Richard Vlasimsky is using AI to create a low-cost way of vastly expanding our ability to detect lung cancer early. His algorithm for detecting lung nodules in chest x-rays just got FDA clearance. Listen to this remarkable founder who’s done a lot with modest funding.
Highlights:
Sal Daher Introduces Richard Vlasimsky
What IMIDEX is Solving
"... how much your studies cost to get your 510k.."
"... We had multiple radiologists looking at the image, and then we had a final truther that adjudicated these independent reads that the radiologist did..."
"... Payers care about catching it early so that they have better outcomes and the costs are much lower ... The providers, being the medical systems or the physicians, typically hospitals, they care about it because of liability issues and also because they could have the opportunity to treat something and they would get the revenue stream..."
"... It's like a very fine-tooth comb picks up all the nits. The specificity relates to how many of those things that you pick up are actually what you're looking for and not false positives..."
Exit Possibilities
"... his definition of a startup is an enterprise that uses resources that it does not yet command..."
Richard’s Entrepreneurial Journey
"... You had this experience of joining a startup, then creating your own startup which was using machine learning..."
Advice to the Audience
ANGEL INVEST BOSTON IS SPONSORED BY:
Transcript of “Finding Lung Cancer Early”
Guest: Richard Vlasimsky
Sal Daher: I'm really proud to say that the Angel Invest Boston podcast is sponsored by Purdue University entrepreneurship and Peter Fasse, patent attorney at Fish & Richardson. Purdue is exceptional in its support of its faculty, faculty of its top five engineering school, in helping them get their technology from the lab out to the market, out to industry, out to the clinic.
Peter Fasse is also a great support to entrepreneurs. He is a patent attorney specializing in microfluidics and has been tremendously helpful to some of the startups, which I'm involved including a startup, came out of Purdue, Savran Technologies. I'm proud to have these two sponsors for my podcast.
[music]
Sal Daher Introduces Richard Vlasimsky
Welcome to Angel Invest Boston, conversations of Boston's most interesting founders and angels. I am Sal Daher, an angel who's very, very curious to find out how interesting technology companies are being built. Today, we have privilege of having with us, a founder, who is doing something very exciting in the field of machine vision, machine learning. Welcome, Richard Vlasimsky.
Richard Vlasimsky: Thank you, Sal. It's a pleasure to be here.
Sal Daher: Awesome. I connected with Richard via my friend and fellow investor, Eric Smith, who is an investor in the company, and I also had a conversation with Keith Hearon, who has been on this podcast, who is also an investor in IMIDEX, the company that Richard has founded. Richard, tell us what it is that you're doing and what problem you're solving.
What IMIDEX is Solving
Richard Vlasimsky: If we were to distill it down into a few words, it would be, we're saving lives from lung cancer through early detection of lung nodules in chest X-rays.
Sal Daher: The way you're doing it is, you've collected a data set of chest X-rays, you've trained your AI algorithm on it, and you have achieved a degree of sensitivity, which has called the attention of people in the field. Also, I understand that the FDA has cleared your device under a 510(k) application for use. Please fill in the gaps here.
Richard Vlasimsky: Absolutely, and probably expand a little bit on the problem as well. Just in terms of context, chest X-rays are the most ubiquitous radiological procedure that's performed in the world. We actually aquire two billion of them every year.
Sal Daher: Two billion chest X-rays are taken a year?
Richard Vlasimsky: A year, worldwide.
Sal Daher: How many have you actually studied? How many have you looked at to train your AI on?
Richard Vlasimsky: We studied three-quarters of a million.
Sal Daher: Three-quarters of a million, 750,000, so there's a lot that you can still cover. In the US, how many extra chest X-rays are taken?
Richard Vlasimsky: In the US there's 240 million a year taken.
Sal Daher: 240 million.
Richard Vlasimsky: That's on an estimated 40 million individuals.
Sal Daher: Excellent.
Richard Vlasimsky: Roughly 10% of the population every year will get a chest X-ray for a variety of reasons. Never for suspicion of lung cancer.
Sal Daher: No, it's always something else.
Richard Vlasimsky: Correct.
Sal Daher: Swallow the golf ball or something.
[laughter]
Sal Daher: Or a kid swallowed, like my grandson, swallowed a paperclip. We can see the paperclip right there inside the kid, a big paperclip inside of a small kid.
[laughter]
Richard Vlasimsky: Yes.
Sal Daher: It came out. We know it came out because they did another X-ray and there was no paperclip, but we never found it.
[laughter]
Sal Daher: I mean, if it weren't made of steel, I would assume, oh, it got dissolved, but I doubt it. It would stand out very clearly in an X-ray piece of metal inside a kid, inside a three-year-old. Oh, I remember that day well. Anyway, getting back to business here, you've trained your AI on 750,000 images of chest X-rays. Where does it get you? Tell me about the reliability of radiologists in picking up serious--
Richard Vlasimsky: Lung nodules.
Sal Daher: Yes, lung nodules, versus what the sensitivity of your AI.
Richard Vlasimsky: It's a great question. I'll preface this by saying chest X-rays are probably the hardest modality to read for radiologists because it's flattened in two dimensions in nature and your chest has a lot of organs in it, so all that is superimposed on it. Compound that with the fact that there's fewer radiologists and there's more radiological procedures being performed, so radiologists are being asked to read double the number of X-rays that they would have 10, 15 years ago. With that backdrop and knowing that the prevalence of this is low, so there's a reverse vigilance bias, typical radiologists catch less than half of the lung nodules in usual care and chest X-rays.
To help provide cognitive support to the radiologist, we developed what's called a CADe device, computer-aided detection device that highlights on the image on the X-ray where a suspicious lung nodule is present to call the attention to the radiologist, where they ultimately make the final decision. To get to our 510(k) clearance, we had to conduct a number of pivotal clinical studies to demonstrate the safety and efficacy of our device. One of those was a multicenter, what we call bench test, where across the US on 12,000 X-rays, we demonstrated an average sensitivity of 83%. Again, 30 points higher than what's typically seen in usual care.
"... how much your studies cost to get your 510?.."
Sal Daher: Very good. Can you give me a ballpark figure if you're free to talk about this, on how much your studies cost to get your 510?
Richard Vlasimsky: It's a great question. Put it this way, to get to 510(k) clearance, we're pretty scrappy, I would say, and we were able to get to it with $3.2 million in funding. When we look at individual studies, the predominant thing that we did when we worked with hospitals and health systems was we did an exchange. We said, "Let's do the study for free and we will give you a license to our software for a limited period of time once we clear the FDA and then it could be used to affect patient care."
Sal Daher: Oh, okay. How many systems did you approach for that trade?
Richard Vlasimsky: We approached 15 or 20 systems. We worked with five different-- Actually, I should say six different sites.
Sal Daher: A typical number of X-rays from a site would be 20,000, 30,000, something like that?
Richard Vlasimsky: Pretty close. A medium-sized hospital and we're focused on the emergency department and outpatient setting. That's where you're going to find the, as our CMO likes to call it, the walkie-talkies, the healthy people, that you're going to make a difference. That's typically around 55,000.
Sal Daher: The point here is that if a radiologist is looking for a lung nodule, they're probably going to go do a CT scan, tomography scan, which is much, much sharper and will give depth, and they'll be able to discern it much better. The reality is that there's a lot of scope to pick up lung nodules in people who are having X-rays for all kinds of reasons, like a paperclip in your stomach.
[laughter]
Sal Daher: Gosh. He knew he was not supposed to have a paperclip in his mouth, my grandson, but it was a tricky situation because it was one of those paperclip holders, I'm sorry to digress on this. I got to tell this story.
Richard Vlasimsky: Oh, please.
Sal Daher: It's one of those paper clip holders where there's a magnet around it, and he's budding engineer, and so he was fascinated by the fact that you could turn it upside down and the magnet would catch the paper clips before they fell out because there's a magnet around it. He was so confident in that, that he turned it upside down and shook it in his mouth like this, just to show off.
Gosh darn, the thing was overloaded and one paperclip went straight into his throat. My daughter, not his mother, his aunt, who happened to be right next to him, reached inside his mouth and could barely touch the paperclip inside his throat but couldn't pull it out and then he went, "Glook," and it went down. [chuckles] Never to be seen again except in a CT scan or whatever they did. I don't know if it was X-ray, I think it might have been just a--
Richard Vlasimsky: I would guess it would have been an X-ray.
Sal Daher: An X-ray because it's very easy to see. There's no ambiguity there.
Richard Vlasimsky: Absolutely. It's bright white light.
Sal Daher: Anyway, getting back, I couldn't [chuckles] resist telling the story. The idea here is that you have a very large population of people getting chest X-rays, very low-burden test because radiation that you get from that is nothing. It's like background radiation because technology has progressed a great deal. It's something that human beings don't do very well at because there's a lot of stuff stacked on top of other stuff, but machine learning algorithms can pick those things up very easily. Now, the images that you got from the medical centers, how is it that you decided what was a lung nodule, what was not?
"... We had multiple radiologists looking at the image, and then we had a final truther that adjudicated these independent reads that the radiologist did..."
Richard Vlasimsky: It's a great question. We had grades of truth that we internally called platinum, gold, silver, and bronze level ground truth. We worked with anything that was silver above in our parlance and silver ground truth was a panel-adjudicated sample of data. We had multiple radiologists looking at the image, and then we had a final truther that adjudicated these independent reads that the radiologist did.
Our gold level of truth was when we had CT confirmation because CT is a much more sensitive modality than chest X-rays, and then platinum level is the ultimate ground truth, is when you have a confirmed malignancy from a biopsy.
Sal Daher: Very interesting. Basically, that three million that you mentioned, is that just the studies themselves, or how about the acquisition of the data set? How much would you say that that cost you?
Richard Vlasimsky: Having summed up the number, I would say, and some of it is hard to evaluate what it had cost when we ran these studies. In a lot of cases, we got the data for free in those cases, but it did cost us a lot of money to pull it all out and curate it and that sort of thing.
Sal Daher: Excellent. You've created this really impressive algorithm that can find lung nodules in X-rays that humans have a hard time understanding. Who cares about this? Who might pay you for this product?
Richard Vlasimsky: The provider and the payer both. The patient cares too.
Sal Daher: The provider is the hospital, the medical system, or whatever, and the payer being the insurer or CMS Medicare.
Richard Vlasimsky: Yes. Providers want to provide good quality care, obviously. We help reduce misdiagnoses. There's an avoidance of medical malpractice component to this, but ultimately, for the provider, if you could think about, and we don't apply to pediatrics, but let's say an adult came in that swallowed a paperclip and they took a chest X-ray and they missed that lung nodule that ended up turning out to be cancerous, that was a patient that they could have identified early and treated early.
The downstream treatment pathway involves rad onc, medication oncology, infusions, surgery, and so it's a significant amount of additional revenue that a hospital's missing when they don't identify these nodules. We help them with that.
For the payer, which has a different orientation, they're thinking more about cost and clinical utility. The earlier you identify someone with lung cancer, the less it costs to treat them and the outcomes are much, much better.
Sal Daher: Hugely better. If it's caught early, it's really a minor thing. If it progresses, it's a horror show, extremely expensive and the success rate is low.
Richard Vlasimsky: Yes. Stage 4, sadly, is what half of lung cancer patients are diagnosed at today and the prognosis is pretty grim. It's 5%-10% survival.
Sal Daher: I know someone whose mother is in that situation and my heart goes out to him, prayers because it's a situation where going through chemotherapy, someone in the 80s, not a smoker, 15% of lung cancers are people who don't smoke.
Richard Vlasimsky: Yes, you just need lungs to get lung cancer.
Sal Daher: Yes, and it's like, "Are you well enough to take the chemotherapy?" Because it wasn't picked up early. It's very sad.
Richard Vlasimsky: That's why we're here.
"... Payers care about catching it early so that they have better outcomes and the costs are much lower ... The providers, being the medical systems or the physicians, typically hospitals, they care about it because of liability issues and also because they could have the opportunity to treat something and they would get the revenue stream..."
Sal Daher: I understand that you-- We'll get into this later on. Keith mentioned that you had a personal reason for pursuing lung cancer. If you don't mind, I'd like to talk about that in the second part of the podcast. Let's continue on the business vein here. Payers care about catching it early so that they have better outcomes and the costs are much lower.
Richard Vlasimsky: Correct.
Sal Daher: The providers, being the medical systems or the physicians, typically hospitals, they care about it because of liability issues and also because they could have the opportunity to treat something and they would get the revenue stream. It is a low-cost thing for them because basically, this is a SAS product. It's like another subscription, another SAS subscription. It's not like a massively expensive pharmaceutical drug that they have to spend a lot of money and refrigerate or nonsense like that. [chuckles]
Richard Vlasimsky: Exactly.
Sal Daher: It just lives on the cloud and it runs in the background. I spoke with Keith Hearon today. By the way, listeners should know that Keith is an alumnus of this podcast with Matthew Stellmaker, episode titled Poly6. Keith and Matthew were co-founders of Poly6 and they exited. It was very nice. Keith has had a couple of exits since then. Matthew has been running Poly6. It's very nice to know that Keith is on the board of IMIDEX, a vote of confidence here, in addition to being introduced to this by Eric Smith, shout out.
Richard Vlasimsky: Keith is an amazing one-of-a-kind individual, insanely intelligent.
Sal Daher: The word genius gets thrown around, but this guy is a freaking polymath. He was an honor student at Duke University and he's a kind of guy MIT missed, Caltech missed. He invented a whole different chemistry for Poly6, had patents on this. It's basically an entirely different chemistry, which found among remarkable, which has caused their exit. It was supposed to be like an environmentally friendly plastic, the chemistry that Keith Hearon had invented.
It turned out to be ideal material for creating negative molds for carbon blades. It became an aero and astro company, got acquired for that. [chuckles] It's incredible. If you talk to him, the guy is unbelievably articulate, has a very good business mind, and got--
Richard Vlasimsky: Amazing.
Sal Daher: Gosh, you run across people like that at this podcast. I'm very proud that Keith Hearon, that's H-E-A-R-O-N, and Matthew Stellmaker, S-T-E-L-L-M-A-K-E-R. Matthew also is a genius but on the business side of things. Two amazing founders. Anyway, getting back. Keith was mentioning that he thinks that there is the possibility of hundreds of millions of people using this because it's so low burden and hundreds of millions of X-rays taken every year. Did you say 240 million something like that?
Richard Vlasimsky: 240 million in the US.
Sal Daher: Chest X-rays. Why not have every single one of them scanned for a chest nodule? Because the sensitivity is very high, so you pick up 83% of chest nodules. What is the specificity on this?
Richard Vlasimsky: We look at it from a little different endpoint so we didn't measure that for our studies. For the FDA, we have to demonstrate that we had overlap with a lesion and so there can be multiple nodules within the chest X-ray. We measured in what's called an FPPI, false positive per image. On average, that's-
Sal Daher: 1.5.
Richard Vlasimsky: 1.5.
Sal Daher: 1.5 false positives per image. That's an absolute number.
Richard Vlasimsky: Yes, and the reality is, what typically happens is there'll be a chest X-ray where it's not a single solitary nodule, there's multiple nodules in the image.
Sal Daher: Typically, there might be six or seven, one of those is a false positive. Since almost every single one of them is multiple, you're getting a signal, there's a signal there?
Richard Vlasimsky: Correct, and just to be clear, the majority of the cases that we validated and trained on were our target for our mission of saving lives, so it's the single solitary nodules which are more of the early stage.
Sal Daher: Ah, so you want a sensitivity so that you can do a low-burden test such as?
Richard Vlasimsky: Low-burden test, get a second pair of eyes, the radiologist makes the final decision.
Sal Daher: The next step would be a liquid biopsy.
Richard Vlasimsky: And that, I would say in the more distant future. Immediately the next step would be getting a CT. The problem with-- CT is not hugely accessible. There is preventative service guidance that the USPSTF has put forward for smokers over a certain age. All the hospitals in the country are trying to figure out ways to actually get adherence up to the screening programs but they're hovering above 5% now. It's been that way for a while.
There are a lot of reasons for that. We don't necessarily get into but we provide a way to funnel people into that screening. By definition, if you've got a lung nodule, it puts you at high risk for malignancy, and so put you into CT screening, make an evaluation about treatment from there. In the more distant future, there are a lot of super interesting companies that have developed liquid biopsy tests. With a blood test, they're able to detect circulating tumor DNA. We see ourselves as hugely complementary to these companies.
They're not actively used right now, their tests are very expensive. They're in varying degrees of FDA clearance but we offer a high sensitivity and they offer a high specificity, and the two together are, for lack of better words, a peanut butter and jelly combination.
Sal Daher: Sensitivity means you pick up everything that's there.
Richard Vlasimsky: Correct. That's the most clinically important thing to do.
"... It's like a very fine tooth comb picks up all the nits. The specificity relates to how many of those things that you pick up are actually what you're looking for and not false positives..."
Sal Daher: It's like a very fine tooth comb picks up all the nits. The specificity relates to how many of those things that you pick up are actually what you're looking for and not false positives.
Richard Vlasimsky: Correct.
Sal Daher: The circulating DNA, the CT, the cell-free DNA liquid biopsies, by the way, cancers give off cancer cells, circulating tumor cells, CTCs, they also give off fragments of cells, which contain fragments of DNA. There's a well-developed industry for detecting cell-free DNA, these DNA fragments that are out there, and doing a lot of number crunching and coming up with theories as to what's happening in the patient's body. If some type of DNA is present, it means that it's there. It's for real, so it has high specificity, but the coverage is not there because you're getting strands, pieces of DNA, that you may not catch enough for the algorithm to recognize it as cancer DNA.
The biopsies have that failing, but your algorithm complements it really well because it's really good at that and what it's bad at. It's very good at picking up nits, but your algorithm also picks up dandruff, sorry to say, disgusting
[laughter].
Sal Daher: I was looking for nits, but I got a lot of dandruff. The liquid biopsies pick up all the nits. If you use those two together, you can know pretty sure that if positive on both your algorithm and the liquid biopsy, it means that's a serious lung nodule. That could compete with the CT scan, the computer tomography scan in terms of-
Richard Vlasimsky: Absolutely.
Sal Daher: -predictive value. That gets us to a conversation about-- We talked about who would pay for the service. Who would buy you? What are the exit possibilities of IMIDEX? It's embarrassing. You're supposed to ask embarrassing questions.
Exit Possibilities
Richard Vlasimsky: [laughs] Like any venture-backed company, we're not for sale, but we're always for sale, but there are multiple exit potentials. Now, obviously, big tech is making big forays in the space. Microsoft, Google, there's also the X-ray manufacturers, so GE, Siemens, there's also the PACS provider, so the PACS is short for Picture Archiving Communication System and that's where all the radiologist workflow is orchestrated.
Sal Daher: That's a new one on me. I wasn't aware of the existence of this sector of the economy, the PACS.
Richard Vlasimsky: PACS and there's 30 PACS providers out there. GE provides a PACS, Change Healthcare.
Sal Daher: What's PACS again?
Richard Vlasimsky: Picture Archiving and Communication System.
Sal Daher: Picture Archiving and Communication System, cool.
Richard Vlasimsky: It stores all the X-rays, MRIs, CT, ultrasounds, and all the medical imaging and it has workflows where it'll send the cases to a particular radiologist for interpretation. The liquid biopsy companies, potentially.
Sal Daher: Perfect fit?
Richard Vlasimsky: A perfect fit. Big pharma, especially the AstraZenecas, and the rushes that are putting forward cancer medications, so we will.
Sal Daher: If they have a treatment for it.
Richard Vlasimsky: Drive dry volume.
Sal Daher: Yes, they need dry volume of people to have early detection to have very high cure rates, and for sure, they're going to want to have this in their clinical trials to make sure that they're having the results that they need. Excellent. Is there anything else that you want to say about IMIDEX in terms of the business and in terms of the problem it's solving or who might buy it and so forth, that kind of thing?
Richard Vlasimsky: I would just say I get a lot of energy from the business by the people that we've surrounded ourselves with and so I took some sagelike advice from a very successful entrepreneur operator by the name of Stan Lapidus.
[laughter]
Sal Daher: Who was giving a speech and his advice was twofold. It was to young entrepreneurs such as me, although I'm old for an entrepreneur and his two bits of advice were surround yourself with the shadow team, your network is going to be very important to you.
Sal Daher: Yes.
Richard Vlasimsky: You got to punch above your weight. After he gave his speech, a whole line of people with cards lined up and I knew I wouldn't have a chance to get him to remember me.
Sal Daher: Tell the audience who Stan Lapidus is.
Richard Vlasimsky: Stan Lapidus is the inventor and founder of both Cytyc and Exact Sciences. You probably know the tests better. Cytyc still enjoys dominant market share with their thin-film pap smear test for cervical cancer and Exact Sciences is most known for the Cologuard test for colorectal cancer.
Sal Daher: Two very high-value tests. The pap smear it's a low-burden, high-value, very important test.
Richard Vlasimsky: Absolutely. I called up Stan Lapidus the next day after his speech. I said, "I'm punching above my weight. I'd like to meet with you."
[laughter]
Richard Vlasimsky: He said, be it building five at Anschutz Campus at 3:00 PM and I was there.
Sal Daher: Okay.
Richard Vlasimsky: We had a funny first meeting because he double booked me and in that meeting, I told him I would like for him to be on my board of directors.
Sal Daher: Awesome.
Richard Vlasimsky: He said, "I'm already on six, thank you." I'm over-committed and I was persistent and stayed on him and he eventually agreed to observe on our board and has been an amazing mentor. Our whole playbook is he's been there, done that twice and so just a great mind strategic thinker and that's just-- Stan's like Mount Rushmore in cancer diagnostics but I just enjoy that. That's one of many people that I get to interact with all the time.
"... his definition of a startup is an enterprise that uses resources that it does not yet command..."
Sal Daher: That was really awesome. This calls to mind the definition that Howard Stevenson, he's America's now but he is an early professor of entrepreneurship at Harvard Business School and his definition of a startup is an enterprise that uses resources that it does not yet command. Basically, shoestringing it, being able to do stuff without all the resources that the large well established enterprises have, figuring out ways to be super efficient which is what I see IMIDEX doing here.
Richard, if you wouldn't mind, what I'd like to do right now is I'd like to do a very brief call to action for listeners to help get more people to find this podcast, particularly this interview with you, and also there's a call to action to you as well, and to people in your company to get this episode found but also after this, I want to get into your previous career as a founder because I know Keith told me that you're an exit founder and that you have a very personal reason. It's a passion that you have for this mission. It's not just any cancer. It's a particular thing with lung cancer and you have basically found a way to make a dent on lung cancer because of a personal connection.
First, I just want to say that if you find this really, really compelling conversation with this tremendous founder, Richard Vlasimsky, do us a favor. Go to the app that you use to listen to this and follow this podcast and leave a rating and a written review. Richard, do the same thing. If you're on the Apple podcast app, leave a review and a rating. We like to ask for five stars. We dare to ask for five stars because our guests are all five stars. I may be in three and a half, but the guests are all five.
That would help us get found and get people to learn the model of things that can be done with miracle methods these days and with a lot of grit and spunk, which is what Keith described you as a person who just like you went at this with unbelievable determination, unbelievable discipline, just didn't leave any doors unknocked on. You were just methodical and relentless and tireless. If you want to promote to other people the examples, like Richard, I would be very grateful if you would do that. Anyway, Richard, tell us, how did you get into entrepreneurship?
Richard’s Entrepreneurial Journey
Richard Vlasimsky: I would say I started thinking about it in the mid-'90s. At the time, I had been working for Andersen Consulting and I was seeing a lot of my technology colleagues jumping ship to startups in the dotcom era and becoming instant millionaires. I had to sit a bit of an allure to that.
Sal Daher: Little did we know. [chuckles]
Richard Vlasimsky: Little did we know [chuckles] what would eventually happen. My last stint with Andersen, I helped with a partnership with an integration software company that George Shaheen, who was the managing partner for Andersen Consulting, sat on their board. That's where I started to really get the bug.
I met with a gentleman by the name of Andre Boisvert who was first name basis with Bill Gates and Larry Ellison. I was like, "This is what I want to do." I moved to San Francisco, joined a startup that I didn't co-found. I learned a lot about what not to do, I would say from that, just viewing from afar.
In early part of 2000s, I co-founded Valen Analytics with a childhood friend and consummate entrepreneur, who's actually also on our board so lifelong friend. I was really into machine learning. I was attending Stanford and Berkeley's seminars on this. They just had outstanding minds and individuals and I saw that there was a future in AI, at that time, I wasn't quite sure what industry. We ended up choosing financial services as the place we were going to start. We were very early.
Sal Daher: This is when?
Richard Vlasimsky: 20 years ago.
Sal Daher: 20 years ago. 20 years ago you had a machine learning startup addressing aspects of the financial services industry. I understand insurance.
Richard Vlasimsky: Insurance underwriting, so helping insurance companies' price to risk. I would say we were extremely early in the technology adoption curve. 20 years ago, it's kind of funny, AI was a dirty word. It was like you did not--
Sal Daher: A couple of false starts by then already.
Richard Vlasimsky: [chuckles] We were a predictive analytics company outwardly, but nothing changed underneath the covers. We built the company up. We ended up selling it to Insurity Corporation, and after that exit, I was reflecting on next chapter of my life. I wanted to do something that would have a social impact. I had at the time a former data scientist that had worked for me, and then a colleague that was a partner both struggling with cancer 1. With putting that together, with the mission to save lives, and then looking at that lung cancer, in particular, as the biggest cancer killer in the world, every 17 seconds someone dies of it. It's basically a 747 crashing daily.
Sal Daher: It's horrific.
Richard Vlasimsky: It doesn't have to be that way. It's half the people diagnosed stage 4. There's only 20% of the people are diagnosed stage 1 and your chance of survival there is pretty good, 80%, 90%.
Sal Daher: If everybody had a chest X-ray, they would've picked it up early, removed it, the nodule, and that's it. So much suffering.
Richard Vlasimsky: It is.
"... You had this experience of joining a startup, then creating your own startup which was using machine learning..."
Sal Daher: That's the motivation. You had this experience of joining a startup, then creating your own startup which was using machine learning. Unfortunately, the gamers hadn't yet gained enough for them to develop the imaging chips that [chuckles] make real machine learning possible. I always thought video games have just been a waste of time. All this brains being wasted and all they were doing is they were providing the market that would subsidize the research into the graphics chips that would then make machine learning possible. [laughs] Technology is amazing.
Richard Vlasimsky: 20 years ago at Valen, the tools were nowhere near where they are now in terms of AI. There was no such thing as a GPU. We had to build a whole infrastructure of distributed computers where we would have 30, or 40 computers at once solving on the same problem and putting it all together. It would take days and days and days and days to solve. Now, with the GPU, and this is nowhere even near the level of processing.
Sal Daher: Richard, unpack that a little bit for laypeople. CPU is the central processing unit in the computer. It runs the operating system basically, and pulls that stuff off memory, but then there's a specialized graphical chip, the GPU, which handles the attractive graphics that were needed for video gaming to become successful.
Richard Vlasimsky: Right. The big difference in the architecture is that a CPU, you'll hear it will be eight-core CPU or something like that. There's not a whole lot of processors, but the processors can do very complicated operations. GPUs, they basically have very simple processors that are essentially calculating one processor per pixel. You have in some GPUs, 32,000 processors. Underneath the covers of computer vision a lot of deep learning are neural networks where a CPU would have to solve that in sequence. Through backpropagation, a GPU can do it all at once.
To try to solve what we're solving right now in a CPU, GPUs, when we fit models will take about a day. On a CPU it probably would take a year, maybe even more.
Sal Daher: That's the first revolution of solving things that would take forever. You solve it altogether. Now we're doing it with large language models ingesting instead of what a GPU used to do. They do many, many times that all at once. That is only expanding. What will these large language models imply for the type of model that you work on in the future?
Richard Vlasimsky: It's a good question. It's something that we had talked about. I will say that the large language models, I think, have a place in terms of summarizing the patient's medical record. It's scattered all over the place. This is the patient, this is all their labs. Give me a summary of that. That could be potential additional indicators that one could use to fine-tune models such as ours. I see that there's probably a big application in that space.
Sal Daher: How about learning from hundreds of millions of patient records? This is what the diagnosis was, this is what actually happened over history. What if a large language model was set loose on tens of millions of patient records?
Richard Vlasimsky: It's a good question. The big question I have, and I think large language models are amazing. The big question I have is the random nature that they have that they don't produce the same result given the same question over and over again. At least in the near term, I'm dubious that they would be used in any way that would affect patient care for that reason. That's why I think things like summarizing the patient's medical record, I think there's great utility in that as well as perhaps triaging the patient. I've got a chest pain and it comes back and asks questions to determine if they should go to the emergency room or not.
Sal Daher: Right. Did you eat pepperoni pizza?
[laughter]
Sal Daher: First question. How many cups of coffee did you have this morning? Did you have pepperoni pizza? Excellent. We've touched on the reason for starting the company, how you doing it, where you are now, and then your inner motivation, your entrepreneurship journey. As we think of wrapping up this conversation, are there any thoughts that you would like to communicate to our listeners who are maybe two-thirds of them are founders, one third are angel investors? What thoughts would you like to leave to them?
Advice to the Audience
Richard Vlasimsky: I think for the founders, I would say two words. Never quit. You're going to have lots of challenges. The difference between a successful startup and a failed startup is the one that quit.
Sal Daher: Yes. [chuckles]
Richard Vlasimsky: When I was in college, my senior project, I worked on a solar car and one of our donors of all things made it really big. He had patented and invented the twist tie for garbage sacks.
Sal Daher: [laughs]
Richard Vlasimsky: He had MRI machine that would help Jimmy Dean sausage cut out cartilage when it was--
Sal Daher: [laughs] Oh, cartilage in my Jimmy Dean sausage.
Richard Vlasimsky: -unsolicited, I always remember this and he said, "With patience and perseverance, you can pierce through a stone," and so that's my advice to founders, keep at it. For investors, if you want to be part of something that can make a big social impact, look us up, reach out to us.
Sal Daher: This could not have finished better, this interview.
[music]
Richard Vlasimsky: Greatly enjoyed it, Sal. Grateful for the opportunity. The timing is awesome. We just got our 510(k) clearance so we're coming out of stealth mode. I sincerely enjoyed our conversation. Great questions.
Sal Daher: Same here. Richard Vlasimsky, I am very grateful to you for making time to be on the Angel Invest Boston podcast. Thank you.
Richard Vlasimsky: Thank you.
Sal Daher: Thank you for listening. I'm Sal Daher.
I'm glad you were able to join us. Our engineer is Raul Rosa. Our theme was composed by John McKusick. Our graphic design is by Katharine Woodman-Maynard. Our host is coached by Grace Daher.