Name: Brainstorm AI London 2024: Boosting Breast Cancer Detection
Uploaded: 2024-04-16
Duration: 10 min 37 s
Description: Presenter: Dr. Peter Kecskemethy, Co-founder and CEO, Kheiron Medical Technologies

Presenter: Dr. Peter Kecskemethy, Co-founder and CEO, Kheiron Medical Technologies

Brainstorm AI London 2024: Boosting Breast Cancer Detection

Photography by Joe Maher/Fortune Brainstorm AI

Conferences

Thank you very much. It's an absolute privilege to be here. And uh I, I thought about that. Probably everyone here has heard a lot about what A I could do, what word could be, how it could be bad for humanity, how it could be good for humanity. But I'm bringing you slightly different things today. And that is a very good news story. How A I in practice is actually bringing impact right now today. It's not a future, it's not a hypothesis, it's actually what's happening now. And I'm also going to bring you a secret, hopefully that you can take away from today and maybe a bit of motivation. So let's get into that. So my company Kiron has developed an A I called Mia that is supporting doctors in early detection, specifically breast cancer screening at the moment. Uh And that has got quite a bit of news coverage. We were on the front page of the New York Times. We were also named after our results to be one of the seven biggest breakthroughs of 2023 next to CRISPR and Messenger RN A. So in good company and there's a very good reason why that is the case. One of the reasons is we are solving a very hard problem. People may not know that breast cancer screening is, is very, very complicated as a diagnostic task. Uh It's so complicated. Some people say it's like finding a snowflake in a, in a snowstorm and, and and actually the practice of medicine covers that it pretty much reflects it. Uh most countries outside the US, uh two doctors are required to look at every single mammogram. So basically every single image that is done for breast cancer screening. And even if the two doctors, we get at least 20% of the cancers missed, that means at least every fifth cancer gets missed. That's like the best we can do across the globe. And in some places, over 70% of the cancers get missed. This is, this is the best we are doing in screening right now as humanity. And you can see a couple of examples of like a hard case on the left and a hard case on the, on the right, on the left, the cancer was missed and on the right, a lot of the times the doctors look at all the little tiny nodules and they're thinking, OK, which one of these is potentially a cancer? Which one is not, it's like extremely hard. Some people call it magic sometimes. So OK, what can, what can A I do? I'm going to bring you talk, you through this particular case, this is an actual case where a woman went for breast cancer screening regularly. So did the best she could do uh multiple years. Uh and she died of breast cancer and she died of breast cancer because the cancer was not picked at the last screening. It was not picked up two years before and it was also not picked up four years before every single time missed by two doctors. OK. This is, this is reality. And what can, what can the A I do? The A I, we ran the many thousands, hundreds of thousands of cases. What the A I could do is find it at the last screening, find it two years before and find it on the case for four years before. So this could have been over four years early detection. And we know that when breast cancer is detected early, it's over 90% chance that a patient survives. That could have been practically a life saved with A I. OK. So that's the impact what we can have on a single patient, right? So let's see what, what does it mean in reality. So what you see here is how a doctor looks at a particular case. This is a medium complexity case. It's not like a higher complexity case. You see a couple of nodules, but also it's relatively clean. Uh In this particular case, again, two doctors missed the cancer and the A I found the cancer. Ok. This is not a very, very hard case, this is a medium case, but the doctors have to go through hundreds and hundreds of cases every day. There's just not enough energy or attention for the doctors to do their jobs. Well, unless they get good tooling and good help. And I believe a I is one of the fundamental ways we can, we can, we can help. In fact, that's exactly how about what we have shown in live use. So what does it mean? It means we can have extremely strong impact on a patient on the life of a single patient. It also means because we're talking about software, this is automatically scalable. Like you bring a piece of software, a box to a hospital and you can have doctors start saving more lives. That's practically what we're talking about. And so roughly, the impact that we could have that, that Kyron is focusing on is scaling it across the globe. So we're estimating we can probably help doctors save about 2 to 3000 lives every single year in the UK, 40 to 60,000 lives in the US and hundreds of thousands across the globe. OK. This is a piece of software that you bring across completely scab right now today. And this is not a hypothesis. This is based on actual evidence, actual the results that we have right now in multiple clinical trials and multiple multiple assessments across the globe. Um So why is, why is this big, why is it, why is it different from the past? Actually, computer science and computer vision has been used a lot, or at least there were a lot of attempts to be used for breast cancer screening, but not with too much success. And there's a couple of reasons for that, but I'm going to talk, talk you through how we had a slightly different approach. We took two things extremely seriously. And I think the importance here is what can you learn from that? How can you translate it to like how you use A I or how you, how you develop A I and the two components of that is taking A I seriously that it's a very different type of technology than humanity had before and taking the domain seriously, right? Like we're talking about a domain where any kind of decision impacts the life, right? Like either way either you can over diagnose or you can potentially miss a cancer. So taking the A I seriously and taking the domain seriously means that you need to collect a lot of data, you need to have extreme level of evaluation. I just need to make sure that when the software goes to healthcare, to actually impacting every day, the patients and the doctors, it has a positive impact rather than a negative impact. OK? So if you want to know the details, you can look it up, I have a number of papers and nature published and also a number of other publications, but I'm just going to distill for you some of the secret sauce that I believe we all need to need to think about. So hopefully, you can use that in your everyday practice. And the secret here is that A I is not just a model, you have a good model that you think is good, it is not going to necessarily or most likely not going to succeed in practice A I is a type of technology that we have never had before as humanity, right? Like if you have a piece of hardware or software, you know what it's used for, you test it in the lab and too much. That's how you're going to use it in everyday practice your software. But A I specifically, it's designed to be able to be intelligent and good on input and in circumstances that it has never seen before, right? This is a completely different. That's the difference from, from between A I and other technologies. So what does it need? Of course, the models need to be good, it needs to be good in the lab, but also it needs to be good wherever else you want to use it. So the algorithms have to be good. OK, fine whenever, but we need to validate that again, both in the lab as well as in real word. However, we never know because it's a black box, whether it still generalizes, does it still work? If I bring it to a specific hospital, you need to validate that locally in deployment. So that's the deployment stage. And also because it's a black box, it might work one day, but it might not work next week or might not work a month or a year later. You need to keep monitoring that it is still providing the same performance. So you need four things for A I to work in practice, you need to get the algorithms right. You need to validate it. You need to make sure it works in practice both upon deployment as well as over time, right? So please please, if you are impacting lives of it or, or, or uh you're having some kind of very, very important impact with A I do make sure all of those work because uh or all of those are covered. Otherwise it's not gonna be very good. So after that, I'm actually going to leave you with the words of uh one doctor and one patient who, who can tell you about what it means for them uh on uh having used the A I um in the UK. I remember when I was a child, they think they would eventually find a cure for cancer, but it will take a long time. And this is all just progression down the right road. We read 5000 mammograms a year. There is a lot of pressure on the service. I have staff shortages, people are burnt out. We have to make our decision first. Then we are allowed to save the opinion of Mia. So we press the little button, it opens up and shows us yellow circles that says, can you have a second look here. I was quite surprised to hear back from the penny didn't drop immediately until they said they had found a very small amount of cancer and they said it wouldn't have been picked up by the human eye. It was too small, but the computer earmarked it and that's how they found it at the moment we are using in an evaluation process. So we haven't integrated it fully into the system. We have modeled that if we did have ma in our workflow, that would be a big difference because it would make our turnaround time go from probably 14 days at the moment down to three days. The technology helps move us forward. This is the beginning large scale. We can make a difference. It's a lifesaver, a life changer. So I will just leave you with that. That I think when we have enough evidence that A I works, it's actually our moral obligation to use it. Sometimes we need to be very careful. Sometimes we need to go through all the steps to validate it. But when we do and it can have a very strong impact, we need to actually use it And, uh, if you want to learn more, feel free to contact me.