The App Store for Medical Devices Is Being Tested Right Now

There are relatively few biotech startups in Japan.

Few investors are willing to write the multi-million dollar checks and have the decades-long patience that is required to really succeed investing in this industry.

But startups find a way, and an innovative biotech ecosystem has started to develop in Japan despite the lack of traditional funding. In fact, we might be seeing a new, uniquely Japanese, model of innovation that we'll call "the innovation supply chain".

Today, we get a first-hand look at how this innovation supply chain functions, as we sit down with Yuki Shimahara the CEO and founder of LPixel. LPixel uses AI image analysis to detect potential problems in patients MRI and CT scans.

The technology itself is fascinating, but Yuki and I also talk about how medical research and medical innovation might be taking a very different path in Japan than it is in the West.

It's a great conversation, and I think you'll really enjoy it.

Show Notes

The real problem with using AI for medical diagnosis AI's deep roots in medicine How safe is medical AI, both in theory and in practice Are we about to see an App Store for medical devices? Why doctors have mixed feeling about AI in medicine How to maintain a competitive advantage in a crowded AI marketplace

Links from the Founder

Everything you ever wanted to know about LPixel Connect with LPixel on LinkedIn Friend Yuki on Facebook

Leave a comment Transcript Welcome to Disrupting Japan, straight talk from Japan’s most successful entrepreneurs.

I’m Tim Romero and thanks for joining me.

You know, we’ve talked a lot about biotech in Japan on this show before and quite a bit, really. We have gone into the fact that the Japanese biotech venture ecosystem is really being held back by the lack of investors willing to write of the large checks required knowing that they won’t see any returns for over a decade. So, things are hard for life sciences in Japan.

However, in the words of Dr. Malcolm, "Life finds a way" or in our case today, "Life sciences find a way."

There’s a growing number of impressive life sciences startups emerging in Japan and they are adapting it and evolving so that they can innovate within the capital constraints they find themselves in. Today, we sit down with Yuki Shimahara, founder and CEO of LPixel.

Now, LPixel applies artificial intelligence to medical imaging and detects a wide variety of conditions from CT scans and MRIs. Yuki is still a PhD candidate at the University of Tokyo but he is running a company with more than 40 employees, so you can imagine, he is a pretty busy guy, but he took some time to sit down with Disrupting Japan and talk about how AI is being used in medicine, the challenges facing life sciences in Japan, and between the two of us, we sketch out a new way forward for Japanese innovation, an innovation model that is distinctly different from that in the US, but that might just be the way forward in Japan. Oh, and as you know, my goal here at Disrupting Japan is always to bring you amazing insights from Japanese entrepreneurs in their natural habitat.

This week, that habitat was a large concrete wall to conference room that makes it sound like we are talking at a vast underground cavern. It sounds a bit odd at first, but if you join us for the next 20 minutes in our underground layer, I guarantee you that you will leave thinking very differently about life sciences in Japan, but you know, Yuki tells us that story much better than I can, so let’s get right to the interview.

[pro_ad_display_adzone id="1411" info_text="Sponsored by" font_color="grey" ]

[Interview]

Tim: So, I’m sitting here with Yuki Shimahara of LPixel and things were sitting down with me today.

Yuki: Thank you.

Tim: So, LPixel is a cloud-based AI image analysis that you are using mostly for life sciences and related research, but you can probably explain it much better than I can.

Yuki: so, LPixel is a start up from researcher realm in the Tokyo University, so our major is bio-image infomatics, so our core value is we combine life science and image analysis. Due to the evolution of CTs and MRI, and microscopy and so on, we have many, many data these days. We do bio experiments and develop microscope, and then analyze image data.

Tim: Well, I think image data analysis, in general, is one of the most interesting areas of AI in machine learning right now. LPixel offers a dozen different products and services, so are these products meant to be kind of different pieces that can integrate together, like AWS? Are they one-off independent products that different customers would sign up for? How exactly does the system work?

Yuki: Okay, so let me explain the one example, image analysis for medical image diagnosis. So, we provide the technology a name of EIRL, so EIRL is image diagnostic support system for medical doctor. So, for now, only human medical doctor diagnoses, CT and MRI images, and the number of diagnostics is increasing. Our solution is very simple: we provide AI for them, we would like to decrease misdiagnosis and medical expenses as well.

Tim: Okay, actually, let’s dig into that right now because that’s interesting. So, I know you have been working with National Cancer Center of Japan to better detect cancer and other types of diseases, so what is involved in getting certification for a diagnostic tool like this? Medical diagnosis is a very different type of business than most AI imaging.

Yuki: To be honest, AI is very new technology for medical doctor. This kind of AI defined as medical devices, that means we need to get approval as medical devices. Yeah, so we need to do the clinical tests, and then we need to make medical doctor decide to use or does not use our system.

Tim: This is what is interesting because I mean, before, you mentioned alike, AI is a new application for medicine, but actually, medicine was one of the very first applications for AI going back into light, the early 80s, the expert systems, so I mean, it has a long history with medicine, but it doesn’t seem that it’s really made an impact yet, and so like at the clinical trials, is there a specific number you have to hit? So, for example, does it have to be 98% accurate with .01% false positives or is there a specific number you need for certification, or is it more complicated than that?

Yuki: oh, that’s a good question, so yes, the answer is very complicated. So, we cannot decide the exact number because it depends on diseases and it depends on yes, do we want to say for getting approval, sometimes, very difficult.

Tim: Medicine is such an unusual field because in some ways, it is so data-driven and in some ways, it is very vague, and so when you’re looking at a diagnostic device, not just a medical device, but a diagnostic device which is even harder to get certification for, so if we are saying something like diagnosing a melanoma, is there a baseline, the we know that medical doctors are accurate 95% of the time? Is there a baseline or is that kind of unknown?

Yuki: It depends on diseases. 100% is impossible in medical field, right? In some diseases, 80% is highest. It depends on the diseases. So, we need to prove that this number is reasonable, so need to compare the human accuracy and the AI accuracy, or I need to sometimes compare with only human doctor or human doctor with AI.

Tim: So, there is no specific number you have to reach. It’s up to you to say, this is our number, and this is why that number is good enough, you just have to convince the regulator?

Yuki: Exactly.

Tim: All right. So, how do you see AI working with medicine? Do you see it as the AI would tell the doctor, hey, pay extra attention to this? Do you think the doctors would ask the AI to double check their work? How do you see doctors and AI working together?

Yuki: As a first step, I think the AI is just supporting diagnosis, so it is kind of the tracking system for now. So, sometimes, two medical doctors diagnoses one patient, but I think it can be the change, the medical doctor and AI.

Tim: You see AI maybe someday acting sort of a prescreening? So, for example, right now, if someone gets an MRI or a CAT scan, the radiologist has to look at it and analyze it. He may not know specifically what he is looking for, but easier time where we might have like, 300 different little AI programs, like this one is designed to detect pancreatic cancer and this one is designed to detect this type of two more, DC every Scan being run through all of these different algorithms, and that maybe after that, the doctors, they can give their advice to the doctor and say, look for these things?

Yuki: Yes, I believe that only one company cannot cover all diseases, so our company is kind of an app vendor, and I think that can be a platform. It is like an app store, so we provide the app on the platform and we can get the profit after paying the platform fee.

Tim: What has been the reaction from the doctors themselves towards AI technology like this?

Yuki: The questionnaire from around 2000 medical doctors, it says around 80% of medical doctors is interested in using AI, but only 1 to 2% of medical doctors use AIA.

Tim: why do you think – because that’s a huge gap? Why do you think that is? If there’s so much interest in it, why aren’t doctors using it more?

Yuki: One thing we think, we need to think about user interface and the user experience, and kind of workflow. For example, medical doctor is very busy, so they don’t want to use new application. I think that we need to integrate our AI, the software or workstation, so what medical doctor used.

Tim: That makes sense, that makes sense.

Yuki: Yes, so that’s why we have strong relationship with the vendor, like Fujifilm, number one share in Japan, and Canon and so on.