This is our interview with Georgi Kadrev which was recorded for the Solutions OP Clinical Trials podcast. You can listen to the original recording here. Please note that the interview below is adapted and not a transcript. This was done to improve readability.
Interviewer: Welcome to the latest episode of the podcast. As you know, I’m trying to invite people who work in various startups in different fields and disease indications from various countries and places throughout the world.
We are trying to understand more about what people are working on at the moment, what kind of challenges they’re having with research and data generation for clinical research, and how they’re getting their products to the market.
My guest today is Georgi Kadrev, a Bulgarian entrepreneur, who has been involved in numerous businesses over the years. He is an excellent mathematician with a background in computer science. He is also the co-founder and CEO of Kelvin Health, a Bulgarian business focused on creating thermography AI technology for vascular diagnostics. Georgi is also a board member of Bulgaria’s digital health and innovation cluster, and he still finds time to be a lecturer at the Faculty of Mathematics.
I’d like to thank him specifically today because he’s taking time out of his annual leave to chat with me. Georgi, thank you for taking the time to join me today.
I’d like to ask you the first question to learn more about your company and what you’re doing right now, as well as any general overviews or information about your job that you’d like to share.
Georgi: Thank you for having me, Olga. It’s my pleasure. I always love to share about the mission of Kelvin Health, and this is a venture, actually a spinoff venture from another company that we started more than 13 years ago, because our background is in visual image recognition, before it was a modern thing as it is today, and the opportunity occurred to help in diagnosing different kinds of infectious diseases.
The actual occasion was the start of the COVID-19 pandemic, an idea by a team of medical scientists to apply thermography. Artificial intelligence analysed the thermogram images to detect COVID, and this was a little extravagant but showed some potential.
However, in the year following the onset of the pandemic, things changed several times. We even experimented with applying tomography to breast cancer screening, which is one of the most contentious applications. We ended up developing a solution to empower vascular specialists, and even non-specialists, to accelerate the diagnosis of arterial conditions. In particular, we are focused on peripheral artery disease and its progression to critical limb ischemia and carotid hypoperfusion.
These two vascular territories are areas where we want to make the detection of problems such as clots or stenosis much easier. The main premise is that these conditions are currently very difficult to diagnose in a wider setting.
At the invasive end of the spectrum, we have X-ray angiography and other forms of angiography, which typically require hospitalisation and carry certain risks for patients, such as kidney complications from contrast agents.
At the non-invasive end of the spectrum, there are methods like ultrasound echo or the ankle–brachial index, but these are not always reliable and still require specialised training for the person performing the test.
So, this came very promising to apply our recognition technology and provide a solution that can empower diagnostics.
Interviewer: That sounds very interesting. It’s a very interesting field that you’ve picked, and I think there is a strong demand, especially for imaging diagnostics. I see quite a lot of startups working in radiology, for example, detecting tumours and similar conditions.
So I think your niche is quite unique. I imagine your idea is for people to use this tool for some kind of pre-screening, and if they discover something, it could then trigger further assessment. Is that how you envision your product?
Georgi: Yeah, to a large extent, what we want to enable is the ultra-early detection of signs of vascular disease. It’s not necessarily “ultra-early” in the general sense, but rather in the context of how the healthcare system currently engages with patients with peripheral artery disease.
The main issue is that when patients report symptoms, such as leg pain or claudication, the disease is often already at an advanced stage. One of the major comorbidities of peripheral artery disease is diabetes, which further complicates matters by masking symptoms due to reduced sensitivity to pain.
So, if we can create something that can be used even at a family doctor’s office or by a podiatrist, something that reliably detects significant changes in blood supply, that could empower the diagnostic process.
This can essentially be a goal to install as part of the healthcare system to detect the patient much earlier on their patient journey, so they can have something much deeper and way better protecting their health and increasing the chances of saving their legs, because unfortunately, late diagnosis peripheral heart disease escalates to ischaemia, and it has a very high percentage of first taking the limb and dying in less than 5 years.
Just to add to your earlier remark, imaging has indeed been a very hot field over the past decade, especially with the rise of artificial intelligence applied to traditional medical imaging modalities like X-ray, MRI, and CT scans. But something particularly interesting in telehealth is the innovative use of less common modalities, such as thermal imaging.
The advantage of thermal imaging is that it can be highly portable, mobile, and cost-effective, making it far more accessible in terms of both price and form factor. That’s why we became so excited about applying our background in imaging, specifically non-medical imaging toward medical applications, particularly the analysis of thermal images.
Interviewer: That sounds really interesting. I think you’ve chosen a great field, very niche and very exciting. I wish you the best of luck with your work.
I know you’re very passionate about artificial intelligence, you follow the field closely, and you’ve been involved in many accelerators. You’re also working as a mentor. Honestly, I don’t know how you manage to do all of that, but you do it.
So, tell me, how do you feel about AI in medical diagnostics? Do you think the field is overheated, or do you believe AI has significant potential in medical diagnostics? I’d love to hear your thoughts.
Georgi: Yeah, I think that with the adoption of every new technology trend, there’s usually an initial surge of very high expectations, followed by some disappointment, and then things eventually settle into something more realistic and sustainable. That said, the trend toward AI-based analysis exists for a reason. With the availability of data, machine learning algorithms can now be reliably trained to detect patterns in images, speaking broadly.
It could be an anomaly, a malformation, a particular spot, or something else, it depends on the therapeutic area. But AI truly holds the promise of automating a lot of the detection work, at least in the initial stages, whether for triage, assistance, or helping to resolve complex diagnostic situations.
An important part of this equation is data availability. For established imaging modalities like X-rays, MRIs, and CT scans, we already have vast amounts of information available in practically any hospital or healthcare system. For thermography, however, it’s both an interesting challenge and, from a business perspective, a competitive advantage. The reason is that there are essentially no large, high-quality thermal imaging datasets with enough patients and scientifically validated results.
We scientifically agree on the results, and we basically just bite the bullet and say, “We’re going to do that.” We’re going to take thermal images of as many patients as possible who are also examined using traditional methods. For example, in our case, the gold standard affects your angiography if they’re going to go there and have the angiography done. Then we pack ourselves with this process and take thermal images of the patient before and after angiography.
To wrap it up, I think AI, especially in imaging has evolved to a point where both the technology and, to some extent, the datasets are mature enough to deliver real value. It can save a lot of time and help reduce the inaccuracies inherent in any human-driven process. Of course, machines also make mistakes, which is why it’s crucial to define exactly where in the diagnostic workflow such a solution is applied, and how much trust is placed in it. But as an assistant, AI can be an extremely valuable tool.
Interviewer: Yes, that’s true! I agree with you. I think imaging, particularly in diagnostics, holds huge potential for AI.
I’m less certain about areas where you have to rely on source data from hospital medical records and use that to train models. I’m not sure how accurate that would be, because data generated by humans can vary a lot in quality, and there are many pitfalls in working with it.
But when it comes to imaging, I believe it can be really powerful. On that point, I completely agree with you.
Georgi: Yeah, if I may build on that, I’d say for imaging it’s a no-brainer. For the analysis of structured data, it’s more questionable, but to a certain extent, I’m still optimistic, mainly because of the rise of recent trends. Well, they’re not that recent anymore; it’s been almost two years since the explosion of large language models, which have shown enough capacity to extract structured data from unstructured sources.
To basically extract structured data from unstructured data, and as I mentioned, we’ve seen this with our prior company for more than 30 years, regardless of whether it was in the medical or any other field. Any business problem requiring artificial intelligence always comes down to defining the problem in the right way.
What is the data format you truly desire, and can you confirm that whatever the output of the artificial intelligence process is accurate? If you can, then you start to build confidence and trust. Of course, context matters, and how you use this knowledge and insight.
We should think of AI as providing additional data points or guidance that can influence decisions but not yet make them independently. That’s why I’m still sceptical about AI making a final diagnosis within the next couple of years, maybe even five years. It could happen, but either way, any insight we can extract from data that already exists is highly valuable.
If we take it with a grain of salt, it’s always something that might give us a hint, so it’s a great value add because it makes the process more informed. We’ll either be a little more proactive or receive a better diagnosis.
Interviewer: Yeah, or at least point us toward the problem. That’s a great idea, and I think it will have an impact. We’ll see how the field develops. But for the field to develop, people need funding. Since you’re operating mostly in the European Union, what’s the funding situation for biotech startups there? What’s your experience? Is it easy for people to raise funds, especially for medical diagnostics, or is it not very popular right now?
Georgi: I think it depends on current trends and the priorities of both the venture capital community and the European Commission. At the moment, I wouldn’t say there are still opportunities, especially at the smaller end for pilot projects or accelerator programs. Some even provide grants or awards to build prototypes or do early development work.
But when it comes to larger funding needs, like for clinical trials or the heavy administrative work required to secure reimbursement, which is the cornerstone of any diagnostic. It requires significant resources, and the venture capital community is still not very supportive of businesses with long time-to-market R&D risks.
The interesting thing about healthcare is that from the outside, it’s often perceived as very high-risk from scientific and research perspective as well as a regulatory one. Few people are aware that, in the long run, they are relatively low risk in terms of business and commercialisation.
So, you’re doing a lot of heavy lifting. To the point, but once you prove it, because the process of proving is so difficult, by definition, you’ve bullet-proofed yourself, and as part of the regulatory approvals process, you’ve had interactions with leaders and actual potential customers.
With your product, clinicians, healthcare workers, and healthcare systems themselves, you’ve essentially been able to build a solution that is quite personalised and works pretty well with its dinos. So it’s more of a let’s say – I’m making that way, I believe there’s a huge potential for MedTech and HealthTech startups, but the funding opportunities from the VC in the European Union are still limited. On the grant side, there are opportunities, usually ranging from a few tens of thousands to a few hundred thousand euros, through project grants, accelerator programs, and so on. For larger amounts, the challenge is that there are not many healthcare-specific funds. Grant programs like the EIC, the European Innovation program, are extremely competitive since they compare robotics to healthcare, space technology and any other fields. Healthcare projects, despite their strong social impact, are often at a disadvantage because of their long time-to-market.
So yes, a lot could be done to improve the growth of healthcare innovation in Europe. These things take time, sometimes a decade or even a generation, but I do believe there is space and opportunity for healthcare innovation to flourish here.
Interviewer: That’s very interesting, actually. I find it strange that they don’t have dedicated funds for healthcare. Everyone in research is talking about improving access to drugs and medical devices in Europe, and yet there aren’t specific funds to support startups working in that very field.
I find that very strange, but it’s not unusual. I recently spoke with another entrepreneur in the U.S., and we discussed drug development. They’re seeing something similar, early-stage startups struggle to get funding.
Most of the money tends to go to later-stage companies, where investors feel more confident about guaranteed returns. But the problem is, if you don’t support early-stage companies, how on earth are you going to have late-stage companies later? At some point, the market will be saturated, and more investment will have to flow into the early stages. I hope this happens. I don’t know if you feel the same way.
Georgi: Yeah, well, it’s an interesting dynamic. It depends on whose perspective you take.
I believe we absolutely need competition and small independent vendors that eventually grow into strong players, even strategic ones. Right now, much of the innovation in healthcare and related fields happens through acquisitions. From my perspective, that’s not the ideal approach, and there are many reasons for that. But let me explain how I see the process.
Research often starts in universities or university spin-offs. In some cases, like ours, it can also come from a company spin-off, which is less typical, but it’s given us some leverage.
Now, some serious research is typically done through universities, university spinoffs, or, in our case, a company spinoff, which is more of an exception for a type of typical situation, but that is what our situation is and continues to be, giving us some leverage to not be directly dependent on venture capital while still being able to do some long-term research with a long-term market.
But let’s assume we’re the small guys. We push forward toward regulatory approvals or reimbursement negotiations, scraping our way through. Then, a big player who perhaps didn’t support us earlier to accelerate their problems, so they’re keeping an eye on us, and they are acquiring us, and still manages to commercialise the solution.
There are pros and cons to this approach. There’s a lot of perceived risk from the outside if this is going to happen. So I think this is one of the reasons why investors are stricter or less risk-oriented.
As for the lack of healthcare-specific funds, there are initiatives like EIT Health, which do provide support. But again, that’s mostly for very early-stage work. What’s missing is funding for the next jump. You’ve got a prototype, maybe proof of concept, maybe even a pilot with a hospital, but now you need resources for expensive clinical trials and reimbursement negotiations. And often the response is, “That’s great, get the clearance first.” But that clearance is exactly the value gap startups struggle to cross. So the support isn’t always that helpful in practice.
Maybe, over time, successful entrepreneurs will change this by setting up funds themselves. That’s something on our own long-term agenda as well.
Interviewer: That’s very interesting. It often feels like everyone wants you to have already done all the work, and only then they’re willing to give you the money. That seems to be the general rule.
Speaking of the work, this brings me to my last question. In your view, what are the biggest challenges for biotech startups in generating clinical research data?
In regulatory forums, we often hear many questions from companies working with AI, diagnostics, software, and medical devices. People ask how to design their studies, how many patients they need, and so on. Based on your experience, what do you see as the biggest challenge right now in setting up clinical trials and generating reliable data?
Georgi: Well, there are many challenges. That’s what makes this field not only difficult but also rewarding, interesting, and timely.
So starting on the more general side, one of the biggest challenges is the lack of experience among entrepreneurs. And this is nature, the people coming forward with innovative ideas are those not burdened by experience and past failures in a particular area.
As innovators, we start out a bit idealistic, then become more realistic over time, but we still carry the vision to bring something new, unique, and unorthodox to practice. By definition, though, we lack direct experience at the beginning.
As a side note, we’re currently part of a U.S. market-access program, one of the most successful medical device accelerators called MedTech Innovator. We just enrolled at the end of June, and this will run through November.
So I’m saying that because a large part of their mission is to actually help entrepreneurs bridge that gap of being experienced with all the many moving parts and building a healthtech solution. That lack of experience, I think, is one of the biggest burdens for entrepreneurs.
The second challenge, especially in AI, is the lack of standardisation. This is still a new field, and regulators are struggling to catch up. What’s acceptable? What’s not? Do you need to prove your data resources? Does it only require testing as a black box or sandbox and measuring the outcomes? Can you actually change the system because we know about the design increase and background as well. But what happens if you want to improve your algorithms with all the data you’ve collected? How can you speed up this process?
These dynamics create innovative solutions that are frequently unacceptable under the current definition of the regulatory processor. Of course, regulators are sensing and feeling that they don’t want to block science and technology. So it takes time again, as with any regulatory process, especially in healthcare, where the stakes are so high because we’re dealing with human lives and health.
The lack of funding is also a challenge that we just referred to. Startups don’t get many shots at running trials; you cannot repeat the trial 5 times. You almost need to get it right on the first try, and this creates a bit of a holdback effect where you want to set everything right from the very start of the process. This is not a bad thing, but this means the process takes a lot of time.
Especially if you’re a first-time entrepreneur or a first-time health tech professional, you may not be able to afford expensive experts to teach you everything, lead all the conversations, and have people take you seriously enough to support you, engage with you, and interact with different regulators when possible.
That’s one reason we’re focusing heavily on the U.S. market. The FDA and U.S. reimbursement system offer more standardised procedures. You can, for example, request a pre-submission meeting and get many of your questions answered early in the process. That’s one of our very next steps.
Interviewer: That’s very good. I think the takeaway from what you’re saying is that it would be great if we had specialised support from regulators for startups. The conversations are very different when you’re dealing with an experienced corporation versus a startup that’s just entering the field.
But if you’re talking about a startup that needs initial advice, I think we need a different type of advice, which is probably more tailored for startups. This is something that I think could be a potential solution to help biotech startups design their clinical trials and find out what kind of data they need.
Having scientific advice available from regulators could be a very practical solution, and something we could recommend as a way to better support biotech startups.
Georgi: Yeah, it’s also a little too good to be true, but that doesn’t mean it’s impossible. I think programs like MedTech Innovator were actually created to fill that gap, because startups usually don’t have the resources or the network to engage with all the key stakeholders, or even realise that they should. This is often the “middle ground” where consultants come in. But of course, consultants need to be startup-friendly, both in terms of budget and in understanding that startups don’t know everything that needs to be done.
As you mentioned earlier, and I also referred to, we’ve been part of multiple accelerators. MedTech Innovator is our most recent and current one, but before that we participated in MedLim, an initiative by Medtronic’s European headquarters in the Limburg region, together with local organisations in Maastricht. The aim there is to support entrepreneurship and innovation to be developed there. They even provide access to vertical specialists in specific subfields, similar to the way a large organisation like Medtronic is structured.
Still, when we had questions that were more overarching, about how different elements should be coordinated. We might solve one piece of the puzzle, but without the broader context, we weren’t sure we were even solving the right problem or asking the right question. It’s very hard to rely on specialists or consultants for this, because everybody has been so profiled in their own specific field.
It’s really difficult to locate someone who can ask additional questions and perhaps share some if you enquire about something that’s more interdisciplinary.
As part of your joke about me teaching technology entrepreneurship, experience from all generations of entrepreneurs is also beneficial, especially in the medical field. It’s kind of a philanthropic activity for me to share my lessons learned with the new generations of entrepreneurs in a very condensed but still intensive way, without a huge time commitment. I think this is important because it helps new entrepreneurs become more productive and efficient and ultimately pushes technology and innovation forward much faster.
Interviewer: That’s brilliant, and actually very similar to what I’m trying to do as well. I try to share knowledge and give people guidance that could point them in the right direction.
I’ve even worked with some AI companies developing products for clinical research, advising them because, as you said, they often lack real practical field experience. That’s something I can share. I see it as a collaborative effort if people get the right support, we can see a lot of great products reach the market.
I’ll end on this positive note: there are many promising companies out there, and with just a little bit of support, they can succeed and deliver products that will truly change patients’ lives. That’s something important to keep in mind whenever we talk about startups and their work.
I’d also like to thank you again, Georgia, for your time today. This was a very interesting discussion, and thanks as well to our listeners for joining us.
