'AI-vengers Assemble': Extending Lives, Embracing Equity, and Eliminating Burnout With Intelligent Data

IO Learning spoke with Jeff Elton, President and CEO of ConcertAI, and Dan McSweeney, President of TeraRecon, the healthcare business unit of ConcertAI, to discuss the next generation of RWE and AI platforms in oncology, trial efficacy, the unbiased and streamlined nature of ConcertAI, and how it can help both IO professionals and their patients now and in the future. 

Can you tell our audience of IO professionals a bit about ConcertAI and how it works?

Jeff Elton: ConcertAI is an entirely oncology and hematology-focused company. We started off really focusing on a domain that's kind of termed real world data, real world evidence, which really has to do with generating evidence about currently available medicines, how to use them better, how to improve outcomes, begin to get that into peer review journals that starts to inform treatment guidelines and other things. We've been a long-term working partner of the American Society of Clinical Oncology and even during the pandemic we've kind of created kind of COVID-19 registries with them to work with them about health equity and other considerations, making sure that everybody has appropriate access to oncological care. So that was our beginning. I mean, it was all around getting those insights that can guide treatment to improve those outcomes. As we continue to evolve, and as you see with Dan's introduction from TeraRecon, we're starting to develop relationships with people like Caris Life Sciences, bringing TeraRecon into ConcertAI, and starting to align a lot more to the diagnostic side of activity.

And as you know in oncology, it's super complicated, right? Cancer is hundreds of diseases, it presents differently for individual patients depending upon where they are in their journey, depending upon the genomic status of their particular cancer, if it happens to be a genetically targeted or mutated cancer, et cetera. So there we've tried to really start bringing in deep genomic data and also taking a look at radiological data and digital pathology information. And now we really are taking the view that multiple modes of data, meaning I can have a full exome, I can have a full transcriptome, I can actually take a look at the digital pathology, I can take a look at all the radiological, but when I pull those together, it gets me a different insight about who that individual is and the characteristics of their specific cancer.

So now, if I wanted to actually design a clinical trial, I can look to make sure that the new therapeutic entity that I want to go take into the clinic for those clinical trials, is really targeting that patient with that characteristic and not this one over here that likely won't respond to that. And that type of information becomes valuable both as you're going through clinical development, but also becomes part of the information that's available around that patient actually as they're now in standard of care treatment and able to use now approved kind of medicines. So we think of ourselves on  two dimensions; generate the evidence that can start to inform standards of practice and then develop the tools, particularly diagnostic tools, that help guide the therapeutic approach once these medicines are now available commercially.

Your bio online states, "Our mission is to accelerate insights and improve outcomes for patients through leading real world data." What role would you say that big data plays in AI for interventional oncology and, furthermore, preventative care?

Jeff Elton: So for one, you actually can't do AI without data. I mean, the whole idea of AI is that AI has to be trained. If you don't have a lot of data and if you don't have a lot of data where you actually know it's truly representative of the people who are ultimately going to be treated, what you're training on may have its own bias. So even to kind of say that you're going to develop... And the same thing's true even if it's not AI, even if you're doing a research piece and you're doing classical statistical analysis and comparing Kaplan Meyer outcome curves of two alternative approaches, if there was a bias in the data, then it's not clear how meaningful the insight is or you need to tell people what the bias is, in this set of circumstances, this was what was there.

So for us right now, the reason we started as a data company is that you want to get the insight, you want to get that insight at scale and you actually want to have confidence in it, you want to be transparent about it. I mean, we now have 225 publications that underpin and actually support our data. We do work identifying patients for clinical trial eligibility that uses AI, machine learning, natural language processing, et cetera. We presented posters at ASCO on how we did that and what the transparent... We want people to be confident in what we do and how we do it so that in fact they have the confidence to integrate it into how they do their clinical work for kind of doing that. So first and foremost, it is about the data and the data informs almost everything. It informs trial design, it informs who is benefiting actually when a study's over. I don't know if you're aware, we have a five-year collaboration with the FDA and we're probably about a year and a half...well, maybe a year and two thirds into that. Two of the things we're doing are taking a look at populations of patients that were excluded from clinical trials and we're trying to make sure are they beneficiaries in the medicine just like the trial population, because in certain circumstances these patients may be 10%, 15% of the actual population getting the medicine, but we have no data whether or not they actually receive the same benefit, therapeutic or safety, and things of that nature.

So we can use our data at scale. We now have the largest collection of clinical data of anybody in the United States and likely anybody in the world. And actually as part of that, we now represent all 50 states. We have high depth in probably 39 of those 50 states. The fact that we can actually track data across the entire patient journey, all of that's not there just to be big, to be big, but to be big, to take on really rare cancers and really tough questions and get to answers that people can have confidence in.

You say that ConcertAI designs trials more effectively and optimizes trial site footprint. I know you've already elaborated on this, but how does the system work to streamline this process and benefit the industry as a result?

Jeff Elton: Thanks for actually reading the things we put on our website. You never know if anybody does. [laughing] So, if you think about clinical trials and probably given looking at your website too, as I'm kind of talking a little bit, if you think about clinical development is very complex and involves lots of inclusion exclusion criteria, establishing whether or not a patient's eligible. If a patient's at an early stage of their cancer or if they're in a mid-stage of the cancer as opposed to very, very late stage, you probably have standard of care treatments that may work, sometimes they may not work. So when you're making a decision about placing a patient on a clinical trial, it's a pretty deliberate decision to kind of understand that treatment versus the current available standard of care.

Also, and you may even be covering this at IOL, but right now, staffing in clinics has been challenging ever since the pandemic. Retention of key people has been challenging because of the pandemic. Graduate rates of new medical oncologists, even radiologists doing work in oncology, actually are going down. So you're talking about a situation where you want to afford more availability to more clinical trials because it is part of how we give cancer care because there's such a rate of new therapeutic entities in cancer. And you may be aware of all clinical development in the entire world across all disease areas, more than 50% of it is cancer related, of everything that anybody does any R&D on, it's cancer related. So in cancer clinical trials are a viable alternative to that standard of care. So the first thing we try to do is find the patients and actually present them to let people know what trials are they actually eligible for and meet the criteria for.

And we use AI machine learning and we create AI models that read the record like a human does. So we actually studied how a human and a clinical research coordinator actually takes a look, establish that, go into the EMR, goes into the notes, goes into the penant documents, but now we can actually have machine support that human. It's not replacing, it's support that human, but then one individual now can actually review 10, 20, 30 times more patients for study eligibility than before. So even if I have staffing, and budget issues, and all the conundrum that community and other oncology centers tend to have, this really helps support what they're doing. So on the one hand, so we think we do it better than anybody else, and that's why we presented it and put it at ASCO. And we didn't always do it better than everybody else, but over time by improving the models, and the approach, and the value of the data, we think we do it better than anybody else at this point and more accurately.

There's something called a false negative, meaning if a patient's eligible for a study, you find them. And there's something called a false positive, which is if you tell them all these people are eligible, but it turns out that almost none of them are eligible, you just created work. So you're working that corridor between you don't want to miss somebody because somebody's life hangs in the decision, but you don't want to create more work. You're trying to actually do it. On the other side, if you think about how clinical trials work, clinical trials work that every single trial sponsor has their own technologies. And so each study may bring their own set of ECOA, or kind of clinical, like their metadata, their Veeva, their Clario, or whatever technologies they're using up to that research site. So that research site may have five, 10 different web portals with different technologies.

And what they're doing is they're going over to the EMR, they're looking at the EMR and they're typing data in. And so we watch that process and you're like, why is somebody taking data that's in a machine, looking at a screen, and typing it into another screen? And so rather than having the paradigm that every sponsor brings their own technology, we actually said, let's put the technology into the workflow, the research site into that provider setting. Let's read the files that already exist on the patient and let's autopopulate those case report forms so that the people at the research site don't have to do data entry. And sort of the principle here is data that's digital should actually stay digital.

And so what we're trying to do is lower the burden, lower the burden on the site, allow them to have more studies accessible to them without actually having to spend more money and have more work, actually have very high standards of practice and lower the burden on the clinical personnel, so they feel like they get more time with the patient, less time typing in things on the portal and kind of doing it. Half the time these people do it after they close off their visit day, they actually sit down and start typing these in for two to three hours at the end of the day. So when you think about it, and these aren't sustainable, it's not good for accessibility, so we reframed the paradigm. We've now put this into place. We've made some joint public announcements with companies like Bristol-Myers Squibb that are actually using this solution kind of into the workflow to start running some of their clinical trials in the United States.

And some of the results that we have of this are fantastic. And so we continue to invest in this very large scale. We feel a big commitment to making sure the trials are accessible to all patients in all areas regardless of setting, and that we can contribute to the diversity of that patient population and to make sure that the trial population looks like the ultimate population that will get those drugs, which is actually what the FDA wants as well.

You mentioned the fact that there are struggles with retention and growth in this industry as of late. Do you think that this technology will help with some of that burnout that we've seen creating these various issues?

Jeff Elton: It's so funny, I'm actually glad you used that term burnout. In fact, when we were at ASCO, we had a couple working sessions on that. That was the first word that one of the physicians used: burnout. And they said that everything—every incremental thing that you put onto them—takes a group of people, they have a little fragility, and that contributes to that potential burnout. And their actual response—and the whole reason we were having the conversation—was they said, if you can continue to do this and keep improving its performance, you are making a major contribution to the avoidance of that burnout. And you're actually allowing us to get around all the conundrums and challenges we have of recruiting staff, retaining staff. And the problem is, with higher staff turnover, you're just constantly training.

So, you never actually get to the point where you're getting it and also letting them have the time with the patient, not with all the administrative and electronic infrastructure that demands their attention. So that's definitely part of the goal. I appreciate you framing it the way you did.

Now, I have a question for Dan. On LinkedIn, you discussed a podcast you hosted with Jeff regarding ConcertAI. You stated that "addressing burnout, improving the productivity of care providers, reducing unwanted variability, and helping to deliver better patient outcomes" all were highly important to you both. How can this AI system improve upon these areas of the interventional oncology and radiology world?

Dan McSweeney: Again, thanks for taking the time to read through some of the important things we feel we're putting out. So, Jeff and you just talked about burnout. TeraRecon in the advanced visualization world has spent the last 15 to 20 years, and one of its big charters is around radiology and cardiology burnout, which has been a thing for a long time. And it's no surprise that as you get more data, more pull, more cooks in the kitchen, that we're seeing that same type of burnout in the clinical oncology, and in the oncology, and in the clinical trial space. And so we feel, we've spent a lot of time around this, it's really part of our culture is how we design products, how we implement them, how we support them. A lot of it is designed around how to reduce burnout from the actual clinical users as well as improve the workflow.

So what we've learned and that we continue to—Jeff hit on it—that we continue to employ here is really everything that comes in, many things are going to have to be in the workflow. So if it's a swivel chair or if it's, to Jeff's point, you're copying this from this, or if it's just additional work with no additional benefit, it's just going to continue to increase the load. So when we think about how we on the TeraRecon advanced visualization imaging side are able to partner with and complement what Concert has been doing for a long time, when you think about clinical trials, there's really four areas where we feel we help. One is around the image and data management. Those are big challenges with these trials. And another is with the project workflow and management. And another is with the algorithm development and management.

And just even on the first couple image management, data management, and the project management, as Jeff said, these used to all be on pieces of paper and now they're... The industry digitized poor processes and so nothing is connected. And not only is there a human element to that, but there's also a workflow element to that. And then candidly, the nirvana, the goal that we're all looking for is not just to have the same task take less, but be able to have really the right tasks in the right buckets. So eliminate the non-value added work and let's have the people that are trained on clinical outcomes, or on new therapies, or on protocol management be able to focus on those. Because when that happens, we're able to see really exponential gain in the results as opposed to filling out forms and QCing translations between this system and this system.

So we spend a lot of time again, partnering with the very robust offering that the ConcertAI teams already had around image data management, image management, data management, the project workflow, and then maybe we'll get to it today or not, but also around the algorithm development and management. Because each of these life science companies, as they're facilitating these clinical trials, are using big data and AI and they also need a common platform to be able to develop their own AI as well as import and manage existing third party AI that can be beneficial that's already out there in the marketplace. That's how we think about it.

Jeff Elton: I want to add one thing to Dan's statement because he's saying something important and you probably know this just from your work in this area at IOL. Healthcare providers are all organized very differently. Right? So they have different EMRs, different infrastructure, et cetera. So when you think about doing something like what Dan does in TeraRecon, they're trying to be a layer that says, okay, you may have Phillips imaging, and GE imaging, and your Siemens imaging, but if you work within that TeraRecon layer, you're independent of that one manufacturer's imaging instrument and you don't necessarily have to go up their software stack. I can train and kind of deploy, so all of a sudden now I have a consistent diagnostic and other experience. So that becomes super important. It's one of the reasons why, and in fact, everything we do has a little bit of independence from it and we try to make sure that there's an openness.

Also, the AI architecture, it's a clinical AI. So these are softwares, medical device, 510(k), so these are actual medical device class AI models that are being deployed, but it's an open architecture. So the individual healthcare provider makes a decision of which of these modules do we want to deploy as an augmentation of our own decision and we let any third party components come in, we actually open it up. And if they develop their own, which there are, and I've noticed like your editor I think is kind of from MD Anderson, et cetera, well, the MD Andersons and the Memorial Sloan Ketterings actually do a lot of their own internal algorithm development, which don't have to be FDA approved. If they're deploying it on their own patient population, they can deploy those too. So this idea that you can streamline, simplify, unify this, but also give them control to practice the model they want to practice actually becomes a super important kind of part of this. And there aren't very many people that make things more uniform and simple, but also give you more control at the same time.

That'll make it so much easier for a lot of these oncologists to be able to have a system that they can manage on their own and make their own.
 

Jeff Elton: And make their own, that's actually, that's a great word. We want them to clinically make it their own.

Another aspect of ConcertAI you shared with us at IOL is that the system helps advance equity and diversity in clinical research. How exactly can the integration of AI help in this regard?

Jeff Elton: So several different ways. We have a little scenario that we'll sometimes simulate for people to give them an illustration. Multiple myeloma, for instance, does disproportionately affect Black Americans and they have a higher rate in the disease state. If I deploy say a IE criteria, inclusion exclusion criteria, that says hypertension above a certain level is excluded or a hypertension here, that population also has a much higher rate of hypertension. And so in a way, depending on how I design my trial, hypertension's not really one of the most critical elements for a multiple myeloma study where I'm actually kind of looking at configurations of cell and it's kind of blood based. I'm actually looking for kind of response, minimum residual disease, and a bunch of other characteristics so I can unwittingly exclude a population that is disproportionately negatively affected by picking one variable.

There may have been a rationale for it, but it's a modest and kind of subordinate and less important rationale. So how do I design things from the very beginning to be inclusive of the people and the patients, of those peoples that really are kind of suffering that disease state. So it starts at the beginning, starts at the design kind of itself, and we have tools that show that, reflect that in terms of what it looks like. Second thing is most clinical trials and studies had historically, and this is not to say this isn't important, it is important, but they're historically done at a very narrow number of very large academic cancer centers. So if you're at a Memorial or MD Anderson, I'm here in Boston where I have the Dana Farber and the Mass General, half of their staff spend... And if you look at how they get paid, they're paid from multiple different kind of pockets of money.

And part of that is research-funded and part of that's care enterprise-funded. So they're paid to do research, but the patients they see tend to come from wealthier, better off, more economically robust areas, and they don't look like the 80-90% of the population in the United States. So, we need to make sure that the studies are accessible in the areas where that 80-90% do live because those are the people who are going to be treated. And again, I'm not saying it's not, in fact, it's extremely important that we have the academic centers because they do a lot of work on phase one, and the initial proof of concept, and validation of the biology of some of these new medicines, but as you go to phase two and phase three, you really do need to begin to pull that up.

Second thing is we do a lot of work and we have medical in addition to clinical data, we also have social determinants of health data that we also syndicate. So we can take a look even within an individual geography, and you may even see this even with some of your own readership here, sometimes they don't have great information about the status of their own populations that are in their kind of primary service regions. So we can bring our data and enrich and provide insights about their own populations that actually help do that. We actually actively do that, and we do that as kind of part of our own responsibility towards them to help them achieve the goals we want. In fact, even at ASCO, I didn't talk to a single healthcare provider institution that didn't bring up the world equity to me.
100% of them did, 100% say we're incorporating that into our practice model, into what we're doing. And I'm like, I mean, one that's phenomenal. That was not the case if we went back five years. It would not have come as one of the topics of conversation that they'd be sitting in the hallways at ASCO going into. So I think part of this becomes on how you work, how you enrich, where you do the work, and et cetera for doing that. So every data set we build, we actually have the ability to bring social determinants of health to it. Every software solution we do can actually look for subpopulations that are adversely affected on this and we can do it on the basis of economics, racial, ethnic considerations, kind of everything else like that. So what you end up doing is you build it into how you work and you just make it part of the model as opposed to at the end of the day, but it has to be in lots of different parts about how you work and what you do.
 

I love the fact that it's not just about equality and inclusivity, but also the concept of equity.You're not making everyone the same. You're actually recognizing the fact that certain cultures respond differently to certain cancers and cancer treatments.

Jeff Elton: Yeah, you're not making them the same. You're actually recognizing them for who they are, and you're letting them be who they are, and you're making how you can interact with them let them be who they are. And you could take that definition the same way into lots of little category headers that you might want to kind of put on that. That's super. Yeah, that's a very important way of reframing that.

Currently, you have collaborations with Caris and PathAI as well as engagement with Medexprim to advance transatlantic real world evidence offerings. How do these partnerships and alignments help push the company’s goals forward? 

Jeff Elton: Healthcare is super complex. Data sits in lots of little pockets, even if you're in a healthcare provider enterprise, some of it sits in an EMR electronic medical record, some sits in a laboratory information system, some sits in Dan's world into a PACS archival system for images. So you're always getting around this problem of stove-pipe data. So all these little verticals, but patients work through that system on a horizontal. So when you know want to solve a problem and when you know want to get underneath that problem, part of our job is to sew together and knit together kind of these verticals so that in fact you get a view of the patient journey on the horizontal. And that's a big part of who we kind of work with and how we try to bring it together so that we're really completing the picture.

If you were, and I don't know your background, but if you were doing a lot of work in say, advanced statistical analysis or something, there's a term called confounders. And if I only have a limited number of variables, you may say it's an outcome that seems to indicate the trend was this way, but you have a lot of things you can't explain. So what we try to do is understand what are those things that can eliminate the confounders so that you can actually have more confidence. So as an example, if I am able to link together the genomic data and the transcriptomic data, and so just an example, Caris' diagnostic test for solid tumor work is 22,000 genes. Most people that do this are 498 genes. You're talking about the difference between 498 and 22,000. The reason why the 22,000 matter is that a lot of cancers have a single gateway mutation that's responsible for the majority of it, but then they have secondary, tertiary, and other mutations that may be part of the resistance they may have to certain drugs.

And if you don't have that broadest view, you can't really make the determination of what treatment approach has the highest likelihood that the patient's going to respond to. And sometimes there's an example that we use of connecting in the transcript the RNA and the DNA data that can show the difference that certain patients may respond in colorectal cancer to FOLFOX and others may respond best to FOLFIRI and it matters what order and what sequence you put it in, and that's knowable. Yet not everybody does the testing and the diagnostic process to know it's a 30% outcome difference, not like a 5%. Also, 30%, by the way, in outcome terms, that's the equivalent of a newly approved drug and the difference it provides over the former standard of care. So, these aren't small things that you try to bring together.

By working with people like Caris that we kind of come away and say they have greater depth, we have great clinical depth, and by getting together with some of the other partners, that's what we're trying to assemble through our pattern of those particular partnerships. The other thing is, I mean, healthcare always is founded on the principle of competition and not cooperation, but at the end of the day, you kind of use the patient as the centering. And if you keep doing that, then parties do get together and they actually kind of bring together the right assets in the right way and you actually get the right cooperation for the right outcome.

In an ideal world, what do you hope ConcertAI and TeraRecon will be able to achieve within the IO industry in the next 10 to 20 years?

Dan McSweeney: I think there are the goals and objectives that exist out there that are the goals of many, many companies, right? Providers, life science companies, AI companies, tech companies, and a variety of NGOs all are striving for these same goals and objectives.

I think where we feel can, I would say, have a unique version of the difference we can create really is around when you think about, first of all, when you think about where we're focused. We're focused on the post-diagnosis perspective, but spending a lot of time on the pre-diagnosis area and on clinical trials and then even more further upstream into drug development, population health, so that we can try to provide benefit after diagnosis. But the real secret sauce is how do we prevent these things way far upstream.

When you think about the stable of data and resources that we've developed now through Concert, the legacy ConcertAI, life sciences business, the partnerships that they have, and through TeraRecon we think we are really uniquely positioned to, as Jeff talked about a little bit, to look across all of that workflow and across all of these disparate verticals within this very complex GUI of the healthcare clinical trial perspective and drug development perspective and really be able to make meaningful contributions. So when you think about the EMR data, 10 years ago it was we need EMR data to drive better clinical trials. And then now we're talking about genomics, and we're talking about digital pathology, and we're talking about social determinants of health, and we're talking about image, we're talking about images, and the desire and the need for additional inputs into the system to be able to drive better results.

And now I'm not even speaking from a workflow perspective or burnout, just overall better outcomes is complex. And that's where we feel that at the ConcertAI level, that we're pretty fortunate to have this unique proposition of not only just having access to all of this data, but then having the platform expertise, experience, and clinical tools or AI tools to be able to aggregate that data, to organize it, to curate it, to annotate it, segment it, track it, and be able to really have it be focused towards driving better outcomes.

Jeff Elton: Dan, I appreciate your comments on that. I think maybe to build a little bit on it, we're sitting here and all things we want to be able to turn cancer into something that starts moving towards something that becomes less acute and more chronic and then eventually move earlier in the stage of the disease to the point where you can start thinking about cancer prevention. That's what we are all trying to do in terms of where it is. So treatments are becoming more effective. We are seeing life being extended and some of the work that particularly the clinical communities that are in IO and particularly are doing it, which can have a radio or a device based approaches and things of that nature where you're focusing an intervention on a part of it, these are important overlays where you have an insight, you have an imaging insight, you may have a molecular pathology insight around things in tissue.

Partially, what you're always trying to do is constantly narrow down the cells you have impact on, constantly get more specific to that so that you're only getting those cancer cells, you're only getting the ones that are actually responsible for the driving and the characteristics of that particular pattern of cell growth or hyper cell growth that's kind of underlying in it and be super sparing on anything else kind of around it. Because that's actually what starts to deplete the patient. And when you cause harm to healthy tissues, that's also where you're lowering different things. And even people that are a bit more fragile because they may be older in their life, it actually has its own consequences as we're doing this. So everything here is actually around the insight and precision and everything is around moving that insight and precision a bit earlier kind of all the time.

I actually think right now, if I had to say just based on the level of insights, and that's one of the reasons why we bring AI, and data science, and different solution, the speed with which we can get an insight and the precision of what that insight is, but using tools that are available and accessible in standard of care that can talk to the radiotherapy group, that can talk to the other one, and start to integrate these pieces together, that's actually moving at a rate far faster than what I would've actually have seen say if we were 10 years ago or even five years ago. And so that's great because what that means is clock speed of these innovations and the clock speed of the innovations being adopted is starting to move and then the outcome benefits will kind of start to move. But I think what we have to do is we also have to make sure that the regulatory framework, the reimbursement framework, and the other things actually allow that to take place.

Because sometimes we can do it clinically and scientifically and from the device and the others. So there's a context that this all works in too and we have to make sure the context sometimes either enables or stays out of its way so that we can get that done too. So that's one reason what I'm super excited about, and one thing that keeps us doing what we're doing with the enthusiasm we have for doing it, it's one of the reasons why we partner, and it's one reason why I want to talk to you, and kind of everything else.

What do you think the future of IO in general will hold given your expertise and connection to advancements in the field? And you can even elaborate on other technologies such as AR, robotics, etc.

Jeff Elton: Yeah, I was going to actually go a little bit in robotics. So you do have a lot of things where you have image guided surgery patterns, you're going to have assessment of... And in fact, when you think about the level of accuracy and precision, and so again, everything's getting down to narrow numbers, margins are getting tighter. You're getting down to almost thin cellular levels of actually ability to intervene or to guide a beamline or whatever you're doing. These things are kind of going down. So for one, the interplay between devices and technology, or radiotherapy and technology, radiation oncology and technology implanted drug alluding technologies that keep actually the medicine just going into the tissue of interest and away from any of the healthy tissues. So actually I think again, in some of this area, you actually need the diagnostic tech and the imaging approaches combined with the other interventional approaches as a set.

And they're all AI, machine learning, and visualization layers can see more features now. So there's some work we actually presented and had a discussion at ASCO about digital pathology. And when you use AI based interpretation approach on digital H&E materials and slides, you can create a view of 146,000 unique features on different cell types, textures, depths, and layers in that particular tissue. Each of those can then be analyzed and reduced down to guide to a very specific interpretation of that particular patient's cancer, to then guide the intervention on that. That wasn't possible. We didn't even have the processing capacity. We didn't have the GPU capacity to do those things. Graphic processor unit capacity, which a lot of AI doesn't go on CPUs, it actually goes on GPUs, which is why companies like Nvidia, which you think is like a video game, well, they're actually making some of the biggest, fastest AI processing chips in the industry.

So when you go through all that, we're actually at the... The reason why we're excited about it is we're seeing real results, but we're still actually at the early days of those results. So I think that's going to be an extraordinary, exciting area, and it's actually only going to benefit patients. And I'm not worried and threatened about AI is going to put these people out of work. It's not going to put anybody out of work. It's actually going to augment what they do so that they feel better about their work and the outcomes that they're able to achieve.

We talked a little bit about genomic data. Do you feel that down the road we'll get to a point where we will be able to definitively predict and prevent cancer based on someone's genetics?

Jeff Elton: Between the area of liquid tumor biopsies combined with solid tumor next generation sequencing, you'll know something about the individual as an individual and you'll know something about their specific cancer. And you may or may not, but every time you give somebody a therapy, you actually start... the body and the cancer will become resistant because the same thing that keep us alive, that's allowed us to survive under really adverse conditions, we have built-in redundancies. Cancer cells do too. So they have the same capability as humans. They will do everything possible to survive and they'll figure out how to become resistant to that medicine or that approach and kind of what's there. So when you read the DNA, you actually can read the full story of everything that that person's been treated with and doing because the DNA retains, it doesn't forget the story. And so you can actually start to understand 'how did they get to where they are, what's going to work now?'

And what we're going to start doing is rather than saying, we're going resistant, we're going to know that when we do this, they're going to become resistant. This is what's going to happen, but we're going to know what we're going to do. So instead of resistance being saying the medicine doesn't work anymore, we're going to say, we know they're going to be resistant. We'll monitor that and we now have liquid biopsies to monitor that. So it's just a blood draw as opposed to a tissue, or a needle biopsy, or something else. And then we'll know what to bring in next because we'll have a predictive ability to know what actually we're going to be dealing with at the back end of that. That paradigm is very different and it's not like I have to be fearful of it. I just know it's a journey that I'm going to go through. And I mean that's going to make people feel a lot... It's going to get outcomes of interest, but also I think everyone can kind of lean into it with the right way, the care team and the patient.

What's next for ConcertAI?

Jeff Elton: Definitely scaling up and making sure that digital, heavily accessible, deployed clinical trial, clinical trial infrastructure is available for as many patients as possible. And we'll be partnering with people all over the community, we'll be moving more of that outside of the United States. So not just in the United States, but I think our plans will be to begin moving that out. I think you're going to find us working a lot more on tying together multiple diagnostic [inaudible 00:45:22]. So people talk about tumor boards, well, if you imagine taking imaging data, having a collective interpretation of that particular patient's cancer, and then likely interventions to be making, now let me add the molecular data. What if I took a molecular tumor board, a radiological tumor board, a digital pathological tumor board, and I overlaid and integrated that together? Now my decision layer that can guide the interventions just becomes a lot more sophisticated.

That's definitely the direction we're going into. So that level of sophistication is way beyond normal evidence-based approaches and guidelines. It builds on that, but it also contributes to, it makes it very specific to that patient and that kind of stuff, that and accelerating new medical innovations, that's what motivates us.

For more information regarding ConcertAI, visit concertai.com

To learn more about TeraRecon's exciting work, visit terarecon.com