Informa Tech and Omdia's Felix Beacher interviews Antony Rix, CEO of Lucida Medical

Can AI systems outperform radiologists in diagnosing cancer? If so, why don’t we see such systems in clinical practice? What does the future hold? And would you rather be assessed by a human physician or an algorithm?

In an episode of the brand new Medical AI Podcast, Felix Beacher, head of healthcare technology at Omdia and Informa Tech, addressed these and other issues with Antony Rix, co-founder and CEO of Lucida Medical, a company developing software to help detect prostate cancer through MRI scans.

Listen to the podcast:

The following is an edited transcript of their conversation.

Felix Beacher: Lucida Medical is a really interesting example of a small company breaking into medical AI. Could you introduce the company?

Antony Rix: Lucida Medical focuses on developing AI to help find cancer in MRI scans. MRI is great, it's non-invasive, it's very safe and it's now pretty ubiquitous. Most medium-sized towns will have an MRI scanner. And what's been really intriguing is that medicine has found ways over the last 20 years of using MRI to allow you to actually visualize abnormalities inside the body and find cancer. The vision behind the company is all about that.

Beacher: What gap does your product address?

Rix: We're helping to solve a long-term problem and a short-term problem. The short-term problem is that, as we as we all know, health care systems are greatly overstretched. And if we look at how cancer is being diagnosed, we're really struggling to deliver the services at the moment.

We've got a specific opportunity in prostate cancer and chose to focus on that specifically because there's a need now to deal with the COVID backlog of prostate cancer patients. We know radiology can find prostate cancer using imaging, but only the best elite and super-experienced radiologists can do it quickly and easily and accurately. What we're trying to do there is distill their expertise and make it available to everybody.

There is a longer-term goal − if we can make this work, screening for disease non-invasively using MRI, we've got the potential to find a whole basket of different cancers using AI. We need that desperately because right now we find cancer far, far too late. About half of us will develop cancer in our lives. And in about half of those, cancers are found when they're advanced or metastatic. Survival rates are not great in that situation. If you can find cancer earlier, then it gives us a chance to completely cure it.

Beacher: Why did you choose prostate cancer in particular as the focus?

Rix: The need here is really pressing. Prostate cancer is the most common male cancer in the developed world. About 1.4 million a year were diagnosed in the run up to COVID. About four to five million a year worldwide go through the process of diagnosis.

Historically, we only find about 50% of prostate cancers using the traditional screening test. Using MRI, we can make it a lot more accurate. But today hospitals are struggling to roll that out. They're struggling to implement a mandate now to do MRI prior to biopsy for prostate cancer. And this gives us a commercial opportunity to introduce software that does that better and that aligns very nicely with the needs to help relieve the burden on health care systems and give patients access to better care.

Beacher: How close are you to having a product in clinical use?

Rix: We're quite lucky because this is software, and it's used in a way that's actually very safe. It's used to assist radiologists and there's always a clinician overseeing what the software does. We were able to get our first CE mark regulatory approval last year and we have a pipeline with further regulatory approvals underway.

And we've been lucky to already have had the opportunity to improve the software working in a simulated clinical setting … and that has given us a real impetus. It's shown that radiologists like the software and it's possible to integrate it technically in a way that's very usable.

We're now working with a group of leading doctors in both NHS public health care and in private practice to start trialing the software in their own clinical pathways. And at the same time, we're building further clinical evidence by running ongoing clinical studies to really provide additional justification for using the software and make it really clear that it's both safe and effective.

Beacher: When would you expect to release your software?

Rix: We're working to complete that this year in 2022.

Beacher: What are the main challenges today for companies in this area is trying to get their products approved.

Rix: One of the biggest barriers is medical device regulation in Europe. And we're seeing quite a few companies actually go to the U.S. first because the regulatory landscape can be significantly easier especially if somebody else has already tried to crack a similar problem before. In Europe, the introduction of the Medical Device Regulation in 2017 that came into effect last year does make the process rather more complicated.

But the biggest challenge there has been that it's overloaded the notified bodies, like BSI, who are responsible for doing the approvals. So, companies large and small are currently facing a backlog of introducing medical devices through that approvals pathway in Europe. It’s actually one of our strong suits because we completed the first approval last year. Approval is a prerequisite and is something that slows people down.

The next challenge is getting the decision makers to accept that the product − irrespective of whether it's got an approval – is actually good and cost-effective. And there our focus is really to make sure that as well as having the necessary regulatory approvals … you've got to provide a patient benefit. You've got to have cost effectiveness. And you've got to make it as easy and frictionless for hospitals to integrate your technology as possible.

Historically, companies that have often come from the perspective of developing algorithms have not necessarily focused on those areas. And it's really important that you do right from the beginning.

Beacher: Could you describe how your software would work? How would a radiologist use it and what would come out of that use?

Rix: With Lucida software, a patient is referred for investigation and diagnosis of prostate cancer just as they would be today. The first thing that they have is a PSA blood test and an MRI scan. The results of those two are fed into our software, which analyzes the images and the test results and is able to identify firstly whether the patient has an overall risk of cancer.

And secondly, if there are any potential cancerous lesions that ought to be considered by a radiologist and other clinicians for biopsy. The end result is we're giving radiologists everything they need in terms of images, 3D models, risk scores and the template report that they can use to complete their reports, and take that to their clinical colleagues to decide whether a patient needs a biopsy. And if they do need a biopsy, where to target that biopsy. The results of that will then determine whether the patient needs treatment and how best to treat.

Beacher: Is there any evidence you've gathered that indicates the success of this system could outperform a human radiologist?

Rix: The software doesn't need to outperform a human in every application to make a clinical pathway more effective and safer or more cost-efficient. You often only need to solve particular problems. In our case, we're targeting the problem that's the easiest to describe: How do you find which patients are normal and don't actually need a biopsy and don't need treatment? That's quite easy to describe but it's actually technically quite difficult to achieve in practice.

The early results we've found so far indicate that we have very high negative predictive value − in the high 90s percentages − and what that means is if the software says a patient is low risk, it's very likely that radiologists would agree and it's therefore safe for the software to be used as part of a pathway where your first decision is which patients can I send home now and reassure them that they really don't have cancer.

By doing so, you might be able to cut the workload for investigation and treatment by a factor of two, three, potentially significantly more, save quite a lot of money in terms of health care resources and free up a lot of clinician time so that they can focus on what they enjoy most actually, which is giving patients with cancer the best possible care.

Beacher: Is your system doing anything that a human radiologist couldn't do? How is it adding value?

Rix: Firstly, radiologists are terrified about missing cancer, and we know that they do. A Cochrane meta-analysis of eight major clinical studies showed that radiologists find about 85% to 86% of prostate cancers, which means they're missing 14% or 15% of patients. If we can help reduce that gap by helping them avoid accidentally overlooking something, we can really make a significant difference.

Secondly, it's really difficult to rule patients out because in practice, normal is a continuum and cancers can often look quite similar to some benign conditions. It's not easy for humans to consistently make those judgments. But an AI that can be trained on thousands of cases and operates in a completely repeatable way has potentially got an opportunity to do better.

The last thing we can do is we can really fuse information in a way that a human can't. We can look at all of the MRI images at the same time. We can look at subtle image features, or we can look at the clinical history or really focus the algorithm on making a decision that's optimal in a way that a human never would.

The goal here isn't to make an AI that replaces humans at all. It's just to make the humans more productive, and allow them to concentrate on the interesting clinical challenges rather than the routine stuff.

What we've been told clinicians love about our software is that it highlights the areas of suspicion that they need to look at, which makes them reassured when patients are negative. They've got the calculations, they've got the volumes. They can see that there are no areas of risk and it helps them and ultimately I think it will help reassure the patients that these individuals don't actually need further investigation.

Beacher: How do you see AI imaging evolving in the next 10, maybe 20 years?

Rix: It becomes just a routine part of clinical practice in the way that ECGs used to be read by hand 25 years ago. Now a computer looks at the heart rate and can identify a whole range of abnormalities completely automatically. And then that allows us to move on to analyzing more relevant clinical questions.

Beacher: Do you see medical AI in general as involving ethical risks and if so, what do you think?

Rix: We're already putting in place the processes to monitor how our algorithms are used in clinical practice, so that we can make sure we're not in some sense advantaging one group over another, and overall, to make sure all of the patients whose health care condition is assessed with our algorithm are given the best possible care.

Beacher: I wonder whether we, in the discussions we tend to have about the ethical risks of AI medically, sometimes miss the idea that there are ethical risks with not having medical AI. We've developed a technology that could save lives and if we really put the brakes on it because we're worried about the risk, we are missing the fact there is a big opportunity cost to that kind of reticence.

Rix: The status quo is often the most dangerous thing and not introducing a medical intervention can often lead to real harm. We all have a duty, when thinking about health care, to keep an open mind and consider what's ultimately going to be best for patients and most cost-effective for health care systems.

Back to prostate cancer: There's enormous regional variation in the quality of care for it. The expertise to interpret these MRI scans is not widely available. And if we can help narrow that equality gap we'll actually end up bringing a large proportion of the population up to a level of care that today they might only achieve in the best academic centers.