How AI is Revolutionizing Healthcare Products and Access with Sam Dribin

For the first time in healthcare, human ambition and technology are starting to intersect and AI is one of the main drivers. Listen in on this episode with Sam Dribin to learn about how AI is evolving to improve healthcare results. 

 Sam Dribin is the Chief Technology Officer for CureMetrix, an AI company focused on solutions that support radiology.

Sam has a diverse background in technology including Biotech software development with expertise in the technical issues facing healthcare and life science companies and laboratories.

For the last five years, he’s been on the forefront evolving artificial intelligence with solutions that are focused on AI-based mammography to improve cancer detection. 

We’ve known Sam and CureMetrix for some time as a customer of Qualio, but today he shares his insight into how AI can impact healthcare such as reducing false-positives and increasing access to healthcare everywhere.

Show notes: 

• https://curemetrix.com/
• Sam Dribin's LinkedIn
• Radiology Society of North America

Music:
keldez

Qualio website:
https://www.qualio.com/

Previous episodes:
https://www.qualio.com/from-lab-to-launch-podcast

Apply to be on the show:
https://forms.gle/uUH2YtCFxJHrVGeL8

Music by keldez

Transcript

Robert Fenton: 0:20

Sam I am incredibly excited to get the chance to chat with you today. I know you've been interacting with some other members of our team. I'm a big fan of you and everything you're you and the team are doing a CureMetrix. And I'm really excited to kind of dig in a bit deeper and actually get the chance to ask you a bit about history today. So I really appreciate you joining.

Sam Dribin: 0:42

Likewise. Thanks for having me. I appreciate it. Nice to meet you in person.

Robert Fenton: 0:45

Yeah, it is. It is. Um, I I'd love to know how do you folks talk about what you're doing? I don't think that's a really great starting point because looking at your website, but in your own words, I'd love to hear how you, how you're building CureMetrix to impact the world.

Sam Dribin: 0:59

Sure. So, um, we are an aI Technology company for medical imaging, um, specifically focused on women's health. Um, so our mission is to really help the radiologist, uh, improve the accuracy and detection, um, in classification of anomalies in mammography. So our mission is to save lives and uh, support better clinical decision-making, financial outcomes for hospitals and clinics. Um, you know, and we're really not replacing the radiologists. We're really focusing on support services and, and trying to make, to improve outcome.

Robert Fenton: 1:31

where's the business today?

Sam Dribin: 1:33

Um, busy. Um, so there's a lot of backlog. We actually just had our, um, the major radiology conference, which usually takes up the McCormick center in Chicago. Just went virtual last week. Um, so there's a lot of leads. There's a huge backlog as the clinics have shut down of, uh, you know, mammograms, it's, you know, they weren't super essential. Um, but they, you know, now that we're going on a year of the shutdowns, there's, there's a huge backlog. So we're really finding a lot of interest in using our AI to, to start processing that now.

Robert Fenton: 2:03

So are you saying, is that the, your technology can help manage some of that backlog? Because this is a big risk, right? Is yeah. Stuff is going on. People are sheltering for a lot of valid reasons, but there's a downside if you're not in the lab, looking at things.

Sam Dribin: 2:18

Exactly. And we focus our first cleared product was cm triaged, and it really focuses on that, that sort of workflow. So if you have a backlog of a hundred thousand, 10,000 images, um, or studies, uh, at the very least Everything needs to be looked at, um, but at the very least we could flag the studies and the patients that you know, pop them to the top and say, Hey, listen, you know, really we think someone should take a look at this.

Robert Fenton: 2:43

Yeah. And I just realized we're going pretty deep into the detail here. Maybe as somebody who knows how this works in the offline world, it might be worth taking a step back. So you folks applying AI to like medical image data, right. Fundamentally to recommend. Maybe you can talk about that. You focusing on CURE Metrix. How are people managing this today? I think this is illuminating for people to understand the, the non-technology driven methodology.

Sam Dribin: 3:12

So in terms of the technology itself or the, the,

Robert Fenton: 3:15

the actual process for trial like that backlog, right? So is these in radiology departments? How does this work?

Sam Dribin: 3:22

So technology, um, is essentially installed. We, we listened, we install a little service at the hospital or clinic, and it's called the DICOM listener and DICOM sort of the, the major imaging. Uh, technology that's used widely in radiology and it's also a transfer mechanism. So we install a, um, a service that just kind of sits there. And as the patient comes in and she has her x-rays taken, uh, we're essentially CC'd those images, digital copies of those images. Um, from there we anonymize it, encrypt, do all this stuff to, um, make sure that that PHS stays on ground. And then process it in the cloud. Um, so we transmit the images up. Um, we have solutions to scale up, scale down and what they call inference. So they applaud, they run our AI against these images. Um, from there, the results are sent back down through that same sort of route, um, back down to the hospital or clinic, the information is re associated and, um, depending on the product. Um, so our triaged product is displayed sort of as a work list. Um, so we say, okay, well this patient who came in today, we flag them as suspicious or not suspicious. And so that's at the, it sort of appears as a, uh, an element of almost a, uh, email inbox. It looks like that you can sort on, um, and how the radiologists or system decide to use that are essentially up to them. Um, some radiologist may decide to work on it first,work on it last um, come you know, have, have a cup of coffee and come back to take a look at it. Um, our seam assist product is, um, more heads-up so those same results are sent back down. Um, so that the radiologist, when she's after she's made her diagnosis, generally can push a button on the viewer and our results pop up as certain boxes, essentially, um, say. You know, and it confirms, um, diagnosis could say, okay, well, you know, Hey, I caught that or just bringing attention to something that might've been missed. Um, each box has a severity score associated with it. Um, so that, you know, the higher, the value, the more severe the, the, you know, our system thinks it is.

Robert Fenton: 5:32

Yeah. And before your product, I mean, if I, if I'm correct in saying this, this was pretty much entirely a human error detection system until product like yours.

Sam Dribin: 5:44

Yes, we are. Um, the, certainly the first triaged product past FDA, um, for mammography.

Robert Fenton: 5:50

Congrats on that by the way.

Sam Dribin: 5:52

Appreciate that, um, and our, the CAD products have been around in various forms, but we are one of the first, um, AI machine learning driven platforms.

Robert Fenton: 6:03

You mentioned that you have a triage product and an assist product. And I know that when you're getting into the stepping in with the human or a place of the human to go beyond informational purposes. There's just a ton of like, like minefields. I'm sure you're well aware of this right there, but I guess just take a step back. I mean, for people listening, right. AI and medical image data and, uh, yeah. That, that that's a pretty interesting space right now. How did you, how did you get into that from that? Wasn't a thought thought in CS right? 10 years ago,

Sam Dribin: 6:39

What's interesting is I'm actually, uh, I'm actually a biologist. Um, so I, uh, my background's actually in biology, so, um, I went to grad school, um, and for a PhD in neurobiology, um, and pretty early on, I realized that my interests are way more in the engineering behind the lab bench solution then than the experiments themselves, which you can take. Um, so I went to work for pharma in the pharmaceutical realm, um, and to, to really try to bridge that gap between, you know, biology and technology. And I, I really think that's kind of the future. There's this cross cross modality, um, you know, biologists who can program, um, uh, quality, uh, experts who know biology, certainly biologist easy, but a software. We're engineering side of things. Um, you know, and that's really what we, what most of us at CURE Metrix try to bring. We have physicists on staff, plenty of them who focus on on was essentially by all, you know, biological question, um, as well as doctors and, you know, the, uh, business people, business side and things like that.

Robert Fenton: 7:41

I think that's a pretty common thing. I think you'll notice just from my own. Reading is you look at new fields or new things that become really important. It's often initially pollinated by people that have these multiple modalities and an interest down in the middle. I know something that I heard recently, uh, is that. You can look at waves and trends apparently right now, and use some universities like Stanford. They're seeing the same number of applicants to bio related courses, as they are a CS related course. So if you look at, at that's going to increase the number and frequency of these collisions between these skillsets, I think it's, I think it's super interesting and exciting for me personally, doing what we're doing. Um, looking at the, at the other part here. So. For a lot of people we look at, okay, we can build an algorithm, we can test this. This is showing some interesting data. And for nonmedical purposes, you can kind of wing it with a X percent accurate and it'll get better over time. And it's often a novelty product, right. But if you're doing the cost of getting it right or wrong is very serious and you sounds like you've applied is similar technology framework, very different claims, right? Triage, aid, and assist aid. Could you walk through how you, how you figure that out? Because yeah. That does. That's an interesting conversation.

Sam Dribin: 8:59

Yeah. That's, that's kind of a dive in that. Um, well, so first off, you know, by claim everything should be looked at, um, you know, I think by the, no one on, on staff and a lot of, you know, I think AI companies in general, I in certainly the medical realm, I don't think any of us claim to replace doctors, certainly not in mammography. It's a very difficult thing that radiologists do. We're an assistive technology. Um, so we we're here to help. Um, so it's, it's not a job replacement, it's a, it's an assistive framework. And really going in with that mentality, I think is, is part of it, um, to say, you know, listen, we'll see, but we re here to help, the radiologist, um, in terms of the triage piece, um, honestly there was a predicate, um, that came, uh, became available to us and we took advantage of that. Um, and we realized that we could take our clinic, you know, The CAD products we were working on and leverage them as, as a triaged piece with a very fast, uh, regulatory submission timeline, uh, for in the, in the mammography space. Um, the, the CAD product is a longer haul. Um, you know, th there's, the clinical trials would go through and everything like that. So that's a little bit of a longer piece of that, but yeah, I think, um, having the knowledge, especially at the engineering level of some of these questions though, um, is becoming increasingly important in, in certainly the quality and in the engineering realm, you know, to really a lot of the discussions we have are, say, how are we going to frame this? How should we frame this product? How should we frame these claims? Um, you know, in the AI. So that, you know, we are honest with our selves in our regulatory path.

Robert Fenton: 10:33

How much experience did you folks have when it comes to the regulatory path? before you kind of got too far down this journey? Like, so in terms of the, the going down this path, right, you had this, technology, but I don't think he started off with the claim. I'm curious. How did you make sure that you picked the right claims and you were following the right regulatory path

Sam Dribin: 10:55

advice um, you know, we, none of us I've done this, I've done this in the past, the pharma side. Um, but I wasn't at this level of, of sort of running things. you know, certainly not from the engineering side. So we've had good advice. We make sure that we partner with very good, um, regulatory staff, um, yourselves included, Qualio Included has helped us tremendously with the QMS and, and getting started and everything like that. Um, but partnership is really critical. Um, we need to make sure that we're getting good advice, especially in the startup realm where you know, everything is, are you cleared? Are you not cleared?

Robert Fenton: 11:32

Any big lessons learned, that'd be worth sharing to people maybe coming out of, you know, a new bio program or CS with interested interest in biology.

Sam Dribin: 11:40

I think it's, internally you know, I can certainly say that, that it's important that as, you know, running a team, how important it is to run diversity of, people, of, thoughts of data scientists with developers, with the biologist medical business and certainly regulatory. So I think that having that cross modality knowledge is absolutely critical these days. I think a lot of people are coming up, uh, comfortable with software, uh, comfortable with computers. Um, you know, but to really try to have add some expert knowledge outside of that realm is, is a real value add. Um, and you know, like I said, I think choosing great partners is critical, whether it's on the hardware space, whether it's on the, software space, it's whether it's regulatory, you know, getting help. With our QMS rolling or, or submission, um, it's really important to find the right, right partners to work with. And you know, even though we're small, we have to constantly ask ourselves, you know, we spend a lot of times choosing the right partners for the jobs.

Robert Fenton: 12:38

Yeah. Um, that's wise advice, you know, speaking with you? It kinda sounds like you folks have really executed incredibly well or, or are there any. Unexpected challenges along that path.

Sam Dribin: 12:52

Um, yeah, certainly, we amassing data is always the critical piece in AI and it's doubly. So in, in medical AI, um, you know, these, we deal with movies. We deal with giant pathologies, um, slices. We deal with just. Even logistics of storage. We were on the scale of some movie studios with, with the amount of storage that we pick up. Um, we banked 4 million in growing images, um, that, that are within our company, anonymized and annotated by professionals. And that library is literally, you know, most of our, The cumulative job, um, for training, for validation, for testing the AI. So really that piece of it is, you know, I think starting out you think, Hey, we're tackling cancer. That's hard enough, but some of the logistics of, um, keeping our datasets clean and their GroundTruth, um, uh, clear, it's certainly been a problem that we've, you know, we've tackled very well. I think.

Robert Fenton: 13:52

Yeah. The data problem catches a lot of people. Did you find that getting access to data was very difficult, like, because it's protected and it's protected health information.

Sam Dribin: 14:02

Yeah. Take it partner by partner and know we aligned with various partners and institutions around the world. And to really balance the, you know, it's important to us to make sure that our AI generalizes to, across. Women of different ethnicities and locations and geography and it's highly variant and this one reason that mammography is one of the harder radiology modalities, um, you know, uh, breast tissues vary from image to image, even the same. Um, the same patient may be positioned differently in imaging, in imaging from year to year. So to really build an AI that generalizes to it has been something that, that we've really tried to, um, uh, it's been a real challenge.

Robert Fenton: 14:46

Yeah. And I'm curious, Sam looking at the performance data today, how does the the AI, your run, compare to the human, the human review, if you can share.

Sam Dribin: 15:00

Yeah. So we've run papers on faster reading times for radiologists, um, 30% reduction in reading time. Um, fewer unnecessary patient recalls, um, 63% reduction in false positives versus non AI solutions. Um, you know, allowing for more time for accurate diagnosis and, and a whole bunch of things like that. Um, so the numbers are really pointing, you know,

Robert Fenton: 15:26

false positive is huge. Right? And if you accept some of the data's shared that false positives cause such huge harm. And I think you will be able to correct me on this. I know some of the reasons why some preventative testing, like people getting breast checks to bond. The biggest risks is false positives in populations that want to wait until certain ages in a lot of conditions. 63% seems huge. That's very impressive. Where do you think this can go? I hate to ask you, where will this be in five years question and clearly you probably think about this a lot.

Sam Dribin: 16:01

Yeah, we definitely do. Um, you know, I, I think AI is here to stay. It's, it's hard to, um, argue that right now. I think, um, we are pretty much the, you know, it's, it's been a game changer. Um, we're accelerating diagnosis. Um, there's increasing momentum on the, on the consumer side. You know, um, to, to kind of get where there's a double read and things like that, international, whether it's the lack of mammography of trained mammographers, um, in certain areas, you know, and really it's, it's also, AI is also a game changer from the regulatory perspective. Um, you know, the FDA's is constantly dealing with how to, to advise companies on, on, you know, uh, the regulatory side of it. Self-learning, um, everything like that is, is a big issue on the FDA side and you hear, it's almost collaborative, a lot of the discussions that we've had with them early on, um, where they're looking for feedback from us, for what, what they should be doing, uh, you know, kind of a, an incidental level. So.

Robert Fenton: 17:01

That can surprise people. Um, the big Goliath that is the FDA. I mean, this is new for everybody, and I believe they've been pretty open, when they don't have the answers and are trying to figure it. Sure. Yeah. Yeah. You mentioned one thing I want to touch on for a second and personally, look at even our mission of Qualio. You mentioned access to some of these, like people want things better, faster, and cheaper all the time. You mentioned something about, well, this technology can actually help access this, the little, the, the, this. Expertise by AI. I'm curious. Did you have any thoughts on how that could shape in the future?

Sam Dribin: 17:38

I think, um, some of the things that, that I get most excited about is really international in that front. Um, and to really bring some of this, you, you know, the huge backlog in some, some countries we're right near the border of Mexico. There's a massive backlog of Images, um, through nobody's fault, it's just, there's a lack of trained mammographers there. Um, yeah. So if things like are triaged can flag early on, um, or if it's added to a van, um, a mammography unit that might present in a rural area, maybe we could close that loop so that, the patient won't be lost in the shop like you were saying. And to try to, to a degree, democratize a little bit more, some of the, the diagnosis that we're seeing internationally and the standard of care.

Robert Fenton: 18:25

Yeah, that gets me excited when I hear, I hear that a lot, because I think that's, that's part of their, the real value that I can add. I'm curious, Sam, we're kind of coming up today in end of time here, but I heard so far talking about choose the right partners, right? It's I don't think you say it often enough. Loud enough. You spoke about the data. We need good data, particularly diverse data sets. Um, What else? What else would you for people interested in the space in general, or if you were speaking to yourself before this journey, at Cure Metrix I mean, any advice or any resources you point people towards.

Sam Dribin: 19:04

For AI in general or for mammography?

Robert Fenton: 19:06

Yeah, I think it's, I think it's more how technology, particularly things like AI can be applied in healthcare. Right. And maybe it's medical image side. That's interesting.

Sam Dribin: 19:17

There's a whole bunch of topics and, uh, I think much of it is public of the radiology society of North America. RSNH, um, I believe it's rsnh.org, but we could check on that, um, after, but yeah, there's a lot of, and, and they just finished their conference, uh, virtual this year and there's, it's a major topic. Um, there's whole floors of it. Um, you know, in various modalities and, you know, talks given. So there it's a very active topic out there. Um, so I think that's a, that's a very good resource, to kind of start. Getting oriented in that realm. So.

Robert Fenton: 19:49

Yeah, we'll make sure to add that to the show notes. And you mentioned you were, you were at the conference, but you were probably not at the conference at the conference. Well, tell me how the virtual conference went.

Sam Dribin: 20:00

Um, yeah, I have to admit, I was, I usually am on the ground there, um, in the booth and, and talking and, and demoing. And I think this year we've had an excellent business development and sales staff and medical group that, ran as well as marketing that ran primary. And this year I leveraged them to come get me if needed aside from the talks that I wanted to watch. Um, but it sounded like it was okay. Um, you know, I think we're all making the best of it. Um, but certainly there's, there's active online traffic for that. Huge is usually a huge conference.

Robert Fenton: 20:32

Sounds like you think that conferences in person are going to come back.

Sam Dribin: 20:35

Hard to predict the future on that front right now. Right.

Robert Fenton: 20:39

Uh, I'm looking forward to going to a conference again. Um, I really appreciate you sharing your, your similar story here. I'm really excited to keep it close and, uh, I am. Excited as a company, as Qualio helps supports you in the work that you're doing. So thank you for, for trusting us. And, uh, I'm really excited to see where you know, where the future will take you and to help to follow you along that path.

Sam Dribin: 21:01

Appreciate it. yeah. And it's been, Qualio has been just a great help so far.