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

Good day, everyone.

Good afternoon. Sorry about that, Mike. I was just going to do a quick introduction. We're starting the medical device track in the afternoon session of the fourth day of the 41st Cowen Health Care Conference, and we're thrilled to have CEO of iCAD, Mike Klein, here to run through a presentation of the story. There's been a lot of progress made since last year's health care conference, Mike. Great to have you here today, and I'll just hand it over to you.

Thank you. And sorry about jumping in early.

No problem.

Thank you for the invitation, and I want to thank everyone for joining. It's my pleasure to talk about the iCAD story. And with that, I'm going to jump in and just take a moment as I put up the safe harbor statement, which I'm sure you've all seen before, regarding forward-looking statements. And we'll move forward by taking a look at the first slide, which is a bit of an overview of the company. iCAD is a company that finished last year at about $30 million, just under that, with a substantial growth rate into 2021. And the 2 parts of the business that you see on the page here are what we call our discovery business and our recovery business on the right. That's a nice way of saying or a different way of saying our AI or artificial intelligence business. And on the right is our detection -- or rather our radiation therapy business. They're both based on disruptive technologies. They, on the AI side, are based on a new way of reading mammography or mammograms that are now being generated with 400 images versus the 4 of just a couple of years ago, and we help find literally that needle in a haystack out of those images. And on the right, it's a radiation therapy product that's a miniaturized X-ray source, smaller than a matchstick that can fit anywhere in the body, approved for anywhere -- or cleared for anywhere in the body and with good reimbursement. So 2 sides of the business. 75% of our sales are on the left-hand side, and that's an 87% to 88% margin business moving towards north of 90%. And on the right-hand side, a roughly 60% margin business. And together with a combined margins of -- in the low to mid-70%. And we see this business growing at a rate of north of 35% for the foreseeable future, and that also allows us to grow with both the recurring revenue model as well as an upfront license model, and I'll get into that in a moment. I'll first start by drilling a little deeper on the AI side of our business. You'll note that on the AI side of the business, that we -- if you've been following the story a bit, that we hit 1,000 installs after about 22 months of our launch of the product. And we believe that we'll get to this level of 2,000 far quicker. We actually believe that the adoption rate will be larger than 50% a year. But given that some of the revenue will come in on a recurring revenue basis. So we might take $1 million order, as an example, as a onetime sale or take it as a $400,000 sale over several years and continue taking it, $400,000. We see ourselves in this range of north of 35%, but allowing ourselves -- which allows us to take sales at $400,000 and still grow considerably. So we do expect this growth of installed units to be at 50% a year, even though the revenue will lag a little bit. But we benefit over time by being a recurring and predictable revenue stream. This is a product line that, again, has mid- to high margin, 80% margins, moving towards north of 90%. It's an extremely large market, and that market doesn't even include China, Japan and Korea that literally would almost double the size of this market. Those are the installed cameras that we apply our product to. We are effectively -- as Photoshop would be to your own camera, we are to imaging and, in particular, mammography imaging. Again, the AI business is 75%. We anticipate turning a profit this year. We have substantial cash on hand, and we believe we'll spend a modest amount -- a moderate amount of cash on hand before we will turn profitable. We -- you can see the slides here on our profit timing. We may put more capital to work because we do have a significant amount of capital to be able to invest. And we believe that we can move the top line even faster given the cash position we are in, particularly in global markets and also in the United States, given that our sales reps each are carrying a substantial pipeline of sales. And we believe that we can with a few more sales reps, which is what we intend to do this year, actually grow the top line at a potentially even more accelerated rate. A little more detail on our product. We believe we are still not only in a first mover advantaged position, but a sole mover advantage, in that we have yet to see a competitive system anywhere in the market that's been installed. And we sell to all of the channels, from direct sales to distributors to OEMs and also to picture archive computing systems, PACs. Those are the railroad tracks that folks run their images and data through. So with that, we'll dive a little bit further into the story and take a look at the evolutions that have happened in AI. The first generation was film. In fact, on the left, the analog. And I ran a company that actually took us from film into digital, which was ultimately sold to Hologic. And that capability enabled the images to go on screens. You can see the screen on the left there, 2D mammography, 4 images. But what's happening with the third revolution, which started 4 or 5 years ago, was not simply 4 images. This is now 400 images. Each of those images broken down into 1 millimeter slices and making a stack of 80 to 100 images. And imagine the challenge of having to find that 1 or 2 slices out of a deck of 400 -- almost 400 cards that you have to flip through in order to find that single couple of slices, 1, 2, 3, 4, 5, depends on the size of the lesion. So what we do, just as one would sequence the genome, we sequence the images and are able to pull out the 1, 2, 3, 4, 5 images, and that's all you need to look at to see where the cancer is. And if you click on it, it'll take you exactly where it was in the deck and you could toggle up and down. At the same time, as we've been introducing this software for 3D tomosynthesis, of which there are 10,000 systems in the market that all need our version of, let's say, Photoshop, we're also moving beyond just detecting cancer and into the future, which is being able to risk predict or risk assess what will emerge as one moves forward. Now one of the things I've learned being in this field for over 20 years is that anybody can claim that they could be a professional at this from a -- if you are willing to accept the fact that you may make mistakes. But -- that, I mean, one of the ways that you could never miss as a radiologist or reading film is just mark everything a cancer. And if you do that, you'll be right 6,000 -- 6 out of 1,000 times. But you'll also be wrong 994 times. So it's one thing to find the cancer, but it's another thing to rule out the things that aren't because all of us could just say everybody's got cancer. So what AI software does is it helps you rule in what is a cancer and rule out what isn't. In fact, what you rule out is just as important as what you rule in because if you begin to show marks and indications of things that are just wild use cases for your reader, they'll not use the software. So the goal and the magic is to mark just the cancers, and that's called sensitivity. And not labeling something a cancer that is a cancer -- that's something -- sorry, rather not mislabeling something a cancer would result in a recall. So the goal is, as you see on this chart here, is to be as far north as you can be because that means you're more sensitive. And be as far to the left as you can be because that means it's very low false positive. So that's what would make for an outstanding radiologist. In this case here, you can see these are 24 radiologists that were used as part of our FDA filing. And we had to have these 24 radiologists read hundreds and thousands of images, and they read the images without the software, on cameras, GE, Hologic cameras and then with the cams. And here, you can see, sensitivity is all over the place. You see somebody pretty low. You see somebody pretty high. You see some people out to the right, which means more false positives. And you see these circles as different size, and that's the amount of time one spends on reading. So with that, I'll move to the next slide, which will provide a little more detail on this FDA study. As we got into the study, you can see it here in greater detail, you can see that all of these dots -- and now we're looking at the right-hand side of the page. All of these dots represent 24 radiologists. They all move up into the left. Notice also the conformality and that they all sort of cluster together. So even more nascent readers become more professional readers with AI. They all move over the lines. It's actually what the algorithm, the score is. So the reality here is that with the doctors reading the algorithm, the doctors are better. And this is where we often use the statement, for those who are afraid of AIs, that AI is not going to replace the doc. You can see on the right that we work with the docs. But what will happen is that physicians who don't use AI will likely be replaced by the physicians who do. Because as you can see on the right, you not only are moving up. You're also moving to the left, which means more sensitivity moving up. Moving to the left means fewer false positives. And the circles are getting smaller, which means you're actually reading quicker. So with that, I'll show you how we did when we submitted to the FDA. When reading with our software, physicians will find 8% more cancers. And we're talking about 40 million women a year getting screened and 300,000 cancers. So we're finding 8% more, which is a big number when you add it over that population. 7% fewer recalls. And this may be, to some, the big money shot. These are radiologists that cost $500,000 -- $400,000 to $500,000 a year. And when you reduce their reading time, it's literally like reducing their cost by 50% or $250,000, or you can have one radiologist do the work of 2. So this is one of the huge advantages -- the huge health care economic advantage of this offering. Typically, that would be $250 an hour, reduced down to $125 an hour. And the software -- when you look at those numbers, $30,000 of software gets paid for in that -- literally in 6 to 8 weeks. And this is one of the huge value propositions. So here, you see it summed up. We find more cancers, we call back fewer cancers, and we also are particularly good at reducing reading time. And even complex cases, we reduced even further 57%. And this score of AUC, area under the curve, that's sort of the miles per gallon score in artificial intelligence. And you can see here that this score is a summation of how much you find subtracting what you miss. It's almost like your old SAT score. And with this, we do substantially well, and we actually are now able to, because the FDA post all these, have these scores, compare to other products that may come to market. Now we'll be launching an even better product in the second half actually by -- in the second quarter, and it'll actually improve these scores. But we also know that there may be some folks who will come to market at some point. But when we take a look at those scores, and I'll show you these in a moment, they're really first generation, probably where we were about 4 years ago. And we'll tell you a little bit more about that in a moment. But I want to dig a little bit deeper into why people would go with this solution. When you move from 2D to 3D DBT, digital breast tomosynthesis, that's 3D, you literally are seeing an incredible increase in the number of images. These studies, and this is a conservative estimate, 280 slices instead of 4. That would mean 28,000 images in a week, 5.6 million images in a year when you used to see 80,000. You're seeing more images in 3 days than you used to see in a year. And this is from a real site that runs -- and that's the real CEO, MD of the 10 sites. They also -- with that many images, they were calling back 30% of their patients for unnecessary biopsies. We're able to take that under to a 0 level compared to the way it was before using our software and, in fact, reduce their recalls 7%. So their recalls went up and biopsies went up 30%. We had a reduction all the way back to 0 and then another 7% below their score of just reading with the camera or the 3D digital breast tomosynthesis system alone. So moving on to the next slide. You'll see this is one of those millions of cases. If you're a doc and you're reading 28,000 slices a week, this is just one of those slices. And try to find the panda there in the -- the soccer ball among the panda bears. And I've just given you a little hint that it's sitting here in the column that's missing. And if you look really carefully, you can see that it will be right there. And the way this operates is the way AI does, which I almost call the Michael Angelo algorithm, which is when he answered the question of, "How did you sculpt the statue of David," he said, "Well, simply, I just removed all the marble that didn't belong." In this case, we removed everything that had 2 eyes and nose and a mouth and ears, and what's left is the soccer ball. And that's sort of 1 level of 27 levels of a deep machine learning-convoluted algorithm, as referred to, that separates signal from noise. It's a lot of complex work that we can do in seconds, and it saves the doctors' minutes. And when you're reading 100 cases a day, that's a big deal. This is, again, our score. And you do need to hit this trifecta of sensitivity and reduction of recall and -- as well as the total score of adding them together, AUC, the area under the curve. And if you do that, you can produce a winning score. And just to drill down -- so it's not simply just about the finding of cancer. It's reducing those false positives, and those false positives cost the health care system billions of dollars. And you can imagine, that's not very helpful during COVID when you're really struggling to get as many patients screened as possible. And since there are 40 million patients screened a year, and we missed a whole quarter, we're already 10 million women behind. And we're only screening 2 an hour versus 4 an hour for proper spacing of patients, and we can't afford any unnecessary callbacks. So it costs billions of dollars, but now it actually causes patients to be delayed in their detection if they're pushed back by people being erroneously called back. So this is an extraordinarily important COVID play. And here, you'll see that when one launches a competitive product -- and this is, candidly, the performance of one competitor out there, a major company in imaging that had a product that we went against in the fourth quarter. And the results were that while we decreased the recall rate, this was their FDA claim, they increased the recall rate. And while we reduced the time for treatment, the competitor's technology that's out there -- and it's so important to reduce the time because here you can see the score card. If you can reduce a $500,000 physician's time, you can save over $260,000. So back to the score. If you actually increase the amount of time that -- and we're, again, reducing time by almost 53%. Your increasing time, which is the case for the competitor, you're increasing time 12%, you're going the opposite direction. You're making that $500,000 radiologist cost more. You're going to need more of them instead of half the time and half the FTE. So again, when we compare the miles per gallon of us versus a competitor out there, you can see why there's been no competitive systems installed. And we've gotten to 1,000 installs, and we think we'll get to 2,000 a lot quicker. You can see these claims on websites, on the FDA site. They're not my numbers. They're the FDA's numbers. So I'm going to move a little quicker now and get a little more deeply into the story and talk about what is a very exciting part of our future and what's leading to rapid conversions and at this point, why we grew 70% in Q4 over Q3 and why Q3 grew a full 44% over Q2. And it's the introduction of this risk product. This risk product allows us to examine the various risk factors that a patient presents with to see if we could predict their future. This is based on 8 years of work of 70,000 women that were screened every year over 8 years correlating 8 risk factor -- I'm sorry, 27 risk factors, of which 7 or 8 really made a difference, where we can actually predict which patients would come back with a cancer in a year or 2. And what we found is that this is almost like having a GPS system that instead of telling you what's going on around you right now, as you can see on the left, this is literally -- we've looked at this intersection, you might say, for 8 years. And we know all the risk factors and we could tell the probability of there being an accident. And in fact, you even help us make it smarter by telling us when you get to that accident that maybe there -- if it's there or not. What we do is we literally tell you what will probably be there. And when we do it, what we're really telling you is what is actually there that you may not be able to see but we can see on the algorithm. And by saying that -- I'll be giving an illustration in a moment. What we're really saying is that there is a disease present that is below the visual field of the physician. And although we can't show it to the docs today, because if we did, they would literally say, "I can't see anything." What we're presenting is data that's below the visual field. And when we took this 27 risk factors and we laid it out over 70,000 patients at the Catalan Institute, we found that this risk factor was different for women. And it was different in a very significant way, where people who are at high risk, as you can see here, had a 750% greater chance than women who are at average risk. And this was illustrated in this paper in radiology. And I'm going to give an illustration of exactly what this means from an imaging perspective. Right now, women are pretty much told you have either a 90% -- they're either in 2 categories. Either everything looks fine, and that's 90% of the time; and 10% of the time, let's go get some supplemental screening, like an MRI. So either 90% of the time we'll see you in a year; or 10%, we'll see you for an MRI. What we found through the study of all these women at Catalan Institute is that there's more of a spectrum and that women are across this spectrum. And the people that are in the upper part of the spectrum, which are actually 36%, have substantially higher risk even though you may not actually see that there is a cancer present today. Now I'm going to show you an example of this in a real case in a moment. But I have to answer the question first. Why does this matter? And you can see here that -- the obvious gain of finding cancers early and the obvious financial gain. So the survival benefits of early on the left, the financial benefit. It costs 4x more to treat a patient that on the left, in a late stage, has only 1 out of 5 chance to survive. Now that's a shame because all of these could be found early. So the way this product works -- and we put it into the hands of the physician. Instead of a binary black and white outcome, indeed, it's a 3D color spectrum that puts everybody with a unique risk score. So we now have 2 products. One that's $30,000 for detection and another -- that's another $30,000 for prediction. So this is how the scorecard that physicians might get. You scored 39% on today's risk. Your density was B, which is at a moderate level, moderate to low, sort of towards the midpoint. And this would be a score of someone with a low risk. And here, you can see this patient with similar characteristics. Everything looks the same today, but they have high risk. And that is because we're seeing things that are below the visual field that may not be seen for 2 years. And we can't show it to them today, but they would dismiss it as a false positive, and how do you biopsy something you can't see. So we give them the score, and they're able to take a look at what the probability is. These are all the scorecards we provide. And this is the case study I wanted to get to. This woman had no visible cancer risk in the image taken in 2014. There's no radiologist that ever -- has been able to find a cancer on this image and we couldn't even find it with our detection product. But then when we introduced our new prediction product, we actually said, "Wow, in the same case, this person is a moderate to high-risk factor." And with a moderate to high-risk factor, you could correlate that over to the data I showed you earlier, that score equals a 7.5x greater risk of cancer or 750% more -- 750%. So when this patient came back in 2016, this was what we found. But this is also what we could have found in 2014 with today's tools. We find today what will emerge tomorrow. And this has become a huge play for us during COVID because when you've got 10 million patients and a growing backlog that some suggests is now $12 million to $14 million, you want to know who to call back. People are using our prediction tool to -- on last year's images to know today what they could have known last year. And they don't want to be in a position where they call somebody back too late and didn't know that they had a high risk score. So what this does is it takes us into the realm of risk-adaptive screening, where instead of just high-risk and average-risk patient output, you -- and a binary black and white outcome, we wind up with a full color spectrum, a predictive capability, and we wind up at screening that looks like -- this is what it would be today, which is age-based screening -- I'm sorry, this is what we'll move to. From age-based screening to different screening regimens for different type of women. So this is personalized care. This is care in COVID, accelerating us down the path of risk-adaptive screening away from everybody gets screened at 1 year no matter what. And now you can use these tools on a differential basis for the patient. So in my last moment or 2, I'll just talk a little bit about this area of risk-adaptive screening, which is where we're moving. We're taking data that's there today. It's a $30,000 product that's sold into -- that's providing what you see here. And I'm going to show you some economics in a moment about how the market builds out. We -- again, here's how it will look in the future. Instead of on the left being just a binary outcome on the right, you'll have multiple ranges of scores for patients. And when this product, which sells for $30,000, is added to the current product for $30,000 -- we used to have a market that we thought of is just 10,000 units that were 3D. That was going to move to 18,000 as the rest of the sites adopt. But now we have a product not only for $30,000 plus service and not simply 10,000 times 33,000 or a $330 million market. We see the market much broader now. We see the market as having these 10,000 3D installs, and now we have an offering for the 2D units that are out there. And it's 2 offerings. It's a detection offering, what you see today. Kind of like the GPS system, what you see today, as well as prediction for the future. So when we look at the market, we believe that the market now is not simply 40,000 units times one product at $30,000, but basically 2 products -- 2 at $30,000 for a total of $60,000 or $2.4 billion. And that doesn't include service that would add another 10% to the scores. Now this market is coming in 2 ways to us. We either can get it on a license basis, straight upfront, or we can take it on a subscription basis, which is another way of saying, instead of me taking a $1 million order this year, I may take it as a $400,000 order, back to my earlier point, but I might get that every year. So after 5 years, I've made $2 million. So you can see that when you take it on a subscription basis, you may have a lighter score in terms of revenue in year 1, and this is $343 million per year. This is basically $2.4 billion spread out over 7 years. And now I'm showing in the subscription model how it could come in and if it was all on a subscription basis. Now we only expect about 10% of the business this year to be subscription, about 20% next year and then 40%. But we love this model. You can see that in year 4 you're doing better with the subscription model to margins go to above 90%. It makes us predictable and helps the multiple for the business. And again, instead of $30,000 for one license, you might pay $12,000, but it also allows us access to those smaller sites that wouldn't have bought a $30,000 license. So this is basically the story line, which is what I call frictionless procurement. We have a product that is going to move into 40,000 cameras with 2 offerings of a total of $60,000 that can be taken as either upfront licenses or on a subscription basis, however the customer wants to acquire the technology. And with that, I'll end on this slide, which is, why are we so sure we'll be successful? Well, for one, we've got 7,500 installs, which is probably 7,500 more than anyone else at this point. We're already screening 50% of the U.S. patients. We're collecting tens of thousands of images each week. And the biggest barrier to entry is not even the images that we have, which is the largest breast cancer repository in the world. It's that there are 6 mammo cameras, systems from GE, Fuji, Siemens, Hologic and some smaller ones in Europe. There are 15 workstations and 20 PAC systems. They're like railroad gates, tracks with all different gates. And when you add them up, that's 1,800 combinations. We work on all of them. It's taken us 15 years to do it, and that's probably the biggest barrier to entry of all. And we continue to screen patients, continue to get smarter each day. Tomorrow's product is smarter than today's product. And we believe that as this field moves more to software and data than it does just hardware, like the -- like your computers, the differentiation happening with the software, and that's where we are in the area that you have. So that's the detection part of our business. That's the primary part of our business. That's what I wanted to highlight today. And with that, I believe I'm pretty much over the time limit. So I want to thank all for listening in. I want to thank Cowen for the invitation and for Josh Jennings, all he does and the great team over at Cowen. Thank you very much.