An artificial intelligence tool that reads chest x-rays without oversight from a radiologist got regulatory clearance in the European Union last week, a first for fully autonomous medical imaging AI, the company called OXOPIT said in a statement, it's a big milestone for AI and likely to be contentious as radiologists have spent the last few years pushing back on efforts to fully automate parts of their job. Okay. So this first article, this is actually kind of cool. The EU approved a fully autonomous medical imaging AI that can read chest x-rays without a radiologist. Pretty cool so far. But before I go any further, let's see what this company, OXOPIT, is all about. So let's go to their website. All right. So this is the OXOPIT website here. I guess we'll click on about us to learn something about this company. At OXOPIT, we create AI tools for global medical imaging. Our mission is to harness the latest AI and machine learning developments for better patient outcomes around the world. By utilizing artificial intelligence, we aim to address the global radiologist shortage and improve diagnostic quality. I'm intrigued. Go on. Also, look at this team. They look like a much a nice people. I'd like to work for them. Maybe hit me up. All right. So now we got the gist of OXOPIT or OXOPIT, whatever it is. Let's get back to the article. So going further into the article, the tool called ChestLink scans chest x-rays and automatically sends patient reports on those that it sees as totally healthy with no abnormalities. Okay. So it checks out the x-rays. If it's normal, we're good to go. If it's not normal, what happens? It goes on to say any images that the tool flags as having potential problem are sent to the radiologist for review. Okay, that makes sense. Most x-rays in primary care don't have any problems. So automating this process for these stands could cut down on radiologist workloads that OXOPIT said in informational materials. Okay. So basically this AI tool called ChestLink that was created by OXOPIT looks at the chest x-ray, analyzes it. If it's normal, it sends out a report. If it's abnormal, it goes back to yours truly, the radiologist. So one could say it looks like they still need us after all. Okay. Let's get a little closer look about ChestLink from the OXOPIT website. So per the website here, the first fully autonomous AI medical imaging product is ChestLink. It identifies chest x-rays with no abnormalities and produces finalized patient reports without any intervention from the radiologist, reducing radiologist workload and enabling them to focus on cases that present with pathologies. Do I want to learn more? Of course I do. ChestLink is the first fully autonomous AI medical imaging product with the CE mark. People are probably like, what the heck is CE? And I didn't know either until I lit it up and it's basically like the EU's version of FDA approval. It's not as difficult to obtain as FDA approval is, but it's kind of similar. So we'll just categorize it as such. And per the website, actually they are trying to get the FDA approval for this particular AI device as well. All right, more on ChestLink here. So an autonomy in medical imaging is not driven by technology, but the current systemic health care shortcomings the platform aims to address. It is safe to say that developed countries, the radiology departments are understaffed by one third. That's true. So here we basically just state these facts. Two thirds of the world population does not have access to radiology services. That's actually crazy. I believe that because in my residency at the University of North Carolina, we did a rotation in Africa and the country of Malawi, and there was one CTSD enter for the whole country that covered like hundreds of square miles. So I do believe that two thirds of the world population does not have access to radiology services. Another good point here, they say is demand for medical imaging examinations grows at a steady annual five to 10 percent rate. That's crazy. And I also totally believe that because volume continues to grow up takes eight years to train a radiologist. A little more than that. I would say 10 nine minimum, mostly 10 in the U.S. And depending on the institution, up to 80 percent of chest x-rays may feature no abnormalities. That's could be true. We'll touch on that a little bit. All right. So here is the ChestLink workflow diagram here. So we have the chest x-ray that comes in hot, goes into the oxyput AI, which analyzes it. There are no findings on the chest x-ray. It generates an autonomous report. Who knows what the heck is on that report? I wish they had an example to see if there is suspected findings. They send it to a radiologist who reports it as usual. OK, so let's get back to the article. So what does our professional organization think about this, the American College of Radiology? Well, they touch on it here. So the ACR and the RSNA or the Radiologic Society of North America in 2020 after FDA workshop on artificial intelligence in medical imaging said that autonomous AI wasn't ready for clinical use. So far, those two organizations have stated that AI programs are too inconsistent and often didn't perform as well on groups of patients outside of the original environments they were built in. Interesting. However, oxyput countered this with a statement saying that ChestLink made zero clinically relevant errors during its pilot program at multiple locations. Now, that's an interesting way of saying it made no mistakes. It made no clinically relevant mistakes. It didn't say it didn't make any mistakes. So it may have missed things, but they just weren't clinically relevant, which makes you think on what those items or things that they missed actually were. Did they like accidentally call a rib, a mass or something? You know what I mean? I wish I had more data on that. All right. So that is the brief overview of this article and oxyput in their ChestLink AI software or program. Now let's get into some of the questions I have about this. The first question is arguably the most important and it is brought up many, many times. Anytime we talk about AI and that is who is liable for the AI software. Is the company oxyput going to be liable? Is the hospital going to be liable? Or is the likely scenario that the liability is going to be placed on the radiologist, aka me and if the liability is placed on the radiologist, which I feel like it probably will be. How is that even fair? Say the liability was actually placed on the radiologist like myself, I would have to make sure I go back and double check all of that work. I would have to read all of those chest x-rays again after the AI analyzes it because ultimately it's my name and my license on the line. I have to ensure that no mistakes were made by the AI software, which is fine, but that essentially negates the whole point of having AI in the first place. The whole point of this is to allow radiologists to free up more time. But how does it free up more time when I just have to go back over the AI's work, especially when I can read a normal chest x-ray in a few seconds? So why don't I just read it to begin with and not use AI software? It's almost like I would be signing off a note that a resident wrote without reading it. It's under your name as the attending position or radiologist in this case. And if there are any errors in the note, you have to make sure you correct them before signing off on it. The second question I have is, will this actually help radiologists? I think the answer is yes and also no. That's the most classic answer for me. Yes, because on one hand, it may free up extra time where we don't just spend reading normal chest x-rays. However, these chest x-rays take about 10 seconds to read. And I'm talking about the normal ones here. And of course the 10 seconds per x-ray adds up when you're reading a ton of these normal chest x-rays. But to my prior point, what if I just have to over read all of these chest x-rays after the AI does? Then I'm just back to square one. Three, how much does this software actually cost? For starters, reimbursement on a chest x-ray is pennies. Essentially, we don't really get paid anything to read chest x-rays. So this software would actually have to be pretty affordable for it to make sense in a radiology practice or hospital or it would have to save money in the long run. So if it saves time and money in the long run, it might be worth it. The last question I have is, what do the patients think about it? Do they trust AI software enough to read their chest x-rays? Would they rather have a board certified radiologist read their imaging or a random program? I personally would rather have a board certified radiologist. So does this actually reduce the radiologist workload? I personally am a board certified interventional radiologist who is also board certified in diagnostic radiology. So in my work days, I mostly perform interventional procedures and also read various MRIs, CTs, ultrasound x-rays, et cetera. For me personally, chest x-rays are a very small portion of what I actually read on a daily basis. In terms of the normal outpatient chest x-rays, I probably read about 30 or so a day, which at about 10 seconds per x-ray, it takes me about five minutes to get through, maybe 10. Now this is not every signal practice. I am a private practice radiologist who works at a community hospital. So my case load and complexity may be lower than other hospitals. It also may be higher than some other hospitals. However, if you looked at the tertiary care center where I did residency, we would probably see around 200 chest x-rays per day. And I bet about 50 to 75% of those were normal outpatient chest x-rays. So we'll say maybe like 10, 20 minutes worth of work at the most. And I know people are going to say, it's impossible to read a chest x-ray in 10 seconds, but okay, maybe so, but sometimes you can if they're super normal. But we'll say 20 seconds per chest x-ray for the sake of the argument. So let's double everything I said. So maybe it takes 20 minutes to read all these normal chest x-rays. So the Oxford website stated that it clears up extra time for the radiologist to focus on more pathology forward exams like maybe a post-op CT or a staging cancer study. So maybe in that extra 20 minutes I could read, I don't know, two to three extra CTs that I wouldn't have gotten to if I had to read those normal chest x-rays. And that of course depends on how complicated they are. So yes, it may free up extra time for a radiologist, but is it worth those extra two to three CTs per day? Maybe if it is affordable to the practice or hospital and doesn't require more extra or hidden work from the radiologist, like over-reading every single chest x-ray, then it might be beneficial. If this is something like Bitcoin mining that just kind of runs in the background 24 seven, then it might be worth it. One thing that I do like about this software and that I touched on earlier is that it may help areas of the world that are lacking radiologists. Just let the AI read the normal x-rays and let the radiologists handle the abnormal ones. One thing I am worried about though is that maybe this just becomes the next EKG AI software. And anytime someone says AI is going to replace radiologists, I simply just mentioned the EKG. EKG artificial intelligence has been around for like decades now. And guess what? It still isn't reliable. I mean, we're talking about a line here. Like it can't even read a line accurately. So if the radiology equivalent of an EKG is a chest x-ray, let's just hope we don't have the same issue. Is it going to be not trustworthy and constantly having to be checked by a physician? Probably, but only time will tell. And let's not forget, we've had AI software in radiology for quite some time now. Breast imaging has had AI software to help them find abnormal masses and mammograms and tomography and MRI, but it's not widely utilized because it's not perfect. In fact, a lot of breast radiologists don't even use it. Some breast radiologists use it, some don't. And I can tell you one thing that none of them rely 100% on it. So what do I think about all of this? Am I worried I'm going to be losing my job to AI? Is a computer going to replace me? Well, probably not, or at least not anytime soon. I've always been a huge supporter of AI and radiology because if it's successful, it can help radiologists tremendously. Imaging demands of radiologists have been slowly increasing over time and I don't see an end in sight. It's almost like we're getting to a point now where we can't keep up with the demands of the imaging. There are just too many people being stand. After all, we are human and there's a limit on how fast we can read, safely that is. So this is where AI comes in. It's supposed to help relieve some of that burden on radiologists and free up our time to focus on things that demand our attention like very complicated studies. Again, if it's done correctly and reliably. I still worry about the liability though. Are insurance companies in the US going to pay for this? Or will the US population actually want this? I really am curious to hear your thoughts and I want you to comment below what you think. So in conclusion, I definitely want AI to be incorporated in radiology because I feel like it'll help us quite a bit. And I'm not scared about being replaced by AI because I just don't see that happening anytime soon. Maybe not even in my career. I'm not saying it won't happen but it might not be during my time. After all, we still have other jobs that could be replaced by AI relatively quickly but they haven't been. Like say for instance, cashiers. Some of them have been replaced by automatic checkouts but we still have them. What makes you think that radiologists are going to be replaced before cashiers? Are radiologists really the next person to be replaced before other jobs that don't require as much training? And this isn't a knock on cashiers. I'm just throwing that out for an example. Artificial intelligence in radiology is absolutely amazing and it's only getting better but it's currently nowhere near what it needs to be in order to replace radiologists in dealing with patients and their health. Like I said, on an AI video I did probably like two years ago link up here but just a warning it's cringe worthy. This was in my beginning stages of YouTube. Is artificial intelligence actually taking over radiology? But I said the same line on that video. I'm not anti-artificial intelligence in radiology and I don't think we should be worried about being replaced by AI. But if you're a radiologist, you better embrace it because it may help us out more than we think. As always, thanks for watching this video. I wanna know what you think in the comments below about this article and about AI in radiology. I think it's cool. Obviously I told you everything I think now I wanna know what you think. Make sure you gently subscribe to my channel. Follow me on Instagram and TikTok if you don't already and as always I'll see you all on the next video. Bye.