Proactive Health Screening & Unlocking Hidden Radiology Data

Medical imaging volumes are rising faster than the workforce can keep up, leading to massive backlogs and dangerously delayed diagnoses. 

In this episode of MIAI, Stephen Pech and radiologist Dr. Chee Chong explore how AI can transform standard, reactive imaging workflows into proactive screening platforms that catch diseases long before symptoms appear. 

What You Will Learn:

The Backlog Crisis: Why massive volumes of unreported scans in modern health systems are directly stalling patient diagnoses.

Data Left on the Table: How standard human limitations mean critical markers for cardiac health, bone density, and liver fat are often ignored or missed on routine scans. 

Invisible Integration: The necessity of building seamless, automated AI tools directly into PACS servers so radiologists aren't bogged down by extra workflow steps. 

The Funding Paradox: A look at who pays for preventative AI technology when the massive financial returns take years to materialize. 

The 10-Year Vision: How AI will blend a patient’s entire imaging history to give GPs and specialists a predictive, holistic tool for lifelong lifestyle and clinical guidance.  

Transcript

SPEAKER_02 0:00

On LCT images, there's so many images and you you cover so many regions that your report could be pages and pages in length. That becomes information overload as well for the clinician who has to then read pages and pages to try and find a simple answer which you could have covered in the first paragraph. They're obliged to read all those pages. And sometimes people refer to certain practices or certain doctors because they know they're going to get a short report because it's their time having to read pages and pages of reports versus reading a short paragraph. So I wonder whether AI can help us not just to take more, but to be a lot more succinct in our reports.

SPEAKER_00 0:43

Welcome to MIAI, the podcast about medical imaging and artificial intelligence. Sit down with the most brilliant voices in radiology and AI to discover how technology is reshaping healthcare delivery today.

SPEAKER_03 0:58

Welcome to MIAI, the podcast where technology meets medical imaging. My name is Stephen Peck, and I'm here with radiologist Dr. Chi Chung. Hello everyone. And we explore how AI is reshaping radiology. Today we're talking about proactive health screening. Some of the key points we'll cover are the cost of late diagnosis and why we still diagnose disease too late. There's valuable information hiding inside every radiology image. Can we get it out? Make use of it? Also, can we turn existing radiology workflows to be AI-affied? Can we include AI in them and use that to create proactive screening platforms? This discussion is not about hype. It's about what's already possible and what is the impact for radiology practices for health systems and patients. So let's kick off, Chi. Why do we diagnose disease too late?

SPEAKER_02 1:54

Well, there's a lot of issues, I think, with our health system in Australia, which is a developed nation with lots of good funding for health, but radiology emergy volumes are rising a lot faster than what workforce is able to plug them up at. We've got multiple centres, public and private, with thousands of unreported cases. I think only recently in New South Wales there was a case where there was about 50,000 unreported cases. So if that being reported, they're not being diagnosed. 50,000 unreported cases of what? Of imaging. CTs, X-rays, ultrasounds. You and your peers hadn't got around to them. This is uh I don't have enough friends to report them all. That's that's the issue. So that's that's one issue. And we're talking about, you know, Australia, which has a very good health system, good standing in the world, but still we're falling behind, failing.

SPEAKER_03 2:48

Okay. So you've pointed out one key point there. Not enough radiologists. Not enough radiologists.

SPEAKER_02 2:53

Well, AI can probably help us with some parts. What do you think?

SPEAKER_03 2:56

Yeah, of course, I think uh can absolutely help. And I guess this is a key part of what we want to talk about today. There's information hidden in those scans. So we need to get that information out of them, and that's a key thing that AI potentially can do. We can be scanning for things beforehand. The reality is that there are clues in those images long before you're getting symptoms. So if we can pull that out early while it's there and hiding somewhere in that scan before the symptoms come up, that's that's a great opportunity.

SPEAKER_02 3:24

I think you're right there. I think as a radiologist, I have my limitations, I know my limitations. I know there's probably lots of data in the studies I look at that is being missed by humans. That's probably another way that we're diagnosing things too late by not finding things that's already latently on the film.

SPEAKER_03 3:44

Yeah. Just quickly to go back, G, I've seen you play golf, so I definitely know your limitations. But go back to your point, I agree. Did you want to expand?

SPEAKER_02 3:56

You talked about one point about being a late diagnosis, so we've touched about there's not enough radiologists to service the studies that are being performed right now, and so there is delay there. Ideally, you'd have the scan done and get a report within 30 minutes such that the clinicians can start treating if required on those studies, but we are lagging behind specifically. You've talked about, I guess, are we seeing everything that's already on the image? Like there are diseases that are on the images that we aren't catching. But I guess we also need to talk about what happens when we have a late diagnosis. As you can imagine, you don't need to be a doctor to know this, and I'm sure you're well-versed with it. Late diagnosis means a poor result. It's like catching anything early.

SPEAKER_03 4:44

Yes. That's the ideal, right? We've got it we want a health system which is telling us the right information when we need it and helping us avoid stuff that's coming up later. And at the moment, almost all of medicine is reactive. Hey, there's a problem, help me fix it, rather than let's fix the problem before it's a problem, let's stop it. So the whole of health is looking and when you compare those costs, I'm not saying anything special here, it's been said a million times. Everybody listening to this probably already knows this. The cost of fixing a problem after it's a problem rather than stopping it happening, can sometimes be an order of magnitude more. And the health system, when it is belted like it is at the moment, with the aging population and with uh higher demands in terms of healthcare, those numbers, the costs ramp up. And as a society, we need to decide how much effort are we going to put into fixing stuff and who is gonna get that effort and who's gonna get that value, you know, what diseases and which people. It's a pretty hard area to to work in and to make a decision from a social political perspective. One way to minimize having to make those decisions is to focus dollars on rather than rectification, to focus on prevention early so that there's a lower cost to society and we can do more with less. And obviously AIDS leads into that. The second part of that, and why we're having this conversation on this podcast, is because AI unlocks some of that potential. We're now able to find things for two key reasons. Sometimes we can find things that we couldn't find before. The best radiologist, maybe with a lot of time and a lot of information, might have been able to find particular things, but you're in everyday practice, you're not looking for it. So we we might be able to find things more effectively or find things we couldn't have found before, hiding in data. But we can also, with AI, potentially do it so cheaply and do it in a way that you don't need to be told to look for it. So even at the moment, you get a scan and you're told this is a potential issue. Can you check the scan to see to answer this question? And you try and answer it, and you give a great answer, you do a good job, you pass it on. I'm sure you would you tell me, Chi, you're looking for other stuff while you're there, and you know the other things to look for while you're there. But that's always got a limit. You can't possibly look for every single potential clue. Maybe we can look for a lot more with AI.

SPEAKER_02 7:11

Yeah. Look, I I think I think you're right there. There is balance. So on RCT images, there's so many images. You cover so many regions that your report could be pages and pages in length. And that that becomes information overload as well for the clinician who has to then read pages and pages to try and find a simple answer which you could have covered in the first paragraph. They're obliged to read all those pages. And sometimes people refer to certain practices or certain doctors because they know they're going to get a short report because it's their time having to read pages and pages of reports versus reading a short paragraph. So I wonder whether AI can help us not just to detect more, but to be a lot more succinct in our reports, or allow a clinician to query later on. So let's say the patient's got 50 reports which are kind of 50 pages long each, it's 2500 pages worth. Maybe AI can synthesize that such that it becomes a query. This patient has back pain and it's on the L4 nerve. Is there something in the reports previously that explains this? And then we'll do its search and come up with an answer saying, Oh yes, they've got a disbulge detector on this image that compressed this nerve. Here it is. I don't know.

SPEAKER_03 8:38

You've hit on a great point there in terms of the ability for AI to be able to improve the time that it takes a radiologist to do their reporting. You can't possibly analyze all the past records of a patient to hopefully find a tidbit of information that helps you make a slightly better diagnosis. You can't do that. I mean, you can do it, but efficiently in terms of the best way to allocate your time to create the most value, I'm sure that doesn't make sense. But what we're talking about here is AI being able to give you a little assistant for to be able to go and do that and be able to pull out all the effective information for you, summarize it and tell you if there is something in there which you should take. I think that's a hugely valuable approach for AI. You've got only a certain amount of radiologists. Their time is valuable and costly, and they can't allocate that effectively. They can't do one scan a day instead of doing a hundred. The value is in you doing a hundred to a certain level rather than one to it to an extreme level. But AI can fill that gap. What we're talking about today, though, isn't that. What we're talking about today is the proactive screening and the value in that and how that's going to help us avoid the problem of diagnosing diseases too late by pulling in that information. So can we park that one and have another chat about that another day? Because I would love to go into that more. And in fact, it's really a value out of AI, which is so little risk, but a lot of value and raises the bar on quality. Rather than everybody talking about, oh, can I trust AI? You've now got better quality in less time using AI rather than the worry of it reducing quality. Is that how you take it?

SPEAKER_02 10:21

Yeah, I think I think I agree with you there. You touched on a good point with time pressures. We were already spending too long, arguably too long on each case if we have got thousands and thousands of unread studies. And is that balance of do we spend less time on each case to provide a better result for each one? Or do we have to shorten our time even further? But hopefully not drop quality, which is where AI might help us maintain our quality but speed us up.

SPEAKER_03 10:52

Yeah. So if we leave that topic for another day, Chi, let's go back to the ability for AI to help us capture those disease signs earlier, proactive screening. So the first part is about why we diagnose disease too late and what the cost of that is, the impact. The second part's about what data there is that we could potentially harness, we could mine to be able to get that information. And as I understand it, chest x-rays, CTs, mammograms, there are already a lot of them out there. And they're taking imaging of a large part of the body. And if you're just looking for one particular problem, they could be leaving a lot of money, if you will, on the table, a lot of value on the table that's not being picked up. There's a missed opportunity for cardiovascular, lung, bone insights, etc. What's your take?

SPEAKER_02 11:47

Yeah, I agree, Steve. I think there's a lot of data that is happening on our images. You mentioned chess x-rays, CT, and mammograms just briefly, but there is a lot of data there that we just don't see as humans or we just ignore because it's just too hard, which is where I think AI can step in. So you mentioned previously CT chest screening and looking for cardx calcium scores. I think that was our last episode. That is, as you mentioned, on every single study, but we don't report it because it's uh too hard to quantify, or it's just a guess from a human. There's bones on a CT. We don't report it routinely. There's the liver abdominal organs, the liver assessment looking for fat infiltration. Again, we don't report it. Now, interestingly, the CT lung cancer screening program, which the Australian government introduced this year, it is mandatory for you to report certain things such as the calcium burden in the coronary arteries, bone density, and also the liver density to look for whether it's fat infiltration. So in that instance, I guess the government is aware that there are hidden things on these images that we are performing, but we just routinely do not report it because of a information overload. B we don't have the time to do it. Or C, we just don't see it because it requires AI to make that correlation for us.

SPEAKER_03 13:22

Yeah, we're we're seeing from the same hymn book here, mate. Um, I uh completely see this a decision that we made based on the time and and the way place that it's best managed. So that idea of a helper to be able to go and get you more information, which is at low cost rather than at the radiologist cost and even super cheap from an AI is absolutely valuable here. I think the challenge here is to do it in a way that doesn't change patient behaviour, process, and radiologist and doctor behavior too much. And obviously the the biggest one of that is that you're not prov you're not requiring new scans. First of all, that's a that's a the base level here with using existing scan data lying around to be able to identify issues. And if we can use that existing stuff, existing information, those existing scans, we don't change behaviour in the process. There's a huge door opened here just to be to us being able to add that extra information. I've used the example when we've talked previously, Chi, around the driving through to fill up your car and getting extra information as you check out, just to maybe on the bottom of the docket it says, by your way, your brakes need updating, they're down to one millimeter, your tire tread is down to two millimetres, and your oil it needs to soon soon be filled up and give you an update and then give you the option to be able to take advantage of that information automatically without you having to do anything. That's a great user experience. We need to be able to replicate that same experience from an AI perspective in radiology. And if we can, I think we could really make it make a difference in how we're helping society with identification of disease.

SPEAKER_02 15:02

Yeah, I I think there's a great analogy, Stephen. I mean, like you want to know that your oil is low because if you don't opt up, you're gonna be up for a lot more costs later on. You want to screen issues and find issues with your calf or your body a lot earlier such that it doesn't cause havoc later on.

SPEAKER_03 15:21

If you could wave a magic one chi and if AI was outrageously intelligent and you just could find any link, any pattern recognition in data, and you think about the data we've already got, the scans you've already got, the stuff that's already been positioned, if it was so insightful, what pattern is hiding in there that if AI was powerful enough to pull out, what do you think could be pulled out?

SPEAKER_02 15:46

I think we have to go back and see what makes a good screen test full stop. One has to be disease less prevalent that's causing a lot of issues. And secondly, we need to find something that we can intervene in. So I think anything cardiac related is probably a great starter. Yep. So anything to do with the heart. So we do love chest imaging, be chest x-rays, be it CT chests, sometimes MRIs crossed through the heart. If we can find heart disease a lot earlier using what we have now, with the tools we have and images we have, I think that's a good outcome for society.

SPEAKER_03 16:24

That's it that's a good point you've just brought up. You need to work back from what would you like to find rather than what can you find, what would you like to find. If I've got a tiny bit of disease that doesn't do much impact and I can identify it, that's great, but it's not really a high priority. If you have a big disease that does a lot of impact and it's very common, such as cancer or cardiac, then obviously they're high priorities. So there's no soft tissue information in an X-ray. I'm assuming there's not enough soft tissue information in an X-ray to be able to pick up cancerous stuff, is there?

SPEAKER_02 16:57

Not routinely. Depends on the x-ray itself. You can probably pick up bone changes, but to find cancer on abdomen x-ray a bit hard. Finding uh lung cancer on the chest x-ray is doable, but has to be reasonably advanced before you can pick it up, which is why when we're doing screening for lung cancer, we're using CTs, low dose CT and I x-rays.

SPEAKER_03 17:22

So if you take into account MRIs, you take into account CTs, take into account x-rays, take into account all the types of radiology tests that come across your table, your desk probably, if you could wave a magic wand, what else would you be looking at? We've talked about cardiac.

SPEAKER_02 17:36

Cancers. I think is is a strong area finding cancers while they're a bit earlier rather than later. Although that being said, that's the job of a radiologist. Radiologists should be picking up cancers on a day-to-day image rather than having to rely on AI to do it. If we're missing them, then that's that's bad on us.

SPEAKER_03 17:56

But is it surely I mean, there's a cancer that that starts before you can pick it up. There's something before a radiologist's eye can pick it up, which would be better. If you can get them at half a millimeter, it's better than getting them at four millimeters, right? Or at or at two centimetres. So can AI help there?

SPEAKER_02 18:14

I think AI can help, and I think there are some C's that show that AI in some aspects has a bad detection rate. I'm not over the convinced because what a human eye can see at this moment in time, the radiology is still running a 512 by 512 matrix. And so if know anything about computing, which I'm sure you which I know you do, Stephen. 512 by 512. 512 by 512 resolution is not a huge resolution for a CT. It's not, it's not, it's not a high quality image. I mean, how many pixels do you have on your camera? Uh a little bit more than that. I think even the phone display is a bit more than 512 by 512 on your phone. Yeah.

SPEAKER_03 18:56

No, no, it's not, it's not. But okay, so that you get that on X-ray, but surely on MRI it's a lot higher than that. No, that's a CT. MRI is even lower. Wow. Okay. But this way, MRI, surely they join MRIs together, don't they? Are you saying the whole thing is 512 by 512 and that's all you get?

SPEAKER_02 19:13

The image resolution is 512 by 512, yeah. So each slice is only 512 by 512. That's not very high at all. No. And so therefore, it's not like you see something that's microscopic that a computer can see whereas a human can't see. If you're seeing a pixel at 512, you can tell the pixel. Are there better scans possible? Look, I think you can get higher higher resolutions. The the problem is uh data overload. To get high and high resolution means more radiation. So you to get more you you pump more X-rays through, you get get higher resolution. Or with MRI, you have to keep it longer in the table on on the scan table to get higher and high resolution.

SPEAKER_03 19:59

So From your point. With the radiology scan at only 512-512, there's it's it's a lot of a guess. But if you're given context, if the AI was giving context from every scan that had ever happened before, or joining multiple scans together, now we can start to see a pattern a lot richer, and we might be able to get insights that you can't get just from a scan. Even with AI, at that resolution you just can't you you're still guessing, but given context, it starts to narrow down the likelihood of things and gives you some real high probabilities. Is that an argument? I mean that's a fair argument. Is there anything else that's a magic wand scenario that you'd look for if you could if AI could you know?

SPEAKER_02 20:45

Look, if AI could cure cancer, yeah, sure. If if it can diagnose cancer earlier, yep, love it.

SPEAKER_01 20:50

But it's a magic wand. Yeah, everyone love that magic wand. Okay, but is it possible?

SPEAKER_03 20:57

You say it's 512, 5'12, it's not possible.

SPEAKER_02 21:00

Oh no, I I'm saying 512 by 512, I don't think they can outperform a radiologist. Like if you had to see JPEG that was like thousands of pixels and you had to find two or three pixels in there that were different colored compared to the rest of it that's white, then yeah, sure, AI is gonna outperform it. But if the pixels are so big, such that an abnormality is so clear at 512 by 512 resolution, that I think it's probably gonna be equivalent to a human uh for resolution. That's what I'm saying.

SPEAKER_03 21:30

That's exactly where I was going. At 512 by 512, the ability to be able to pick up something like a cancer that can't be picked up by a human eye actually isn't that high. Like it's isn't gonna be much higher, yeah. It's not much higher. But where there is potentially value is in by adding context of analysis. We talked about being able to understand the hi context and the history of a patient and the fact that that takes a long time. If I could use AI to be able to do that and then give identify patterns across multiple scans that would just take too long and using AI insight across all of them, that could be something that would add the context that would make it able to do a job better than a human eye, even on 512 by 512.

SPEAKER_02 22:14

I think I think you've got something correct there, because a lot of things that we do find is it change over time. So what is cancer? Cancer looks normal one day and then abnormal the next day. So let's say one pixel or a few pixels have changed on its own it's it doesn't mean much. But if it's changed from something else previously that has a lot more context, a lot more significance than if than than it was saying two, three, four years back. So if that change occurs then that's significant. But if that's the same a way it looked five years ago then it's insignificant. So yes I think AI has a role in identifying or helping us identify cancers if it can pull a lot more context than what we can.

SPEAKER_03 23:03

Okay so we've identified some areas where AI can work. You've identified cardio you've identified the ability to be able to dive into history to be able to get context that we couldn't get before. We talked about osteo in the past and the access to information about bones. Anything else you want to add She about where we could be looking for proactive screening?

SPEAKER_02 23:26

I think there's richness in all the soft in all the images that we do and there's lots being missed. I think there's a lot of diseases out there and there's a lot of people having issues and there's a lot of mobility and mortality that occurs every single day and I think um with that richness we need to get a bit better with AI to identify those issues earlier to help our to help keep a healthy population and stay within the health budget.

SPEAKER_03 23:55

Okay so talking about the health budget it's a great segue to the next discussion. We can do this now. Great we've got the images we've identified that there's some heart data hiding in there we've identified where to do it who's going to pay for it. And the first thing probably to point on that is that even though it potentially can be very cheap because once you have the models you just need to scan it with an existing model and be able to identify issues potentially. But if we're going to the point of where we're looking deeper and deeper and we need a bigger model or like you've brought up before we've got a model which is taking into account a lot of data points past scans etc it could start to be more intensive process and instead of cents in the dollar for each review it could start to be a bit more and at that point who's going to pay for it even if we know it's valuable and even if we know it's worth the cost benefit is better somebody still needs to decide to do that over the current process and I'll start off by listing a few of our favourites that would be in line for this there's radiology companies you know the actual clinics there's people who are providing the hardware and the software you know and the machines there's your insurance company and then obviously in single player scenarios like in Australia with government provided medicine medical systems, the government and the health system where's the best place for this to fit?

SPEAKER_02 25:22

Great question Stephen because we all know that AI where's the benefit lie the benefit lies in patient outcomes. That's the whole goal of all what all of what we're all doing here with AI is if it does improve patient outcomes there's no point to it. So there's definitely a benefit there. But finding out who pays is a harder question. So radiology companies make money by scanning patients and each patient in Australia when they scan someone they get a little bit of Medicare rebate so the Medicare will pay them some money per case. So the more cases that you scan the more money revenue you'll receive from Medicare. So for them whether they use AI to so AI can do one of two things to help them out. If it brings in more scans there's interest there because it increases revenue if it increases accuracy such that they get sued that's for misses that offsets the costs of the software then yes that that's a benefit for them. And if you can uh report more scans then yes that's another benefit for them. The government a bit more indirect I guess is is the issue they don't want to hand out more money for AI but they will be an ultimate benefactor with more cost savings because they screen for diseases earlier. Screening earlier means earlier treatment means lower healthcare costs in the future similar for insurance companies but it's hard to link paying for screening to the final outcome.

SPEAKER_03 27:04

Health systems like government run health systems and even insurance companies when you have a long term or even a short to medium term incentive to reduce costs screening programs are used already quite extensively especially in government health systems to be able to identify problems. We were just talking about mammography previously and the screening the AI screening programs that are working there but that was a screening program that's existed already. And there's already uh in Australia there's free BMD scans for Osteo although at a later age there's a lot of screening programs which work. So the idea of a screening program being valuable despite it costing more at the start is already well entrenched. And I would have thought that would just get easier to justify with AI when it's so much cheaper. Again, if we're talking about scanning existing records which were already produced for another reason the impost is tiny compared to the communication that's required to be able to say hey would you like to come in your identify tracking your database who's the right age who's the right uh considerations who should we communicate to how do we get them into a to get a scan then we have to do the scan then we have to report the scan then we have to communicate it back we have to go through all of that infrastructure is already there when you're talking about an iterative extra piece of information that's so little I I would have thought that it's a screamingly obvious value proposition when you're just getting that extra piece of information. You're getting told about your tires and your and your oil when you're driving through filling up your car. You're already there you've already got the right information it's getting a scan for extra for no extra impost. My point then is that surely that cost is best borne by somebody with that long-term incentive like the government and isn't just hey yeah okay we'll do it at the push it's like yes we desperately want to because you hear an enormous amount of proactive preventative healthcare action and about how that's the future of healthcare because from a healthcare system we can't afford to wait. We can't be spending all this money on treating these problems for the aged at such a late level we have to get in early because the costs are so much lower at an earlier stage. I I'm probably preaching here a bit but doesn't that make sense that the government is isn't is the perfect and potentially insurance companies the perfect person to take on this they've already doing it.

SPEAKER_02 29:26

I think you've got a great point there Steven but the flip side is our three year election cycle. So you'd have to screen these patients and say it might cost you two million dollars to screen extra on your budget but your ten million dollar saving comes in year six, seven eight and so therefore if I could say to you spend two million now and I'll give you six seven eight million dollars in five years' time you'd be like yeah heck yes yeah but but go governments are the best people set up to be able to do that.

SPEAKER_03 29:59

Nobody else has got that timeframe whereas they do and I know it's hard for governments like it's hard for everybody but I know they're already like they're already doing all these screening programs. In every country we've we have with a modern healthcare system even in the states with without that publicly led health program there's still big screening programs.

SPEAKER_02 30:20

So there's massive screening programs and uh I think we are one of the last to be honest Steven I think that's one of the issues with in Australia with one of the last instituted lung care as a screening program. US has had it for many years. Yeah. Canada has had it for many years. UK have had it for many years. Only now did we start last year screening tools for radiology in South Korea it's very easy to get rebates for screening in South Korea. We still don't have one in Australia yet. So I think the government is the best place to do it. There's enough justification in a three year election cycle is the harder question. Bit more lobbying Steve bit more lobbying from yourself.

SPEAKER_03 31:03

Okay we'll try okay so who's going to pay for it? I think there's incentive there for multiple people interestingly I still think there's an incentive for a radiology practice because of the being on the cutting edge of technology being able to provide a better service and differentiate through the better service being able to brand recognition based on adding more value and being innovative and being on the cutting edge of AI technology or within technology in general. There is a value proposition there for the technology companies who are providing the platforms, the scanners the software behind it to also provide services like this because as proven by the fact that they think there is too because they've been busy buying AI technologies to integrate into their platforms and into the integrate into their systems. So I think there's interest and a value proposition for all of those areas. Chi let's move on what about how we use it in existing workflows and we need to think about that if we if a proactive AI-led proactive screen is going to work. I've spoken already about how that's essential you can't impact the process the way that I think this works is you just take an existing scan in the existing workflow. You get the opportunity for AI to review that and add information provide extra information to the radiologist who can then take that into account in their existing report and everything else stays the same. You've got the same person who pays you haven't added an extra step in terms of the the consumer's experience to the patient's experience you haven't added a step in terms of the radiologist experience. You haven't changed anything in terms of the how the PAC server works you haven't changed anything in terms of how the doctor at the end who's requested the report works they're just getting extra information. The only thing that changes is the ability to take off of the PAC server the scan, review it and put it back. What do you think? Is that important and is that how it works in your world?

SPEAKER_02 33:07

Look I think AI should be seamless and invisible almost to radiologists for acceptance we're under enough pressure time pressures at the moment extra steps aren't going to be taken favorably unless there is a very strong incentive to add it on. So for example if it was a revenue step where recommendation led to more scanning yeah that that would be taken a bit more favorably if it meant increased accuracy resulting in being too less yeah that's that'll be taken favorably. I think you've touched on some things about being seamless. In some instances it would be great if it came straight back up of the packs and that result was automatically copied into the result such the radiologist doesn't have any extra steps to do that's the kind of screening tool that would be looked on favorably by a radiologist I think. So for example Stephen you mentioned the calcin score on a CT if you're screening the CT chest I think if that number could be automatically plugged into the report at the bottom after the radiologist reported anything and there's zero input from radiologists fantastic outcome for both patients, the radiologist and for the health system.

SPEAKER_03 34:25

This brings us on to what I think is the most exciting part about this proactive screening idea is how it changes the role of radiology and the future of radiology and the practice full stop. It's less about a question answer for a traditional study and more about broader complex decision making. It's changing from reactive to proactive health management and it's broadening the whole practice of radiology to be looking for general health information. We've got so much information there I'm not just answering question I was asked I'm saying hey here is something for you here's some extra information. I've looked at your history I've looked at this scan we didn't just answer the particular question you asked we said hey this is the extra information we can find in the scan and this is the extra that you weren't even looking for. And also within the context of who you are this is the information that we were able to pull out all of that. Here's some stuff that can help guide you the practitioner who's taking care of the patient and the radiologist who's reviewing the scan to be able to help that person more. It broadens it and I think that if there's opportunities we can't even see now that that will open and it's opening the door to a broader view of radiology. What do you think G?

SPEAKER_02 35:42

I like that feature Stephen I think the role of a radiologist will change AI I suspect that in the future AI will touch on all parts of a patient journey. So imagine if it's not just a filling up the petrol it's a full tune up every time you go to the service station. So patients records are all electronically stored. They come see a radiologist the radiologist has all the imaging data AI can merge all those say oh look you've got this kind of pain it's described like this you have pain every time you move out the side let's look through the images AI has gathered all the information through all your CT studies your ultrasound in the past then it correlates with this current study goes oh look you've got bursitis these are the medications you've used in the past this is what you alluded to the recommendation for you is an injection or for you we've studied the records of all other patients in your cohort and what you respond well with or instead of injection for you your tailored suggestion will be anti-inflammatory tablets or something along those lines. So I think I like this how holistic treatment that we can offer the patients. And again is that proactive health management maybe it finds out that you're gonna have shoulder issues in about five years' time because it synthesized all that information. Previously it compared you with as of your same cohort to other patients or other people in your same age group and says you're gonna probably develop bursitis in about five ten years time recommendation is you change your lifestyle to not do too much overhead activities or when you go to gym don't do too much weights in in certain movements. So I like it like these guys saving us a lot of money.

SPEAKER_03 37:31

I'm excited you know what's interesting Key you just made me think about the fact that a lot of what you just described is traditionally the domain of your general practitioner of the clinician the physician who's taking care of the patient but it's really hard for those people to do they can't review every scan they can't review every piece of information that's come across their desk even if you're the family clinician who's been with them for 50 years that depth of information is really hard to be able to pull up. What we're doing here is at the point of time that you get this informational rich report which is a X-ray or a CT or an MRI or a BMD or or many other reports at the time you get that there's that's the point of time where you need to mine that information and provide that information and if we get the analysis of that and put it together and provide it into the existing pathways to the radiologist to be able to include in what they're finding and then pass on to the family doctor, the whatever you call it for GP here in Australia, that opens the door so much more. But it also means there's a few more questions along the pathway and I think the the vision that we're creating here we should say is probably a 10 year you know vision or or something else. But in the short term what we can do is proactively screen for specific diseases within existing scans like we've been talking about earlier today in that we say hey this this is some really basic information I can find straight away fit into the existing pathway. And that opens the door to the broadening of the whole radiology practice as a first step onto that broader vision. Does that fit with what you're thinking? That definitely fits with what I'm thinking too. Good summary all right so thank you Chi that's one of my favorite topics proactive healthcare to improve radiology practice to improve our patients' lives and to also improve general proactive preventative health healthcare really valuable. Also thank you for listening we really appreciate you coming along and hearing us out if you have questions we'd love to also get that from you so please ask Chi or me any questions you'd like create any comments you'd like to share and you can find us both on LinkedIn if you'd like. Please also join us in the future for another episode of MIAI covering all your AI technology and radiology topics and as always subscribe follow and join the conversation thank you