Rx_Michael Abramoff: Audio automatically transcribed by Sonix

Rx_Michael Abramoff: this mp3 audio file was automatically transcribed by Sonix with the best speech-to-text algorithms. This transcript may contain errors.

John Shufeldt:
Hello everybody, and welcome to another edition of Entrepreneur Rx, where we help healthcare professionals own their future.

John Shufeldt:
Hi everybody. Welcome back to Entrepreneur Rx. Now, I'm really excited to talk with Dr. Michael Abramoff, who is an MD Ph.D. He's a founder and executive chairman of Digital Diagnostics, which is an autonomous AI diagnostic company. And they were the first to get FDA approval to use its autonomous AI. Michael, welcome to the show.

Michael Abramoff:
Very nice meeting you and very, very great to be here.

John Shufeldt:
Excellent. All right. So I always start off with this why medicine. How did you get into medicine?

Michael Abramoff:
I was actually always going back and forth between engineering and medicine when I was in high school. So as you can hear from my accent, I was trained in Europe, did my medical school there, and actually started doing an engineering degree, decided, well, there's not enough interaction with humans or with patients, decided to medical school and then found out, well, I actually need to engineer the more exact, precise part of engineering. So essentially I did a master's in computer engineering, what's call it now, as well as medical school. For a while, worked in neuroscience, so really mimicking the brain with a computer, this was in the late eighties. This was my master's thesis actually. Later decided, now I do want to contact with the patients, went back to do a residency, fellowship of retinal surgery. But even during my residency I managed to do a Ph.D. in AI, machine learning. So I've, I've always tried to combine the two, that was hard. People always say, well, you know, it's a great combination, but it wasn't, really the only jobs that were at the time was in EHR, you know, why didn't you write an EHR? Well, that to me is not particularly exciting. So finally decided, well, you know, I've been mimicking the brain of the computer so long, why don't I use it to actually benefit patient outcome? Because I was seeing all these patients who came in for what is a diabetic eye exam too late, when there is also permanent damage and there's ... to do or too early. But there was really nothing wrong with them. And I decided, well, a computer can do a better job here, make it easier for these patients to get the care they need.

John Shufeldt:
Now, I was reading something relatively recently and they said people who have pre-diabetes, by the time you're called pre-diabetic, you already have a lot of damage done to your microvascular system. Is that true?

Michael Abramoff:
That may be true in some cases, especially in type two diabetes, you know, because we don't exactly know why, when it starts. So by the time it's diagnosed, you know, even as pre-diabetes, there may already be and has a long time of hypoglycemia. And what I think, by the way, I think is oral neural components of diabetic retinopathy and diabetic complications in general. And this neural component also starts developing as a result of most varying glucose level as well as hypoglycemia.

John Shufeldt:
So when did you get interested in the hypoglycemia effects on the retina? So was that during your fellowship, the retinal fellowship?

Michael Abramoff:
During my residency, no. I used to have, I really see AI as being applicable for two things, for research. We're discovering new mechanisms of disease as well as for managing patients and improving patient outcomes. In my mind, these are very different. So but for a while I was combining them via the large research group funded by R1s, making new algorithms for analyzing retinal images, various types. And that allowed me to discover that there's this big neural component, neurodegenerative component to diabetes, both in the brain and in the retina. And then the other one is what I call autonomous artificial intelligence, which is essentially the AI making a medical decision, which is what you just talked about, where we worked with FDA, would see a mesh with essentially all stakeholders in healthcare to get it to patients, and now, it's now seeing improved outcomes in patients.

John Shufeldt:
Because then you're seeing improved outcomes really because of its early diagnosis, diagnostics. You can look in this machine and the machine can basically predict using it using AI. Yes, this is a precursor to what will become diabetic retinopathy.

Michael Abramoff:
Well, it's exactly, it did diagnosed as diabetic retinopathy, and another component, which is called diabetic macular edema. What is important there is we already know how to treat these patients once they have it. So what we are proving and even randomized clinical trials now, with one arm is traditional way of screening people. And either way using the AI, you just see better uptake, better compliance, better referral rates. And we already know that if patients get to the care they need with a provider, with an eye care provider, ophthalmology is retinal specialist, then they have a better outcome. There's no, you don't need to prove that over ten years. You can just prove that they get to the care they need and the treatment they need.

John Shufeldt:
In as a treatment, at least early, tighter control of glucose?

Michael Abramoff:
Very important glucose levels, you know, diet, you know, losing weight, things like that. That's the early stage of what we call the metabolic stage where you have more metabolic treatment and then eventually you get to a more advanced stage. If that is not enough, where we have very good treatments, especially in the last 10, 15 years with what you may have heard about anti-vascular endothelial growth factor and treatments for that, right? They really make a giant difference.

John Shufeldt:
In, you know, I probably get a little bit out of my skis here, so I have to back up and talk to you about your entrepreneurial journey. What is a neurodegenerative component of this that you recognized early?

Michael Abramoff:
It's interesting that, as you know, there's peripheral neuropathy and diabetes that is well known. What we didn't know that is it was, well, it's interesting if you go back in history, 150 years ago, people realized that there's, there's a vascular component in the neural component to diabetes in the retina and in the brain. You can see it just by looking at stains from retinas and from the brain. But that was sort of eclipsed by all the focus on the vascular component of diabetic retinopathy. So for many, many years, all outcome measures, all endpoints were about this vascular component, which is macular edema, new vessels growing where tissue to grow, vitreous hemorrhage, all these things that blind people, the neural compound that was sort of not really fully taken into account as endpoints in research. We didn't exactly know what to do with it. And then we were able to show in what I call very elegant studies both in humans and mouse models that even before there's any vascular change in the retina, you already have neural loss. And so that means that unlike what we thought before that yeah, sure, you have ischemia, you have other vascular changes, yeah, obviously neural cells are going to die. It actually turned out that the neural cells were dying before you saw any vascular changes. And that sort of made us think a little bit differently about diabetic, what we now call diabetic retinal disease. And I'm actually part of a big group that's revamping our severity scale for diabetic retinal disease. This has nothing to do with the entrepreneurial activities, of course, this is really the research, and ultimately, it will lead to better outcomes, better patient treatments, but that's years away.

John Shufeldt:
Do you think that using this technology we can have an earlier diagnosis for people who are classified as pre-diabetics? In other words, you better start jumping on this now because we already see the changes in your retina. In other words, is this a canary in the coal mine?

Michael Abramoff:
It may well helpful, but I'm not going, like you say, over your skies to, to claim too much here. Let's call it interesting right now, and it has potential. But, you know, we are years away from using this as a treatment, as a diagnostic that is much more obvious with autonomous AI, where are we talking about diabetic retinopathy. Diabetic ... where we know exactly what to do.

John Shufeldt:
When did you first make this leap? Because my sense is when you actually made this leap, you were way ahead of the curve as far as other people thinking about how AI could impact the physician's diagnostic acumen. When was that aha moment for you?

Michael Abramoff:
I think AI in healthcare is from the sixties, right? And so it's this thesis being written about my sinful and AI that helped prescribe antibiotics. It just, people still use teletypes at the time. And so it was exciting that you could type a diagnosis into a computer. So most of the excitement was about that rather than the cognitive parts of the say. So that died a slow death. Then in the eighties, that was the time when the first neural network started one layer, maybe two layers, I worked on there, the computational capacity just wasn't there and the data wasn't there. So we worked on these simple neural networks, and there was another way where people were trying to make diagnoses or at least help in the clinic, that never went anywhere either. And I think the more, people talk about deep learning and about new algorithms and about GPUs, I actually think that the biggest reason why we're now successful is that we have way better data, digital data, digital sensors that have high-quality data. Before we sent these algorithms with pretty noisy data coming from clinicians, right? For me, listening to a patient and then typing them in and all the translational noise that you introduce that way and then you never get a good performance. But now we have very reliable, objective data in radiology and ophthalmology, in many other fields, and that's why we get so high performance, and now people are actually comfortable patients. Doctors are comfortable with using this in a clinic without even any human oversight.

John Shufeldt:
What other applications does your current technology have?

Michael Abramoff:
So these are very limited but large scale. So this is not a what you call general intelligence, which makes all sorts of analyses and differential diagnoses, this is more about routine care, especially in chronic care. So we have one for diabetic retinopathy, most important cause of blindness, easily preventable if caught early, we're doing it now. Another one is for diagnosing melanoma, squamous cell, basal cell carcinoma in the skin. We're also relatively straightforward in terms of what you're automating, no large scale. I mean, many people need this. You don't want to build an AI, you know, with millions of dollars for a disease that only maybe a few thousand people have. You know, orphan diseases like that are not the lowest hanging fruit for especially autonomous AI right now. Sure, for research, yes. But not for improving patient outcomes and commercializing this.

John Shufeldt:
But when I was back in law school, I remember reading the article and the professor talked about this, that when Big Blue came out, Big Blue was able to diagnose pancreatic cancer by looking at a retinal scan. Do you, do you ever hear this? They never, they could never figure out why it could do it, but it did it with very high accuracy.

Michael Abramoff:
Yeah. So we are getting into explainability and black box algorithms and understand what they do. If the, yes you can, ok, the retina represents about, you can look at the vasculature and you can look at the neural tissue, you know, without anything but using a finished camera. So that's exciting because a lot of systemic diseases are represented in the retina, all the way from Alzheimer, MS, cardiovascular disease, stroke risk, etc.. So yes, there is a lot. You can see just from the retina because you see the tissue without any radiation. But it doesn't mean I'm convinced that I can diagnose a very specific disease like pancreatic cancer from the retina. Sure, there may be changes related to we should change in the retina because someone has a malignancy and there's metastases. Yes, that, you know, but it's pretty specific, right? Yes, we can see whether someone is at increased risk of stroke, but it doesn't mean I can tell you next week you're going to have a stroke. So the specificity, I don't think is there yet, but I'm excited to look into their algorithm if they're willing to share it with me. But I would be a bit skeptical when we don't know the mechanism, why the retina tells you that very specific thing, because I can explain you why we see Alzheimer, I can explain to you why the vessels look different in diabetes, in someone with cardiovascular disease, but not with this specific disease. I wouldn't know what the best genomic feature would be.

John Shufeldt:
Yeah, it just strikes me. And in his point was that, you know, they went literally millimeter by millimeter across the retina and they cannot figure out what this algorithm was picking up that, that was able to diagnose pancreatic cancer with such a high degree of sensitivity, so that was interesting. All right, switching topics, entrepreneurism, when did you all of a sudden go from, wow, this has commercial applicability and I'm going to make this leap because you could have just sold the IP, I suspect, but you made the leap of faith, as I ... do it myself. That's a departure.

Michael Abramoff:
Yeah. So you're thinking of pharma, and I thought the same way, you know? With pharma, you fold, you find a new way of folding a protein and someone came in, you pattent it, and Pfizer licensed it and off we go, right? And that's the very traditional pathway for a physician-scientist. With AI that didn't exist. And so I thought the same way. And I was actually I was told, well, what I thought naively, if I just do enough scientific publications that this works, people will pick it up and see obviously we, you know, how, this is how we can improve patient outcome. Nothing happened. Then I heard, well, you need to patent it, like you said. And then, you know, there's a whole system in academic centers where there's a patent mill. And so I now have 20 patents, but no one came in to pick up the patent and commercialize because there was little there was no market, there was no industry, there was no reimbursement, there was no, FDA didn't know what to do. So then people said, well, philanthropy is the next solution. And that again, works great if you have a protein, a new way of folding a protein that lowers death in children. People love donating money to that. Here, it turns out that the biggest hurdle was FDA, and FDA just requires mostly to do reams of paperwork of tens of thousands of pages of paperwork. You know, donors, philanthropists are typically not excited about donating for paperwork. You mean, you will proudly show another pile of paper, they don't get excited about that. So philanthropy was out, I realized, and that means that the only way to get this to patients, which I care about, you know, I'm not a scientist for the science, I'm literally a scientist for getting as a physician a better patient outcomes. I said clearly I need to do it myself, look for angel funders and network and get people excited. And that was enough to raise another 20 million to start talking to the FDA. And that was the big step because people at the time were thinking, you know, this is never going to work with the regulatory framework we have, we'll just break the system. And there is a big company on the West Coast that we're all familiar with that essentially tried to do that. And I said, no, I want to work within the system, I'm a physician, I think this can work, let me work with FDA. So I went to FDA in 2010 and said, hey, I want a computer to make a diagnosis. They said, oh, no way. And so, you know, a very collaborative ..., appeared into my life where we developed, how do you we do a clinical trial? How do we think about liability, right, to medical liability? How do we think about racial and ethnic bias? How do we, all these aspects, what I now call an ethical framework, needed to solve to make FDA comfortable with this concept of a computer making a diagnosis without human supervision. They were not new to AI, there were many radiology AIs that were being marketed, but they were all assisted. They were always, hey, here's a radiologist. There's an AI to look at the same scan and look at the same mammogram. And now, and now, supposedly, and sometimes yes, sometimes no, the diagnosis becomes better or more specific or whatever. They were familiar with that. But that is a ... of AI, this autonomous idea was very new, but I think in my view, it is necessary to get to patients where they are, which is, you know, in many times and with health inequities especially, it's very important to go to where patients are rather than have them come to you, which they don't in many cases. And so I think that, that was a big step for FDA. I'm very grateful for the collaborative way we have been working together since, and now I work very closely with writing papers about ethical frameworks and how this can affect regulation. But that was a long road because it was not enough to just get FDA de novo authorization, as it's called, because many more steps were needed to get it to patients, which you can get to if you ask me.

John Shufeldt:
So knowing what you know now, would you would you do it again because you were the trailblazer in this one and that had to be the hardest road, the hardest road to hoe.

Michael Abramoff:
I don't know. You know, looking back, it needed to be, someone needed to do it. And so it just happened to be me. That is not any big accomplishment. It just, you know, I, I wanted this to happen. And so I went ahead where maybe a smarter person would have said, no way, let me focus on something else now. And so I. You know, it was a lot of fun and we achieved a lot, I think, so it's fine. I don't want to look back too much and say, well, you know, I had not done this, what would I have done? It needed to happen.

John Shufeldt:
Yeah, it definitely needed to happen. And there definitely needed to be someone like you to do it seriously because you've opened it, now you've opened this pathway for the diagnosing melanoma, for all of the things that autonomous AI could do where we don't need a physician or at least improve the outcomes.

Michael Abramoff:
Well, we improved the accessibility, right? And now those patients that need a physician, a specialist, they can get it. Try to make an appointment to a dermatologist, right? So six months away. Now you say, oh, primary care can diagnose this, now the dermatologist knows that that patient needs to care. The same for me as a retina specialist. I know that AI is referring those patients that need my care and not the ones that there's nothing I can do for them because they don't need me.

John Shufeldt:
Yeah, interesting. So you raised 20 million early on, was that 2009, 2010 era?

Michael Abramoff:
No, no, no. Are we talking to. Yeah, 2010, 2011 from Angels. And then ultimately as soon as you got FDA approval, now VCs become interested, right? So we raised another 30 and that was to bring it to market because there was no reimbursement, there was no standard of care that allowed this. There was no quality measure that supported the use of AI, it was all, every rule, every regulation said you need a physician to do this, in very unexpected places. So we needed to solve all of that. That means getting a CBT code, for instance, which I had never been done, never been considered. Getting HQA, will rise ... language to support the use of an AI, in autonomous AI, to close the care gap, will never be known, and then go and working with CMS to get Medicare reimbursement for insulin ... which had never been done and they did that last year. And then January one of this year, we got CMS reimbursement at around $55 dollars, which was very exciting because that is cost saving for them, but it allows so many more patients to get this because now primary care providers are excited because they can build for it, right? It's a big question, if you are a primary care provider or any physician, one of the first questions with the new technology is, can I build for this even in a value-based care environment?

John Shufeldt:
Well, they didn't sugar out it early on and start as an assistive tool and then work your way into an autonomous tool as opposed to going right for the, swinging for the fence out of the gate.

Michael Abramoff:
So I saw others drive it or shade it or think that way. It's very tricky because there is, you become so accustomed to assistive you never need to solve these breakthroughs because assistive is familiar and I use it in my clinic. But you didn't need to create a reimbursement, you didn't need to create, all the things I mentioned that didn't need to change thinking, so I think the only way was to go straight for autonomous. Otherwise, I see many, many others, researchers, companies do that and they, they think they may ultimately go there. It's very, very hard because you have, you have an existing time, right? You have an existing successful revenue and now you would take a giant risk. And then it's just very hard once you are in there. And... And so because from the beginning we said this is going to be autonomous, we are going to do this every, many people push back, right? You may have heard my nickname is the retinator. And so my dear colleagues in the biggest ophthalmology journal of Ophtalmology Times, the chair of ophthalmology at Hopkins in 2010 that a big editorial, the retinator, revenge of the machines because this .... was going to cost jobs, a lot of quality of care. And looking back, it was great because it made me realize that you can look at it from an engineering perspective, hey, let's solve this, right? Let's make it work. But that's not enough because people have concerns, physicians have concerns, patients have concerns, ethicists have concerns, ... have concerns. And you need to solve all of that because, yeah, you can solve the technology, but if you don't solve the, the societal aspects, the ethical aspects, the concerns, then people will still not accept it, then they'll use it. And so it forced me to deal with that, which was very healthy. But at the time it was, it was brutal. Yeah.

John Shufeldt:
Well, I mean, the retinator was pretty, pretty damn good. That's not a bad nickname. So, I mean, you're a professor of ophthalmology, University of Iowa. What do you see? This is why mentor a lot of these med students, pre-meds, and I always tell them, I said, you have to look at the impact or how AIs can impact your particular field as you're going through it. So I'd say I'm not sure I'd want to be a pathologist or a radiologist unless it was interventional these days, because I think it will be the retinator taking over a lot of what they do today, where do you see medicine going? What specialties do you think are at risk for really having a sea change in the way they and the need for physicians?

Michael Abramoff:
I like to give a positive twist on it. I think the specialties that both diagnose and also treats, like you say, interventional or otherwise, I think there's a lot of promise there to typically there's not enough of them, we're not doing enough in, you know, there's many underserved communities, many underserved people. So moving it out to primary care and other places where the patients are as close to the patient as you can be, the least space for actually treating more patients by specialists like me and, you know, and the people you mentioned. So that's what I tell them. I tell them, you know, the more, we will see a shift towards more treatment, more interventions and less low yield diagnostics, I think that's where the biggest bang for the buck is.

John Shufeldt:
Yeah. So if you're a specialist, you're going to have patients that really need your, your expertise as opposed to.

Michael Abramoff:
Top of license, tope of license, better patient mix. So I mentioned this, the retinator and you would think that ophthalmologists are still concerned about this, they're the biggest supporter.

John Shufeldt:
I bet.

Michael Abramoff:
The American Medical Association, American Academy of Ophthalmology was the strongest force in creating this ... because they see the potential for indeed getting more patients that need their expertise. So we turned it around. But again, if that retinator thing had to happen, you need to deal with the concerns that are out there and be frank and open and very transparent about it.

John Shufeldt:
Yeah, that's amazing. Well, Michael, where can people learn more about you and more about Digital Diagnostics? Because you're going to be, and I think people are going to look at your story and look at you and think, I want to be, I want to be like this. I want to change the world.

Michael Abramoff:
Yeah. I always happy to help people on their on that path and they see it on a LinkedIn page and on DigitalDiagnostics.com, on the company website. And I also have of course, a faculty web page at the University of Iowa. I'm easy to find, I think I have a Wikipedia page that I do not do that, but it's there.

John Shufeldt:
Well, thank you very much. This has been really entertaining and really insightful. And I said, I love the combination of AI and medicines. I've been so looking for this discussion, thank you.

Michael Abramoff:
Thanks so much.

John Shufeldt:
Thanks for listening to another great edition of Entrepreneur Rx. To find out how to start a business and help secure your future, go to John ShufeldtMD.com. Thanks for listening.

Sonix is the world’s most advanced automated transcription, translation, and subtitling platform. Fast, accurate, and affordable.

Automatically convert your mp3 files to text (txt file), Microsoft Word (docx file), and SubRip Subtitle (srt file) in minutes.

Sonix has many features that you'd love including powerful integrations and APIs, advanced search, world-class support, collaboration tools, and easily transcribe your Zoom meetings. Try Sonix for free today.