Connect Trailblazers: Kheiron

Peter Kecskemethy and Tobias Rijken met at Entrepreneur First (EF) in 2016. On the first day of introductions, they simultaneously singled each other out as kindred minds with similar interests and complementary skill sets. They immediately started working together, and the vision for Kheiron was born.

Fast forward to today, Kheiron is leading the charge on detecting early signs of cancer using machine learning and artificial intelligence. Their first product, Mia, has won several awards, including the first UK Government Artificial Intelligence (AI) in Health and Care Award. In this Connect Ventures Spotlight, Sitar had a chat with co-founder and CEO, Peter, to talk about how it all began, what their long-term vision is, and what deep tech challenges they’ve faced along the way.

Peter Kecskemethy: A big part of Kheiron’s origin story stems from the fact that I spent my childhood in radiology departments. My Mum is a radiologist and my grandmother is in medical imaging; at the time that meant I spent my days at school, and my afternoons and evenings at the radiology department. I may have started programming very early on, when I was 9, but I actually learned how to operate CT machines even before that. I’ve launched multiple startups, and it’s not a coincidence that almost all of them were either in healthcare, or had a healthcare component. I did it because I felt it was not just patients who suffered in the healthcare system, but also the doctors who were trying to help. They suffer because they are not empowered to do their jobs well. They either have archaic support, or no support at all in running the processes they need to treat their patients. That’s pretty mind-boggling. You can find a restaurant in two minutes with your mobile phone, but doctors have to duplicate, even triplicate their work and still can’t find the information they need in order to efficiently support their patients. I felt that bringing tech into healthcare was extremely important, and when AI became good enough, I saw the possibility of creating a major shift, an absolute curve jump for the industry.

Toby and I founded Kheiron in 2016 with the long-term goal of making cancer as manageable as the common flu. That stems from the realisation that cancer treatment, and generally cancer management, is an information problem.

Early detection and diagnosis, staging, treatment planning, treatment tracking, follow-up, etc., these are all information problems. The key question is not “how do you treat cancer,” it’s “what exactly is cancer?” because it’s very diverse, and it’s very changeable over time. Cancer is a dynamic disease and it’s very hard to detect because it doesn’t give the body any signals. That means that all the way from early detection to the late stage, we are starved of the information we need in order to treat it. I don’t think there is a generic cure for cancer, but there is a way to manage cancer, and it’s having access to good information. At Kheiron, we are building the information backbone to make that happen.

PK: Our first product is a breast cancer screening solution called Mia, short for Mammography Intelligent Assessment. We started with breast cancer screenings for two reasons: first, a screening comes at the very beginning of the patient pathway; second, breast cancer screenings are a standardised process — probably the most standardised of all types of cancer — and that presents itself very well for AI. Breast cancer screenings are large-volume; we know what good looks like, and we know what the ground truths are. Even starting at the easiest, most useful point is ridiculously hard, and to say we’re ready to bring Mia to the masses after many years of work is a big step. Not to mention that with breast cancer screenings, there’s no such thing as a small start: launching at even one site translates into mass use already.

PK: We started the company completely focused on the clinical side — solving clinical needs and problems — then had a phase where we were very focused on tech, and now we’re just coming back to emphasising the clinical side. There’s an element to understand here: we’re working with really deep tech. That’s not just the difficult task of getting an AI model to work adequately in a lab setting; it’s getting it to work out in the real world, which is a problem at least two magnitudes larger.

We created the lab version of Mia in three months, and it was good enough. Then we ran an equal performance on a real, albeit small, population, and we ran it over six months. That’s when we got the product to what we call a generalisable level, meaning the AI uses its training to process inputs we expect it to see in the real world. This reduces the risk of systematic performance drops for significant demographics, ethnicities, environmental factors, or even different screening systems. It is extremely important for AI in healthcare to be robust and reliable. Most AI is actually brittle, but in healthcare, we can’t afford that.

As Andreesen Horowitz describes in their article on AI, it’s a long-tail problem. You can cover a lot relatively quickly, but the finer nuances take a very long time for the AI to learn.

That’s the journey we’ve been on in order to go live, and we’re now launching with a lot of clinical evidence behind us, going even beyond what regulations require for testing safety and efficacy. We really care about the safety of our deployment, about getting our software into hospitals and everyday workflows. And we’ve received a great deal of interest lately from those who understand this problem and who have seen our clinical validation.

PK: It took nine months of intense work. We had many discussions with various stakeholders on how to do AI in the UK, the US, and elsewhere trying to find the right avenues, forums, and support mechanisms to cut through the relatively traditional NHS and general healthcare challenges. We were pleasantly surprised that the government and concerned fields were also thinking about this, and this AI Award was specifically designed to support the industry and bring real solutions to these critical, everyday problems.

We submitted Mia and managed to beat out our competition, including a couple of high profile, international candidates. Winning this award means we can roll out in fifteen regions across the UK and garner the clinical evidence required for the adoption of AI.

We have many requests and questions around which clinical trials to do, each of which costs between half a million to four million pounds, and we’re working with the NHS, the National Institute for Health and Care Excellence (NICE), Public Health England (PHE), etc. to figure out exactly what is required, and how we can garner the support and mandate to do it all. This will help us in the UK but also internationally, as we’ll be able to show the world what quality AI development and adoption really looks like. This is Kheiron, the NHS, and the UK government spearheading how to bring AI into healthcare in a big way, and that is very exciting.

PK: I think that understanding how much it takes is near impossible because it’s deep tech; no one has done it, and no one has any clue about it.

The first element is getting investors to actually align with what deep tech means, to recognise that it’s going to take time and effort, and a lot of money. You’re basically front-loading your business problems.

Things like defensibility, differentiation, and value creation aren’t an issue once your product is done. Investors need to accept and embrace that; they should avoid pressing for shortcuts, or pushing founders to launch a smaller, lesser thing. Go for it, do the really big deep tech thing, and reap the reward when it’s done.

The second challenge is training to work with a larger staff, bigger operations, and a broader scale. Deep tech isn’t a mobile startup you can launch with six people. You can’t just find product market fit; you need a team of 20–30 people to get the basic tech right first. Then you can start thinking about product market fit, refinement, and proof. That isn’t the natural course for tech groups and I think it’s something investors can help with — taking the lead in operationally supporting a deep tech company. EF actually does this quite a bit, and it really helped us succeed in the beginning.

Finally, trying to see if your deep tech goal is actually feasible is another tough task. You might not get a clear answer until you’ve already put in a lot of effort, but it is still worth doing.

PK: The technology is very complicated and the insertion into the workflow is complex, but there’s no reason for the product itself to be complicated too. It should actually be super simple, otherwise it doesn’t work. Furthermore, our product love comes from the industry itself: by respecting the healthcare industry, primarily developing our product for it, and being very strict about its purpose and quality, we show how much we care.

Two years ago, we held a clinical trial that we could have gone to market with. We chose not to because we felt the product was not yet good enough for patients, for healthcare, or for the industry. Had we gone to market then, we could have negatively impacted the industry, and we weren’t prepared to run that risk. Now we’re ready and I believe we’re set on a good trajectory.

PK: Yes, I think these principles manifest themselves in our everyday decisions. We’re not willing to ship a bad product and we are intransigent on safety. We’ve had many opportunities to release less meaningful, smaller, or less clinically relevant products — all of which would have been high profile and good for publicity — but we kept our product development on point to make a bigger impact.

Furthermore, we did it in a completely non-traditional manner. We ignored the old school way of tackling these issues, focusing rather on radiology, and meaningful real-life problems. We didn’t speak unless we had something worth saying that would move the field forward. And we understood the environment in which we were bringing AI, the skepticism, and the resistance. I think our principles combined with our respect and knowledge of the problems we are trying to solve are what makes us stand out.

PK: I want to get to the point where we can say cancer is mostly not a problem. When 95% of the time, we can handle this thing. Where people can focus their energy on other aspects of life. I imagine happy longevity, thriving healthcare systems, and scientific progress. And I think it’s totally doable.

PK: I think balancing agency, speed, and action with humility and impostor syndrome in a field where there are so many things we can’t possibly know has been the biggest learning curve for me. Just being a founder is hard enough — and a tech founder, too — because you’re running operations, people, management, etc. and you’re learning so many different things from scratch in a very quick fashion. Being on the frontline of something new, I’ve had to understand that the domain itself is very complicated and lean into that. I’ve had to ask myself many tough questions: Is it that we need to be better? Is it a hard field? What should we generally learn from others? What is specific? That balancing act has been the biggest improvement for me, and I’m still working on it.

PK: So many! I’d recommend a number of things, selectively from Peter Thiel’s Zero to One. Definitely The Great CEO Within by Matt Mochary, and Mastering The Rockefeller Habits by Verne Harnish. I also think it’s very important for founders to read biographies and learn from good people — not as templates on how to succeed, because everyone is going to succeed differently, but how they actually handle their successes and failures. There are thousands of books on how to be a CEO, how to manage, etc. How to survive the journey is a very different story, and I think the way a leader handles the normal ups and downs of a business can have an even bigger impact than traditional ‘how to’ books. I’m also in a CEO group where we support each other and feel less alone in the role, and I think that might be even more important than reading anything.

ST: Absolutely. No man is an island. Thanks so much for your time and insights, Peter. We’re thrilled to support the Kheiron vision and look forward to the day we can all say that cancer is just a speed bump on the road.