iCAD, Inc. (ICAD) Earnings Call Transcript & Summary

Good morning, and welcome to iCAD Technology Innovation Day. [Operator Instructions] As a reminder, this webinar is being recorded, and a replay will be made available on the iCAD website following the event. [Operator Instructions] I would now like to turn the call over to your host, Stacey Stevens, President of iCAD. Please go ahead, Stacey.

Thank you, Sarah, and good morning, everyone. I would like to welcome you to our Technology Innovation Webinar. Normally, we would host this event live in conjunction with the RSNA trade show that took place last week, but we're pleased to be coming to you virtually this year. We have a great agenda and lots of exciting updates to share with you relative to the progress that we're making in our business, with a particular focus this time on our detection business. We are also so pleased to have some esteemed guest speakers who are key thought leaders in the area of AI for breast cancer detection and breast cancer risk assessment. Risk assessment is truly an exciting growth segment for the company as we have now commercialized the world's first highly accurate 2-year breast cancer risk estimation tool that is truly going to change the landscape of breast cancer. We're going to hear a lot more about that during the program today. Our speakers for today will include Dr. Axel Gräwingholt. He is the head of the department of mammography screening in the Radiologie am Theater practice in Paderborn, Germany and also co-Chair of the European Commission Initiative of Breast Cancer Guidelines Development Group. We also have Dr. Per Hall, who is professor of the Department of Medical Epidemiology and Biostatistics at the prestigious Karolinska Institutet in Stockholm, Sweden. We also have Dr. Emily Conant, who is the Division Chief of Breast Imaging and Vice Chair of Faculty Development as well as professor of radiology at the university hospital of the University of Pennsylvania. And finally, we have Dr. Joshua Nepute, who is a radiologist at Indiana University Health Frankfurt, IU Health Arnett, IU Health White Memorial in West Lafayette, Indiana. I would now like to quickly review today's agenda before we get started. First, we're going to start the program with an update by Mike Klein on our business with a particular emphasis on our detection business and progress that we've made relative to some of our key initiatives. Then we'll move into a presentation by Dr. Gräwingholt on ProFound AI Risk. This will be followed by a live discussion between Mike Klein and Dr. Per Hall from the prestigious Karolinska Institutet reviewing the incredible work that went into developing the risk model and how it will be used clinically. We will then move on from that to an interview between Mike and Dr. Conant, discussing some of the exciting upcoming research on our risk assessment solution and the potential impact of the solution on both clinicians and especially patients. Immediately following that, we will have a presentation by Dr. Nepute on his experience with ProFound AI. And at the end of the program, we'll have a short wrap-up by Mike, followed by the Q&A. I would like to now turn the presentation over to iCAD's CEO, Mike Klein, to update you on some exciting developments in the company. Mike, over to you.

Thank you, Stacey. Sorry about that. Can everybody hear me all right?

Yes.

All right. Perfect. Okay. Well, we do -- I want to thank everybody for joining us. Monday, we have a really great panel today of distinguished guests who are leaders in the field. I'm out here in Silicon Valley, where we are sheltered in place. You can be sure this is all I'll be doing for the next 30 days, at least in terms of not being out and about, but it's a very busy time, and there's a lot to talk about, and we're excited to jump into it today. So what I would like to do is start by giving us a bit of an overview of the business. This is a slide I've used before, it showed the 2 sides of our business. We have what we call our discovery side or AI side, which will be our focus today. And there's our recovery side, our therapy business, which was at our last Innovation Day, an area of focus. We're very excited today to talk about how we've now achieved 1,000 installs of ProFound AI. And we're going to talk a little bit about how we got there. But also, our focus is going to be how the next 1,000 or 2,000 or 3,000 will unfold, which we see occurring at a faster pace than the first 1,000. In fact, we did the first 1,000 a little bit less than 23 months, and the first 1,000 is always the hardest. And we'll be sharing reasons as to why we believe that we're hitting a very exciting part of the business where the growth will continue, and perhaps even do so at a somewhat accelerated pace. So with that, I'm going to dive into the left side. As you can see, we have already 7,200 installs, both 2D and 3D, and we're building off a strong foundation. And today, we're going to be talking not just about ProFound AI, but also about our exciting new risk offering. With that, I will move into our next slide. You'll see a slide here that I have used before. This shows the movement from film to 2D to 3D tomosynthesis. And what we're talking about today is a little bit on 3D tomo, which is the first 1,000 installs, but we're going to be moving into what we think is an exciting future, and that is the movement into the -- into risk, as you can see here on this slide. Now there's the question I have for folks and that is, how do you be a cancer detection wizard and never miss? Anybody got any ideas? You could put them in the chat box. And the best way to do that is simply just to mark everything a cancer and you'll catch every single one. You'll always catch cancer. However, you'll be wrong 994 out of 1,000 times as there are only 6 cancers per 1,000. Some may suggest that it's as high as 7, some with 2D would say it can be only 5, but 6 is the number. And this is the challenge, marking everything a cancer, you'll be wrong 994 times. So the magic is to not only mark the cancers but to not mark anything that is not. And this is what AI does. It separates the two. It's sort of a pattern recognition, separate signal from the noise. So if you aren't accurate in detecting those cancers, you can have high sensitivity. But as you can see, you can have high sensitivity by just marking everything a cancer. The goal is to have a high sensitivity, which you see on the chart on the right, which is the vertical axis to move up that chart without going too far out to the right. Going too far out to the right would mean that you're -- the price you're paying for high sensitivity would be false positives or what we call lowering your sense -- lowering your specificity. So the better you are at not mislabeling something a cancer, the lower your recall. So the further out to the right you are, you're more likely to recall a patient. So the goal would be, the ideal radiologist would be up and to the left. And that's the point about specificity that if you are able to just call those ones that are cancer, you'd have better specificity, and the true magic is to get both. So this is what we call the daily double, sensitivity and specificity. And if you get both of those, you'll get what we call a high score on what's called the area under the curve, which is interestingly enough, the movement from the area under the curve till upward and to the left. So area under the curve is something I must refer to as the miles per gallon in our industry. This is sort of a key metric. So with that, imagine doing all of this and at the same time, reducing reading time. Well, that's what we've been able to do with ProFound. And this is what has resulted in what we consider a very rare trifecta. It's hard to go up and go to the left, and at the same time, reduce your reading times. So this is what we've done, 8% sensitivity, 7% fewer callbacks, and you can see the reading time reduction over 50%. Now we're going to do this, we think, in an even more enhanced way because we're about to submit our next-generation ProFound. In fact, I think it was submitted last week. So in early 2021, we'll actually be improving, we believe we'll be improving these numbers. It all depends on the regulatory response, but at least what we've submitted is a significant improvement in these numbers. So look for that in early 2021. So with that, again, I'm repeating the -- what we call the trifecta 8% sensitivity, 7% fewer recalls and 52.7% reduction in reading time. Moving ahead on this, a little bit trouble here with this slide here. Moving ahead on this, you also can see the net effect here is the AUC, which is scored here at 5.7. That's the net effect, again, of the sensitivity and the lower recalls, which is also referred to as higher specificity. So all of this, you can see highlighted, again, we really spent some time on this because we're at the point in the cycle where education is critical in terms of patient -- rather, physician adoption. So we spend a lot of time on reviewing what these numbers can mean for clinicians. Now part of the reason why I mentioned that sensitivity is not only important but also specificity and reading time is that here's something I've shown before. This is -- one of our top clinicians use this that has -- let me go back on that a little bit. One of our top clinicians here that has -- on 2D, we're seeing 80,000 images in a year. On 3D, digital breast tomosynthesis, as you can see, they went from 4 images per study to 280, seeing 28,000 images in a day, 5.6 million in a year. And literally, they were seeing 3 -- they were seeing, in 3 days, they were seeing more images with 3D tomosynthesis than they were seeing in an entire year of 2D. So this is one of those images. This is one of those 5 million to 6 million images. Now try to find the soccer ball among the panda bears. Some of you have seen this before. So you might be able to find it this time. But if you look hard, you might be able to find it towards the right side, and if you look really carefully, you'll see it there. And if you want me to go back again and unfold the paper, see if you can find it, now that you know where it is. And imagine doing this every day, 80 to 100 cancers a day, images like this, trying to find the cancer. It's really helpful to have some capability that helps you not only find it, but dismiss the ones that aren't cancers and to do it with lower reading time. So here, you see some of our results showing up. This is a recall rate that in addition to having improvements in sensitivity, we're able to lower the recall. And again, I think of this as the trifecta or think of it as a 3-legged stool. You need all 3 legs for the stool to properly be able to sit someone. So that's key to us. And the reason I highlight recall is that it's an often underappreciated aspect of the trifecta. There's all kinds of articles out about the cost of these recalls. Here's just one, and you can find this on the website just by googling the title and find it on the Internet that these recalls cause misdiagnoses and its estimated cost of over $4 billion a year. So as we -- and that's that recall right there. So as we move forward and people have asked, well, how do you compare to other vendors because there is now another vendor in the market, we feel the need to -- and again, this isn't -- you heard the Daniel Moynihan expression that you're entitled to your opinion, but not your own facts. This is the opposite. This is -- we're just presenting the facts here of what's observed on people's websites and in press releases. So this is a recent player in the market, and you can see their recall rate. Although sensitivity has gone up, it has come at the expense of recall. So it's moved to the right. And you can see the numbers below. The amount of recalls, and you can see that, that represents a 10% increase on a relative basis. Again, it's like the 3 legs of the stool. And the other aspect we talked about in addition to sensitivity and in addition to the specificity of false positives is reading time because these images are instead of 4 images as 2D was, there are 300 images. We generated not only the movement above the line, you see all these dots moving above the line. You see the size of the circle is smaller, so we're generating almost a 53% reduction in reading time. And again, for those who've asked what else is out there, the new entrant in the market, you'll see in a moment, we'll present its score. But I did want to say that the reason this reading time is so important, it's an essential part of our value proposition, is that radiologists can cost $200 to $250 an hour. If you're able to reduce the reading time by 50%, which is what we do in our study, and we've heard even more as people have gotten used to the technology. You're saving about $200,000 a year in radiology cost or $125 an hour. That provides on a $30,000 purchase -- you can see how that provides a very quick breakeven as little as 8 to 12 weeks. So here is the results with what we see with, let's say, another vendor on the market There you see ours showing up right here, showing up that's the reduction of reading time, 53%. And that is, again, 53% relative decrease. And there is the other vendor entry, a 12% increase in reading time. Now I mentioned that sort of the granddaddy of all the scores or grandmama of all the scores is the AUC. This is kind of like what's the bottom line? What's the miles per gallon when you put it all together? Because this is, in some ways, I like to think of this as your sensitivity sort of subtracting your false positives, what might you come up with? What's the net effect on moving the dots above the line and to the left. And there's ours at 5.7, and this is a competitive entry, almost 50% lower than ours. And I say this because we've been out here for 2 years doing this. And tomos launched 5 years ago. So we've got a significant lead out there, and we intend to press this lead with these numbers increasing, ideally, as we move into our next release, as I said in early 2021. You can see the annotations at the bottom and you can find these things on press releases and websites for citations. So the area under the curve, AUC, is a critical dimension of all this. So here you see again summarized and my point at the top is that our next 1,000 to 10,000 installs will be based on creating the right lexicon for a constituent. So everybody understands the difference between sensitivity and specificity, false positives and sensitivity because it's easy to just think about the sensitivity without looking at the total combined and of course, the miles per gallon. You may pay less at the pump. But if you can only go half the distance, that is something to pay attention to, and this is what we're spending a lot of time educating folks on. So we really believe that with AI, the real competition is the ill informed or the under-informed, and we know what the antidote is, it's education, awareness building, working with advocacy groups, societies, patient engagement and also collaboration among the health care teams. So with that, I want to say that we have -- everything we've been talking about so far is 3D tomo. We are moving towards the next round. And I need to throw that away, but we're moving -- because it's still a big part of our business, but now we're looking at risk assessment analytics. So you can see in the technology adoption curve, while we may be crossing the chasm to an early, early part of the majority with the 1,000 installs, we're just getting started with risk, and it's very exciting, and it's as big as the ProFound market. In fact, you're going to see, actually, it's twice as big because we provide this product not only on 3D, but also on 2D. So here it is, and it's based on Karolinska data, we're going to have the principal investigator, the research of -- lead researcher at Karolinska, Dr. Per Hall, who will be talking about it. The findings you'll be seeing today are based on findings just found in the mammogram. But we've also, over the 8 years, did track all these risk factors for 70,000 women over 8 years. It's a lot of images. And in the future, we'll be able to do a lot more with that data. And as we take a look at the results, and I'm not going to steal the thunder from some of our other presenters, other than to say that in this slide, you can see that when women are stratified into percentages of where they fall with the low-risk being 4%, general 60%, moderate -- you'll see in the slide here that, that 9% of patients are over here in this high category, this would correlate to a 7.5x greater likelihood than the general population. That's 750%. And even the moderate category is 2.5%, almost 2.5%. So you can see that while the average may be 5, 6 per 1,000. Over here, far higher than 5, 6 per 1,000. And here, dramatically higher than 5, 6 per 1,000. This is what's in the radiology paper peer review that showed up late this summer. We'll be going into more detail on this. What we have done is taking a look at the prior model, which is pretty much been what I call the black and white model, which is you're either average risk or high risk, at 90% or 85%, you're going to fall into the average category. I call that, I'll see you in a year; or the supplemental screening category, 10%, 15%, I'll see you in a week or 2 up in MRI. So I'll see you in a year or I'll see you in a week or 2. That's pretty much where we're at in many of the cases right now. But what we're doing now with more information is we're at the era now we are introducing -- open the curtains, this whole new paradigm where we move from this black and white binary world to a full spectrum color -- and this is a continuum, even though there are categories, it is a continuum. And here, you see the 9% -- remember, this group here had 750% higher likelihood than the general population. And the group at the bottom had only 4%, a -- perhaps a 50%. And that might suggest that some of these patients may be needed to be seen more frequently, and in certain cases, perhaps less frequently. So let's sort of dive in a little bit further into this. We have prioritized -- we have productized this. It's one thing to work, and we've worked beautifully for many years with the researchers at Karolinska. But now we've turned this into a product. And this is the product we launched. We launched it on 2D in the fourth quarter. And there's some great success stories we have to talk about today. And we also will be launching it in early 2021 on 3D, so both 2D and 3D. So -- and what does this product look like? When a woman gets a score, it's going -- we're going to be combining on a scorecard, her case score today. This is your probability today. You're certainly scored today out of 100%. So if somebody looked at this, they like think, okay, well, probably won't call back the patient or won't need to do supplemental screening. The density is B. So that's on the lower side. They're risk score, and this is the general population. It is actually, in this case, somewhat below the average. This is the person with relatively same score, but they have a high future probability. So this is the difference between 2 women that may get exactly the scores today, look at the difference when we introduce risk. So this is what we're doing is giving this full card of capability to women. So when you pool this all together, what this does is it shows a woman on an age-adjusted basis, whether her scores are lower than average or in certain cases, you may even have the same score. But you -- if you're a few years younger, age-adjusted, that score may be of concern. So suddenly, you can see how you compare to other women of similar age groups. This is what's exciting about this work. And again, we're going to be talking to the folks who really helped develop this with us today and are also implementing it. So here is a case study. I always like to break this down to a personal level. This is an actual image of a woman in 2014. Of course, we read it with our new tools afterwards. But in 2014, this woman showed up, 67-year-old woman. She had -- I'm sorry, going back the other way. Well, I just wanted to show the fancy -- wanted to show the animations again so that gave me the opportunity. But this case study of this woman is a woman that we did not see any cancer in this case. This was something we didn't pick up. We've shown it to dozens of radiologists, didn't pick up, and it wasn't even picked up on our AI tool. So what happened with this woman? Well, when we introduce the scores, the look-ahead scores. This is the risk score, suddenly this woman is scoring in the moderate-to-high category. So she's scoring in the moderate-to-high category. So this gives a little more information than one would have had in 2014. What happened with this woman when she came back? Well, this woman, if you take a look at that score, what it translates into if you correlate it with the published Karolinska data, the one that was in Radiology, that correlates to a score that's actually 4 to 5x greater than the general population, 400% to 500% greater. So if you had known that at the time, you would have actually done a little more work with this woman because in this case, this woman did indeed present in 2016 with a cancer and you could see -- you can see how we outline it. You can see the probabilities of each lesion marked and there's an overall case score. So we mark all of them. And you can see quite clearly with the eyes over here. Now I want to make a point here about this tool because in reality, one of the questions asked is, was this really just -- did it just show up over these 2 years? Or is it possible that it was actually in this image? Well, here is the thing that we find with artificial intelligence and with this risk tool. There are microscopic, there are some visual elements that may exist in the breast image on the left. But because they can't be seen and they may fall below a threshold of clinical utility because they can't be seen, it's hard to biopsy what you can't see, we don't mark them. But that doesn't mean that when given the luxury, you might say, of projecting out 2 years ahead, we can say that, that's a very high risk. So things that are in the subvisual range that aren't going to be marked today, that doesn't mean that there necessarily isn't a problem that will emerge in 2 years because the computer can see more. It can pull this image apart, pixel by pixel, voxel by voxel. And that's what we're seeing here. So what does this all mean? Well, today, with traditional screening, you fall into 2 categories in most part, average risk or high risk. And you're going to -- again, this is what I call see you in a year or see you in -- or if you're high risk, see you in a week or 2 up in MRI or with an ultrasound, maybe even a biopsy if warranted. But with these tools, this black and white reality, again, will give way to be able to see women on this full color spectrum, and each women can have a more personalized analysis. So the discussion changes, and I think you're going to hear Dr. Conant talk about this a lot. How does this change the dialogue with the patient because if you're a low-risk, do you need to lengthen the screening interval potentially? Or should you consider that? If you're a general, should you just maintain? What if you're on the low side to general? And furthermore, if you are in the moderate category, which translated into 2.5x greater risk, well, maybe you shouldn't be seeing it just 12 months, maybe a little bit sooner and maybe even on an additional modality with supplemental screening. But these will be all left up to the clinician. And of course, if you're in the high-risk category, you may want to more immediately avail yourself to some supplemental screening. So again, this is a case, going back to my original point, perhaps not seeing, maybe believing when you have software that's pulling it apart, pixel by pixel, voxel by voxel. Is not seeing for us believing? Well, perhaps with these tools, and maybe it's even like the flight instruments in an airplane. You can't see in the fog so you have to align your instruments. So this trust, this relationship with technology is something we will see evolving in the years to come. I'm sure our phenomenal experts today presenting will talk about this dynamic, sort of the human technology interface, but then interfaced with patients and clinicians as well. So these are the choices. These are the choices that could be availed. So it's not just about mammography, it may be once you know what your personalized screening regimen may be, these all may come into play. So it may be now about the right product, the right technology and the right time can be optimized. And all of the new criteria for reimbursement are going to be tied to indications for use or what's called authorized use criteria, another AUC acronym. But we provide the information to allow clinicians to perhaps better -- to use the tools to make informed decisions, not only based on what they see today, but what may be presented as a possibility for the future. So again, this is how it changes. Now I always ask myself the question, are we really finding interval cancers? Or are we really just finding things that are already there? This is about my earlier point. Well, I go back to that 10-year-old kid who used to wonder why there are only 4 to 6 moons of Saturn. And then as an adult, suddenly see there are 100. And last year, we found 40 more. So did they just grow those moons? Or were they always there, but we just didn't know where to look? Well, we think that, that -- not in all cases, but in many cases, those moons are there, those cancers are there. But some of those cases, they may be interval cancers. The key thing is to find those ones where there are moons and find them early. So with that, since this is an investor-focused presentation, I'm just going to spend a moment, and we'll be going over these more on separate discussions, but this is the way we've previously looked at our market with the first 1,000 installations. We had 10,000 mammo units, 3D mammo units. We were selling at $30,000 plus a $3,000 service contract. We saw that as our market, and we can only move to a larger market as the 3D market continued to consume the 2D market. But what we're finding now is that with risk, 2D and 3D, the next 1,000 ProFound installs will have a far bigger total available market. We've gone from 10,000 to 40,000. Why? because we can now install on both 3D, this first small circle, and 2D as well as upgrade the software for both. So the way the market would look today is 10,000 installed units, add another 30,000 2D because that's what we'll be able to have with both 2D AI and risk that gives us 40,000. And if we use our AI ASP, that's 40,000 times $30,000, that's a $1.2 million market today, not down the road, and that is when you add risk, another $30,000, the total available market is 40,000 times $60,000. That's $30,000 for our AI -- ProFound AI and $30,000 for risk. So that gives us a total available market of a very substantial $2.4 billion. Now we're not saying all of that is going to be available today is what we call the served market, but we're going after a much bigger market than just 10,000 3D installed units. I'm just going to show you this chart of how this looks. If you take a look at that $2.7 billion that I just showed you and look at the license, which is basically buying a singular license, perpetual license, the market there, if you assume it grows at the same rate as our 2D did or even that digital mammography did, it's a 7-year cycle. So that would mean that of that $2.4 million, that $343 million is available per year. But if you -- or a total of $2.4 billion. If you use the subscription model, which is where you actually get paid each year, you might actually pay a little bit less in the first year because they're only going to pay, instead of a $30,000 license, they're going to pay $12,000 subscription annually. But look at what happens, by year 3, those annual subscriptions now begin to crest and exceed the licenses, plus you get more of the market playing because there's no capital equipment cycle. It's in the cloud. You buy it as an expense. And we're using up 40% of the annual license price at basically 2.5 years, not even 3 years, you are breaking even. So this is a model that gets the whole market playing regardless of size, some may not want a $30,000 license for a small site, but a $12,000 a year subscription does it. And it's also recurring model. It's a model that also is very profitable. And we'll see here, this is now the cumulative market. This is adding up what I did on the prior page here. This is adding up all of those amounts yearly to see how it shows cumulative. And you can see, again, the subscription market exceeds $3 billion, in fact, goes to $3.3 billion. If you look at years 8 and 9, we were well over $4 billion, double the size of the market of $2.4 billion, which today has been a $330 million market just tied to 3D. So it's frictionless to go in the -- with the subscription market, ease of product extensions. And again, it's an expense item. And that market really moves. So kind of showing it all together, 2D and 3D plus ProFound, the TAM doubles to $2.4 billion and a subscription or SaaS model moves us towards -- we're now talking about no service. You don't need service people to go in and install it, it grows to over $3.3 billion, again, recurring and predictable and dollars grow faster annually and cumulatively than a onetime license model. So you can see where we're heading. And just real quickly, we're not being quiet about these tools. This is a tool that women want to know about. We are already getting calls because of Breast Cancer Awareness Day about this product. And we've got installations in Europe and in the U.S., and we expect this to move even faster than our AI product. We're hitting all the news media as you can see here, all these different sources, cover of magazines, big features in Breast Cancer Awareness Month. And I don't have time to show a video here, but this will be placed on our website. And the one thing that is interesting is that we've even had sites say to us, so women say, I'm even willing to pay. I had a woman we spoke to 2 weeks ago that said -- because we are interviewing, we've interviewed now almost 200 women. She said, I'm not going to pay more than $27.50 for this. And by the way, this study shows that 87% of women would want to pay for this. Now that's not our model. This is at the site level, if sites wanted to charge. And I'm not saying that's where we're going. And that would be controversial. But this woman said, I'd be willing to pay $27.50. And the question was, what's there about $30 that you don't want to go beyond? And you said, no I meant $2,750 because that's what she's paying for a concierge medicine. So we're basically rolling these tools out. We're selling to our clinicians. Our clinicians will find a way to present these to their patients, and I just show this as an interesting little factoid in that the demand here at the women's level is very high, it may even eclipse the initial clinical demand because people worry about acquisition, costs, whatever. But again, you can see the demand is there and even on a pay-per-click basis or a subscription basis, it will be approachable for all. So here are all the tools wrapped up together. This is the scorecard. This is how we're moving from black and white to sort of a multi-spectrum set of capabilities. And we have new partners like Change Healthcare, which is a dominant PACS player out there that allows us to move more swiftly into this SaaS-based model. And in the future, this is coming attractions. We do believe, while this isn't what's in our current risk tool because our current risk tool, as you'll hear, only relies on the mammographic features. That's all we need. We believe that in the future, we can do even more precise work with the 27 risk factors. So stay tuned for that. We see an exciting future coming our way. And with that, I will end a little bit long-winded, I apologize, but hopefully setting up the rest of the day. And I can now turn the presentation back over to Stacey.

Thank you, Mike, for that great update. Just a ton of information in that presentation. Great information that we'll be breaking down and reviewing in greater detail in future investor events. Okay. And moving along in the agenda, I'd like to now introduce Dr. Axel Gräwingholt, who is the Head of the Department of Mammography Screening in the Radiologie am Theater practice in Paderborn, Germany. He has 28 years of experience in breast imaging, and he actually reads 10,000 to 15,000 mammograms per year, also has 10 years of tomosynthesis experience. Dr. Gräwingholt is certainly one of the early pioneers in the use of AI and breast imaging, and we're excited to hear from him. So welcome Dr. Gräwingholt. I will now turn it over to you.

Thank you, Stacey, for that nice introduction. I'm not seeing my slides now. Okay. Well, I would like to present to you a little bit about the European screening and diagnostics as we have them at present. We have actually 3 arms of breast imaging in Europe at the moment. First of all, the organized population-based screening programs, large programs of just about every European country, especially in Western Europe. It is based either on biannual or triannual intervals in the U.K. and [ more toward ] due to financial resources, they have to switch to triannual. Basically, age groups between 50 and 69 are screened. Some countries start at 45 and some go until 75. All European screening programs are based on 2D mammography, rarely adding ultrasound in dense breast tissue. That's what they do in Austria in some cases and in some Swiss cantons. This is due to the insecurity of density measurement by humans. That's the reason why most other European countries do not screen according to breast tissue with supplemental screening. All European screening programs are according to European guidelines' current recommendations. And then the second way to be screened would be the so-called opportunistic screening for women, which is not an invitation, and it is for women who are not in the age group or who are, for some reason, not willing to participate in screening program. Or if in a country, a screening program does not exist or a national screening program does not exist, they have some regional screening programs, which cannot really be called organized screening programs. And the third way to have a breast imaging would be a diagnostic mammography due to symptoms. Next, please. Well, talking about personalized screening in Europe, we have a very easy approach. Although we certainly do know that there are women at different risk, we do not have any risk stratification. If they have low risk, general risk, moderate risk or high risk, if we could assess that, we say one size fits all. We have one amount of money to spend, and we spend it evenly on every woman in the age group from 50 to 69. And that would mean 2D mammography every 2 or 3 years. Recalls will only take place if lesions are found, and you have very few exceptions on that, in Austria and some Swiss cantons where ultrasound is added in dense breast tissue. I would like to show you now in a case how risk could change the screening in Europe and what would the benefit be for these -- for women to have risk stratification. Next please. This is the case where we retrospectively applied ProFound AI 2D and risk because, of course, it had to be retrospectively done because in 2011, nobody even thought about an AI algorithm like that. Now patient history is that a woman was diagnosed within the screening program at the age of 63. She started participating in screening only at the age 58. The density of the breast, we would consider density C, which is one of the four categories of density. And you might stumble a little bit about the interval, say in 2011, 2012 and 2015. This was actually a 2 year or biannual screening. It was January 2011, December 2012 and January 2015, which means the interval of 2 years here is considered between 22 months and 26 months. So it is a biannual screen. Now in 2011, there was a nonactionable small cluster of calcs in the left breast, which was seen best in the CC view. It was dismissed by 2 radiologists and ProFound AI retrospectively applied, gave a case score of 24%, which is not very high. There were no real marks on that. Risk score was moderate. Low point is 65%. So the woman was invited back after 2 years. In 2012, we had a subtle increase in nonactionable small cluster of in the left breast. It was seen in both views this time. It was again dismissed by 2 radiologists either because they didn't see it or maybe they thought it was a benign lesion. ProFound AI case score went up to 38%. Of course, again, retrospectively applied. The radiologists at that time didn't have these scores. The lesion score at the place where later the tumor occurred was 29%, and the ProFound AI risk score went too high, 1.61%. Now finally, in 2015, we saw a mass with calcifications in the middle of the left breast. ProFound AI case score was 80%, so very high. ProFound AI risk score was very, very high, sky high, I would say. And we did find cancer using ultrasound guided biopsy and pathology revealed a Grade II invasive ductal cancer and a Grade II ductal cancer in situ associated with that, which is a precancer lesion. We did all these mammographies on a GE gantry. Next, please. Now looking at the CC view of all 3 screening rounds. There is a small or subtle change to be seen in the density of the breast tissue in the middle of the left breast. And if you -- next, please, would turn on ProFound AI 2D detection, you would see, there are small marks with an increasing lesion score from 19% in 2011, over 23% in 2012, to 62% in 2015. Next, please. And you can actually see the same on the second view that is always done. And if you apply the markers. Next, please. You wouldn't even see 1 marker in 2011. In 2012, you see 2 markers. One is false positive. That's the one with 25%, but that 29% one is a marker in the lesion where later the tumor occurred. And in 2015, you see 2 markers, 1 of them has a very high probability of being a cancer with 68%, and the other one has a lower probability with 21%. Now next, please. When you look at risk, when we applied risk, in 2011, you can see the woman had a moderate risk, moderate risk to develop a cancer within the 2 years. That's pretty much average of a woman of that age, as you can see in that graph on the right side. If you go one step further, you can see the very, very subtle calcifications in the center of these images. Next, please. And the same, you can see in the oblique view, which is the second one. Again, here, you can see these calcs much better. But still, there is no radiologists to blame, not to have seen that. We go to the 2012 now, next. You can see the risk went high. Much above the age average is about twofold of an increase of a risk. And again, if we go into the detailed images. Next. You can see the calcifications have increased. And in the [ OBL ] view, and the second one, next, please -- you can see that even better. Now going to the next image, obviously, that was the time of diagnosis. You see a very, very high risk, which is obvious because she already had a cancer at that time. And if you go into the detailed images, you see a lot of calcifications. They increased tremendously and even better seen in the next. In the [ OBL ] images, these are polymorphous calcifications, which are almost pathognomonic for a cancer to be there. Next, please. So what did risk give us? As I said, before we did apply that retrospectively, so we couldn't take any actions by that time. But what would the consequences have been for a woman identified with such a risk and such a history of risk. In 2011, the woman have a moderate risk, which is typical for the age group actually. In 2012, she had a twofold increased risk over the age group. And of course, 2015, she had the cancer. Next, please. What would all of this mean now if we are speaking about European screening programs? We're talking a lot in Europe about the future of personalized screening, but at present, we didn't have -- or until now, we did not have an idea of how to change the screening to a better way. Most of the risk models that were calculated like Gail or Tyrer-Cuzick's are not very accurate. So we cannot screen based on them. Also, it's quite complicated to get all the information that you need. But this -- with ProFound AI risk, it is possible to get this information from the images you already have in the normal screening. If we have ProFound AI risk stratification, we can categorize into women with a low risk, a general risk, a moderate risk and a high risk. And next, please. What would that mean? That would mean you can discuss lengthening screening intervals based on preference for the low-risk group, which saves a lot of money. For the general risk group, you could just maintain the regular screening interval, which is, in Europe, 2 years. For the moderate ones, moderate risk women, you could discuss shortening screening intervals and maybe supplemental screening. And for the high-risk women, you can discuss shortening screening intervals, do supplemental imaging and also possible risk-reduction strategies as there are some approaches with tamoxifen therapy, for example. Now in our case, if in 2011, we would have had risk as a product and we would have measured the risk as moderate. We could have said, okay, for this woman, it would be good to come back in 1 year. Maybe we would have detected the cancer earlier. Or we could have added, for example, MRI to the screening, and we would've -- or we might have had detected the cancer already. But certainly, in 2012, the high risk, if we would have added supplemental screening at that time, based on ProFound AI risk, we would have found the cancer by either ultrasound or MRI. This is the way I see personalized screening or screening in Europe in the future because it makes us spend the correct amount of money on the correct group of women for a better screening and a better overall outcome. Thank you very much for listening, and I pass on to the next speaker.

Great. Thank you so much, Dr. Gräwingholt for that compelling presentation on where you see the clinical value of ProFound AI risk. It's always so great to hear about the impact this solution is having in your practice and especially on your patients and how it could change screening in Europe, particularly where the screening paradigm is every 2 years. So thank you so much for the great presentation. Okay. We're going to move along now to the next section of the agenda where we're going to be featuring a live interview format between Mike Klein and Dr. Per Hall. Dr. Hall is a medical oncologist by training. He works with the Department of Medical Epidemiology and Biostatistics at the Karolinska Institute in Stockholm, Sweden. He is also an honorary consultant physician at the Department of Oncology at the largest hospital in Sweden. His research interest is focused on breast cancer screening and prevention. And over the last 10 years, he has established the Karma cohort, which is a prospective screening cohort of approximately 70,000 women. Welcome, Dr. Hall. I will now turn it over to you and Mike.

Thank you.

Good morning, Dr. Hall. Can you hear me all the way over in Sweden?

Absolutely so.

Okay. Excellent. Well, Dr. Hall, I'm going to start by saying that there's an old expression that success has many fathers, but failure is an orphan. But this is an area, in terms of this research, where you can legitimately claim to be the father or one of a limited few that have moved forward with this breakthrough, what I would consider a breakthrough effort in publication radiology. And what I wanted to ask you Dr. Hall is what -- I can always like to get into the personal side of this first. What was it that motivated you? I mean, you've had a long-distinguished career, but what was it in particular that motivated you to focus on breast cancer, and in particular, breast cancer risk moving forward?

Yes. Thank you for these nice words. Maybe we should correct a little bit. I -- we are not the first to do a risk assessment of breast cancer, but we have adopted new ways, as you pointed out in your presentation. Your presentation, focusing on the mammograms and mammographic features that we will probably touch upon later on. But yes. So why -- how did it start? So I started working at the Department of Oncology at the Karolinska Hospital many years ago as a young doctor, you deal with a variety of cancers, but I ended up working with breast cancer was important. And at the same time, I was a little frustrated because all oncologists working with cancer realized that we have a large number of patients coming in with quite late-stage disease. And I was -- early started talking about, how can we sit around and just wait for patients coming in? Shouldn't we identify those with a high risk and do something about it as the cardiologists are doing. Those days, my colleagues, they shut their heads and said, you're not talking about patients, are you? No, I'm talking -- talk about healthy individuals, and they didn't really -- I don't think they were particularly interested in the concept. Anyway, what I started thinking about was that in the U.S., you have something called the Framingham cohorts, '50s and '60s, that cohort was generated in the small village or city of Framingham, New England. And these 5,000, 6,000 individuals were followed and followed and are still followed. And from that little cohort, we learned about cholesterol being unhealthy, high blood pressure. And we now know fairly well what increases or influences the risk of a cardiovascular event. But we didn't know anything about breast cancer or very little. So 10, 15 years ago, I started, well, trying to get funded for establishing my own Framingham cohort. And as was just mentioned, we were able to establish the Karma cohort in 2008, 2009, 2010 after donation, a large donation from a Swedish family. So we have 70,000 women, where we collect an enormous number of information, everything from DNA to mammograms. We collect 1,000 mammograms per day to add up our -- a little bit more than 3 million and use them for risk prediction. So that's the background of it all.

So Dr. Hall, what was it, in particular, about about the 2-year risk versus lifetime risk or 10-year risk, some of those that were out for web. What was there, specific in your examination of the 2-year risk that intrigued you?

Yes. As the [ 3-year ] radiologist on board today, -- no, that we are -- even the most skilled radiologists miss 20%, 30% of all cancers at screening. So these are the so-called interval cancers. They appear in the interval between 2 screening -- 2 screens, which is extremely frustrating. It doesn't mean that radiologists are bad in any way. It's sometimes extremely hard to identify cancer as Dr. Gräwingholt just showed. It'd also be that some of these cancers are fast growing. But we sort of -- we said that, okay, let's -- instead of [ identifying ] a lifetime risk, let's develop a risk model that identifies the women that are missed. So we could identify those women who were sent home with a negative mammogram but shouldn't have been sent home. They should have a supplemental screening. If it's ultrasound, contrast and has mammography, MRI, whatever. But they shouldn't have been sent home. So that was what we were doing and using this large Karma cohort of 70,000 women. The simple question, who were the women that are diagnosed in breast cancer be in 2 years of a negative screen? That was our goal to find that.

Yes. That's -- I like to hear in the back about -- I'm curious when you -- with all the data that you collected over the 8 years, did you expect that at least for this initial launch of product that the mammographic features in and of themselves that, combined with prep stage, would get the kind of yield and produce the results that you've seen.

No. No, I promise you, I didn't, and I was heavily involved in genetics, and I was coordinating a large global initiative, European Commission funded. We were 170 groups from all over the world, and we identified now from 2007, now we're at 300, 400 genetic determinants of breast cancer. I was so convinced that genetics would be the most important factor. But then we realize that in the mammograms, so much of the features of the mammograms, the density, the calcifications, the masses, the masses because it has more dense areas of the density. That reflected the lifestyle factors. That reflected whether you had many kids who are breast feeding. That reflected whether you were using hormone replacement therapy, that reflected whether you have certain genetic determinants. So it actually boiled down to our surprise that we -- to use the mammograms was not enough, but you came a long way just using the mammograms. And as you said, it's -- this makes life much easier because it's not that you're going to ask women a lot of questions, not going to insert that information into a database and to an algorithm. You're not going to draw blood, you're not going to do the genetic, the genotyping. So yes, short answer to your question, I was really surprised. I was definitely really surprised.

Yes. I want to follow that with the following question. It's based on the fact that, since we have a lot of investors on the call, everybody knows or has invested in a company that is, in some ways, a great solution looking for a problem. But now that we have this 2-year risk model, how do you envision that being used clinically? And I recognize Dr. Conant is going to get into that as well. But what -- okay. So the answer is like now that we have it, what do we do with it? So I'm like, you have a great power -- with great information comes great responsibility. So what is our responsibility?

You're asking me to answer this question when our world-leading radiologists are on board at this meeting. Okay, I'll do my best not to mess up. But I look at it -- seriously, I look at it as a decision support, too. So I had a discussion the other week with the Head of Radiology and Mammography at South General Hospital, which is the biggest hospital in Sweden and she said, I want this tool now because I'm employing so many young doctors, so many young radiologists. And I can't tutor them all. We seniors can't do that. So I would like to have this. I would like to start up by using the ProFound AI risk tool as a support for the younger ones. And then I want it, at the same time, want to see how it works. I want to incorporate it in my decision, in my detection support tool. So that's one thing. But I also think that another way is, Axel Gräwingholt beautifully showed, how -- when we talk about risk prediction, we talk about 1 assessment. And we said, that gives us the full picture, which is wrong. Mr. Axel showed that risk changes over time, of course. And some people, they stay low, they go even lower. Some people, they go up to sort of medium level. And some women, they actually change. And this is -- even in the group of people who do nothing else but risk prediction of different cancers, for instance, they never -- they very seldom think about repeated measures. So since it's so easy to use, I would definitely say that, that would change things. And again, Axel, it was absolutely a fantastic presentation.

Yes. I think we've all been programmed over the recent years to focus on density, the importance of density, dense breast. So how does the dense breast fit into the overall risk continuum here? You mentioned mammographic features. But what does that mean in terms of just using density to look at risk? Is that adequate?

No, it's not. So that's also a frustration. Very many of the other risk models, they now finally added density, but they do it in a very simple way. They take what we call the BI-RADS score, A, B, C, D. So A, especially no density; D, there's a lot of density, more or less the entire breast [ stands ]; and then you have B and C in the middle. That is not enough. That's far from enough because our risk tools indicate that the high-risk group, as you mentioned, the high-risk group, not all of these women have dense breasts. Some of these women have calcifications, as we've shown previously. They have uneven distribution in the left and right breast, both in density, masses and calcifications. And if you add on other factors, maybe we'll mention that, then it's even stronger. So surprisingly, we show that some of the women at high risk -- most of the women at high risk have dense breast. But far from all, surely, 30%, 40% have a sky-high risk, but not particularly the dense breast, which also adds to the complexity because what are we going to do with these women, they probably do not need an ultrasound or MRI. They need to be examined more often. That's what we should offer to these women.

So obviously, a lot more on that point, we'll hear also from Dr. Conant. But I'm curious, given all the data and the millions of images generated and the other, I don't -- over 20 risk factors that you've gathered, do you anticipate that there'll be other information that can be correlated, brought to bear? And there'll be a second or third or fourth generation risk model that would go beyond the mammographic images and age and include these other factors? And if so, what could that potentially look like? And how far away is that? Are we talking decades?

Yes. No, no, not decades. So -- so we have developed a Swiss model, but we see it as 3 model. So the first model is age and the mammographic features, as you say, and that performs absolutely -- it's better than any other model, by far. But if you add a couple of what we call lifestyle touch, whatever you should call them, like family history, use alcohol, tobacco, hormone replacement therapy, your age, your menopausal status, you improve the AUC even further. And then you could, on top of that, in the third step, use genetics. And then -- so there are no dramatic changes. What's changing, which is intriguing is that if you add genetics, the group of low-risk women goes from what you showed 4% to nearly 20%, if not, even more than 20%. So you have you have a group of women, 21 women in [ time ] who will have such a low risk, you could consider -- should you really have -- should you be screened at all, at least not every year or every second year as we do in Europe. So that -- I think that we will add on. But then we're testing -- we're testing proteins, hormones, don't really add that much because, again, proteins and hormones in circulating in the blood are reflected in the mammogram. So on the other hand, this could be that we haven't exhausted all the information that are found in the mammogram. So together with Emily Conant's group and now looking into some mammographic features with a little -- slightly different way of looking at it. And if that adds to per risk prediction, brilliant. So we just have to continue searching. But this is done -- you see I think that -- I mean, the crucial part is that you have a really good data set, and then you could come up with all these smart solutions, and we can test it. It's not 10 years away. But -- so we're constantly working on it.

So I have 2 final questions for you, Dr. Hall. The first one is, although I would love to be one of those fathers of success that I've referenced earlier, I can take no credit and that the work was already underway well before I got to iCAD 2 years ago. So what -- I never asked you this question, what was it at -- and they always say, you should never ask the question you don't know the answer for. But what was it that had you -- enabled you to make the decision to want to partner and start this relationship with iCAD 8, 9 years ago.

Yes. So the thing is when we came up with the idea that we want to create the risk model that answered the question, who were the women that's coming back within 2 years, we started thinking what about all these CAD companies, they try to detect cancer, and that's what they do very successfully. What about using the same, more or less the same algorithm, but use it at the risk prediction? So we contacted a couple of CAD companies. And all 3 of them, you were one of them, were very generous, and said, yes, sure thing, you can do that test. But as long as you stick to risk, whatever that is, don't detect -- and don't do detection. It turned out very quickly that your algorithm was, by far, the best in, for instance, detecting microcalcifications. So we worked on our side and we had a little -- contact every now and then. And then 5, 6 years ago, we showed up at a meeting in San Francisco. And there were a couple of iCAD representatives, and said, hey, guys, this is interesting. Could you to tell us a little bit more about it? And then we went from a very nice interaction to really intense interaction. Over the last 4 or 5 years has been really great and it's working supposedly.

Yes. I finally remember our meetings at both RSNA and European College of Radiology (sic) [ European Society of Radiology ]. Hopefully, we'll be able to go back to those again...

I hope so, too.

As we're getting to the details. Last question, I always like to -- and I know someone [ assumes ], you probably doesn't want to go down this path, but it's a little bit of a legacy question. And it's -- when you look back, a couple of decades from now, look back on this moment in time in the evolution of breast cancer and the introduction, evolution of risk, what do you envision to be? What would you hope would be your contribution to it? What would have evolved in the field over this time? And what would be -- what would you like to see at least your contribution to that be?

That's a good question. So I'm getting back to the cardiologist. So I read a paper, 10 or 15 years ago, that says that, the oncologist, the cancer specialists, they should do what the cardiologists are doing. And if you think about it today, it's the most natural thing that when you go to your doctor, you -- they take a blood pressure. They take the test and check that your cholesterol is okay. And that's -- we do that. We don't even think about it. There's no ethical issues about testing the risk for cardiovascular disorders. But it's still really discussed. Is it okay? How should you inform women about the risk of best cancer? And if you want to assess the risk and how should inform men about PSA and all these things, but it's still quite sensitive. So I hope that in 10 years' time, you test yourself for breast cancer as you test yourself for the risk of cardiovascular disorder. And if there is anything, if there's an increased risk, the doctor is going to take care of you, as a doctor do, when he or she is providing pills for hypertension and cholesterol. Yes, that is where I hope will be the case in 10 years' time.

Well, doctor, I want to thank you for your time today. I know it's late your time. I also want to thank you for your collaboration. I think we follow a -- we're following a little bit of a, it takes-a-village approach trying to collaborate with our Risk Advisory Board, of which Dr. Conant is a part and you're a part. But to move the field forward, I think we all need to work and collaborate together as we have been. And we're very much looking forward to continuing that collaboration. Let's think of it as a journey that we're on together to map out the new frontiers of breast cancer assessment and looking towards the future. So thank you so much, Dr. Hall.

Thank you very much. Thank you.

And with that, I'll turn it back over to Stacey, who can begin the introduction for our next presentation.

Great. Thank you so much, Mike, and thank you, Dr. Hall, for your valuable insight both today and over the last number of years as we really try to consider all the various elements involved in fully commercializing this exciting software and bringing it to clinicians and patients. So thank you so much. All right. We're going to continue along with the live interview format here. And up next, I'd like to welcome Dr. Emily Conant. Dr. Conant is Professor of Radiology, Division of Breast Imaging and Vice Chair of Faculty Development in the Department of Radiology at the University of Pennsylvania. She is an internationally known clinician and researcher in the field of breast imaging. She's published over 180 peer-reviewed articles on topics ranging from novel imaging technologies to disparities and access to imaging, the importance of local culture in mentoring and work-life integration. And she is consistently listed in the, Who's Who of American Doctors and America's Top Doctors for Cancer. At Penn, she was awarded the 2013 FOCUS Award for the Advancement of Women in Medicine and named to the inaugural group of The Academy of Master Clinicians. So big warm welcome, Dr. Conant. I will now turn it over to you and Mike.

Great. Thank you so for that invitation.

Yes. It's a pleasure to have you, Dr. Conant. My first question for you is, do you sleep at all? That's quite a body of work, and I know you're very humble of that. I want to pull out a little bit of that today. And I think what -- as Dr. Hall referred to it, we do have some world-leading radiologists on this call. And I'm going to talk a little bit in this discussion about the patient and clinician interface and the whole triangulation of maybe referring physician, clinician and patients. So let me start with this. As you were the principal investigator on the ProFound AI trial, you're also moving forward as a principal investigator for the U.S. risk trial that will also shed new light or additional light on the work being done by the team at Karolinska. So if you could just go into a little bit in terms of what are the -- what are you -- the design of the study? Or what are the key elements within it? And what -- basically, what kind of important information are you hoping to glean about patient care from this effort that you're embarking on?

Sure. First, thanks so much for the prior talks, they were excellent. And I just love, number one, seeing the images that Axel showed so nicely and then hearing from Per's perspective on incorporation of risk assessment. So the things that we really want to do is take the information that's been so nicely shown over in Sweden and really verify, validate it in a U.S. population. Particularly at our institution, we have a very diverse population. We are 50% African-American, and we know there's a different risk profile there. Earlier onset cancer, biologically more aggressive cancers, also different sizes of the breast and different densities of the breast. So we're basically doing a case control, retrospective study initially, to look at our population with the risk applied. However, I have to say, I can't wait to get it into the clinic. We do have AI ProFound right now in our clinic, so the CAD product for tomo. And it really does change the interaction with the imaging. It's so exciting to me to be able to have a robust and reproducible score. I've been talking for years when Axel showed that case and we looked at her breast and the density and the complexity of the texture, you say, she's probably a premenopausal women, and this is her pattern. And when you look at her, and she's actually, I think, postmenopausal. And there is data. It's almost a fingerprint of a woman in that incredibly dense to now digital data that we have. And the first time I heard Per's group talk, I almost fell off my chair because I think it was 2012 at the International Workshop on Digital Mammography. For years, I've been talking about this concept of looking at the breast, looking at the complexity, looking at the texture, looking at the calcification scattered all over the place. And then CAD would come up and I said, CAD's smart, it knows something about this whole women's risk. And the first time I heard that actually spoken about beyond in our reading room casually was by their group. So anyway, we're excited about validating in the U.S. and then really using it in a more prospective analysis as well. And I think very importantly, in the U.S., the new standard of care is 3D, is tomo, and so we're excited about moving that way as well.

I've heard you -- in fact, you may have been the first one, I heard you use the term complex breast, right? And so with the new -- which captures a lot of things. So with the new -- with the study being done, what else do you have to reveal about complex -- what you would describe as complex breast? And the study, are we talking -- again, as I asked Per, we're talking like 5, 10 years from now, how long will this study take that we can actually start getting some clinical utility and if -- what's being gathered?

Oh, no, no, no. We already have the case control study. So in terms of an initial validation, I would hope that we'd get that published by -- journals take a little while once you submit them, next summer or so. So validation in a very diverse population and a small controlled set of images. But that's the first step, certainly. And then there's always looking at a larger population of cases in the U.S. more prospectively, while sort of retrospectively. But prospectively gathering and then looking back on the radiologist calls, the false negatives that, unfortunately, yes, we do have and what happened to the patients. In the U.S., we've had 2 European speakers here. The U.S. is in a conundrum over breast density. And depending on where you live, the information is very confusing to the patient. And so to be able to have a better dialogue with the patient about their individual risk to drive their precision screening, and I think so much of that, as I think Per said and Axel said, comes from the women's images. It's the radiologist that's use that first. So we're -- I really think we're, if not the gate keeper, we're certainly a big part of it in guiding her towards the next step. Supplemental screening is a huge big deal in the U.S., but I actually think that if we just go by density, BI-RADS subjective, we're going to be screening too many and causing a lot of downstream, I think, costs. And then we're going to be missing some women that really are at higher risk that have, as I say, a more complex pattern. So we need to refine it is the bottom line.

What makes me -- your comments make me think that the entire dialogue or the interchange between clinicians and patients is going to evolve. And they do so fairly rapidly. So that's one question. The other is, is this going to change the -- almost the nature of the people who become radiologists or mammographers because this is a much more [indiscernible] sounds, I don't want to lead the women's here, but it sounds like a far more interactive interface than what may have been the case in the past sometimes when you may have just gotten the results mailed to you, perhaps. So how is this going to change things for the industry?

That's an interesting question. I mean I think that people that choose to go into breast imaging have always enjoyed the patient-facing side of it and the interaction. But I do think that with the kind of data that Per's group has shown and with really, I think, the short-term risk particularly, the radiologist is really going to be a much bigger part of that discussion. And having something that you can present to the patient in a very readable, understandable way is so critical. Right now, the American density letters are not great. I have to be careful what I say, but depending on who's listening, but it's very confusing. And I think as we step forward with more robust and reproducible ways to assess the image data, we're going to become even more important, so yes.

Yes. Just building on this. So we hear a lot about personalization of care or individualized care. How will that -- we often think of it on the therapy side. But how is this going to influence, let's say, the screening regimens that you might embark on? And in your study, particularly, you're going to -- it's going to be very diverse. So how do you envision this sort of playing out in terms of the nature of the conversations and each woman having their own somewhat of a personalized care prescription?

Yes, yes. No, I think it's really important because it varies by age certainly. But I think if we go back to that first case that Axel showed that, to my eye, had a very interesting texture to her breasts. She was not very dense, but if you looked at her incremental increasing risk over time, that's a flag, even though she's not dense that she may have needed something more earlier or certainly more frequent screening. So there's a way for this to not only tell us, "Hey, come back more often." But I am, I think, rare in the radiology world in the U.S. saying, "Hey, maybe don't come back every year," if she's got low risk, low flags on her ProFound score, family -- no family history, et cetera. So I think we're going to be getting into this concept of, well, really precision screening using tools like this.

Yes. I was going to ask you that question, which you already, I think, answered about -- because everybody thinks risk is for women at high risk. I think what you're also saying is that there's a different -- wherever you want the continuum, there's a whole range of different personalized outcomes to follow.

Yes, yes.

That's more of a comment. The question I wanted to actually bridge towards, which is one we have to ask while we're in the midst of a pandemic because we're at a time where we probably lost a good part of the second quarter for screening, I'd like to think that we're fully back at screening. I'd like to hear your comments on that, although it seems as though we're screening with proper safety measures where instead of 3, 4 now, we may be doing -- using all the proper safety measures. So how can AI and the notion of risk, how can that actually be utilized as potentially part of the solution during COVID? And can it sustain and be part of a solution afterwards where we may find the 12-month cadence and all the complexity of coming at the hospital shifted in some way?

No. Well, hopefully -- we never have something like this before, but it is providing a natural experiment. We had a significant drop in screening in the spring. We're up doing screening now, but I'm afraid we may -- either patients are going to just not come -- they're cutting back on elective procedures now in my state. So screening may follow. So there's certainly the idea of triaging by risk women and even looking in retrospect, saying, "Hey, last year, you had a risky appearance. We really want you to come in." And so I think that's important. And we've spoken about moving beyond COVID to workforce issues. I mean Axel mentioned it. Being cost-effective, there's beginning to be some really interesting data from iCAD and others, looking at triaging, who needs to be double read, which is the European standard, and then consensus, versus in the U.S., one of us reads -- actually, we gang-read with our residents and fellows, is what I call it. So lots of readers. But is there a way to better triage cases for double reading or not for double reading, for interval screen? There's a whole lot of interesting downstream efficiency, I think, that can come out of this kind of really robust data. So hopefully, no more COVID, Mike.