Conversation with Jukka Tapaninen, CEO at Aiforia Technologies

aiforia, digital pathology

“One thing I’m excited about for the future is that we have been able to demonstrate the real-world application of our platform and our AI models by helping the Mayo Clinic build a prognostic model for colorectal cancer that can give an estimation of patient outcomes based on the AI data. This model is called QuantCRC and Mayo built the model using our platform. It collects 15 different data points from the slides and has tested it in a clinical study across eight hospitals and 7,500 patient cases to build a predictive model that can estimate cancer recurrence based on AI data.”

Conversation with Jukka Tapaninen

CEO at Aiforia Technologies

BIOSKETCH: Jukka Tapaninen is the Chief Executive Officer (CEO) of Aiforia. He joined Aiforia as an investor and board member in 2015 and became CEO in 2020. Prior to joining Aiforia, Jukka held various executive positions over the past 25 years in several international IT and software businesses, such as SAP and HP. His experience includes leading enterprise software sales for healthcare industries as well as overseeing global business development, building strategic partnerships, and scaling up growth companies.

 


Interview with Jonathon Tunstall

Interview date – 02 April 2024

Publication date – 16 May 2024


 

Jonathon – Today I’m speaking with a leader in the digital pathology industry, the CEO of Aiforia, Jukka Tapaninem.

Welcome Jukka. Welcome to Pathology News.

Jukka – Yeah, thank you.

Jonathon – Perhaps Jukka we could start with you telling us something about your career and what brought you to Aiforia and the world of digital pathology?

Jukka – Well, I started my professional life in the early nineties in Finland on the IT and hardware side. I first worked for Hewlett Packard managing Finland and the Baltic countries and then I moved to their Barcelona Division in a global role. Since then, I’ve been living outside of Finland since 1998 in fact. After HP, In 2000, I joined a couple of mid-size software companies and subsequently spent most of my career at SAP. That was over 10 years of having a kind of a European and a global leadership role. My background is really in artificial intelligence and business process automation in enterprise software sales, so I’ve generally taken leadership positions in those types of companies.

In 2015, I made an early angel investment in Aiforia. The company started in 2013 as a university spin-off and then began to operate in 2014 so I became an investor shortly afterwards. I also took a place on the board because I thought, okay, if I’m investing in something, I would like to have some impact and see how the company is going.

I was a board member with Aiforia for five years, and then in 2020, I took over the CEO role, which was when we started to expand on the clinical side of the business and commercialize some new solutions. I also took the company public in 2021.

Aiforia is a spin-off from Helsinki University and one of the founders is a professor at the Karolinska Institute in Stockholm and back then worked as a Research Director at University of Helsinki. The founders could see that a big transformation was happening in pathology, and it was a similar type of transformation to that which happened in radiology a few years earlier, but, as you know, in pathology, the file sizes are much bigger, perhaps a 1,000 times bigger than in radiology. So, there were obvious challenges in the beginning.

So, Aiforia started off by moving the image management system and all the files into a cloud environment for sharing and storage and so on, but already back in 2015 we could see that the real value is created when you understand what’s on those slides. That led us to become the first company in the world to launch an AI-based deep learning platform for pathology.

This meant we could enable pathologists to build their own AI models and that was really successful in the research and university sectors. Also, one big benefit was that we got a lot of users from Sydney to Oxford to MIT, all of the big universities were using our platform for research.

A lot of scientific publications emerged from the cooperation with universities. and that helped to build credibility for the company so that we could show that we have the technology and that it works. Then in 2020, we decided okay, so now we have this platform, we have about 5,000 users and 400 completed AI models, so let’s look at the clinical opportunities as well, and try to do that without neglecting the research or pharma companies. So, we started to build a portfolio of CE-IVD-marked solutions. We currently have five of those AI models, a couple of breast cancer-related biomarkers, and we also have AI models for lung and prostate cancer.

Now, we’re moving forward with that clinical portfolio, we’ve been gaining the first big flagship customers, and we’re starting to roll out the solutions globally. This is a market that everyone has been waiting for, waiting for it to open up at least. As you know, digitization in pathology started fifteen years ago with the scanners themselves, and now we are adding the AI layer along with other components.

The market has been quite slow to open up. Several large companies, such as Philips or Roche, put in big investments earlier in 2015/16 and then had to scale back a little. Now all the big players are back in the game, and the timing is right. We are now moving beyond the development phase into the commercialization.

That means we’ve been able actually to sell AI and implement our solutions to production. There are positive signs coming from the market, for example, we see that all the tenders are now AI-driven, and it seems that everyone wants to use AI in some shape or form. That drives the demand for image management systems and scanners as well.

Jonathon – Yes, AI is very much forming, as you said, a secondary layer of technology following the initial wave of digitization in pathology which started back in the 1990s when digital scanners first came in. Of course, it is also true that AI generally is now in vogue. It’s part of a brand new wave of technology which is gripping nearly every industry, not just medical imaging.

I imagine that these rapid changes have affected the structure of the company. You said you’ve been CEO for nearly four years now, so what does Aiforia currently look like in terms of structure, number of people and territories, that sort of thing?

Jukka – We have roughly one hundred people on the team, we have a very strong science team and a very strong technology team, and about 20 people in sales.

Business-wise, last year, roughly 60 percent came from the US market, and then Finland, our home country, was around 9% with the rest mainly coming from Europe. We have a few customers in Asia as well, but commercially, the US is the largest sector, followed by Europe.

Jonathon – I see your mission statement is ‘to transform pathology image analysis with AI enabling better care for each patient.’ Does that mean that you see yourselves as purely an AI focused company and don’t plan to offer data systems, management systems, instruments, etc?

Jukka – Well, we do have some image management functionality ourselves, and then we are partnering. If you think about the whole workflow, starting from physical tissue samples through the staining to the scanning and digitization, you need to have an image management system, and then you also need to have AI, and all the different components in place. Some of the scanner manufacturers have some IMS capability, and we also have some IMS capability, Then, there are pure IMS vendors. I think this is an area where some consolidation will happen at some point.

But, yes, we have a pure focus on AI. We have the development platform and the ready-made AI models, but for those to be able to work, you need to have the other assets in place as well.

Jonathon – So I guess collaboration is quite important to your business, you need to be ‘agnostic’ and be able to link up with other platforms?

Absolutely, and that makes a big difference. Many people think that you just build an AI model and implement it, but in the case of clinical use, you need to have a lot of different certifications as a company in order to operate in healthcare.  Then, you need to have certain security certifications and standards in place. There is a lot of bureaucracy that you have to go through. That is doable, it is kind of a heavy lift to get those things in place, but we have done that, and it is all part of being a professional software company.

That also means that when you deliver something, you need to have the connections to the IMS or LIS, to the scanners, and all the surrounding systems in place, and you need to be able to maintain and upgrade your software. This is why I think there are similarities to the delivery of ERP systems in that it’s a kind of process automation case that you need to be able to scale up and continually manage. Just simply building an algorithm is not a solution in this market.

Jonathon – Yes, absolutely, and something that has also struck me about this market currently is the level of competition. I was at a conference last summer and counted 18 companies offering AI based image analysis in digital pathology. Some of those are tiny, they perhaps just offer a single algorithm and then other companies are quite competent in this market. So, let me ask you, how does Aiforia stay relevant and different in such a competitive marketplace?

Jukka – Well, I would say that from a technology perspective, we have a solution that actually does all the quantitative analysis. There are plenty of solutions that are just handling the screening part in order to identify which cases might be cancer – and which not – so as to prioritize your daily work list. In our case, we build an AI model at a pixel level to make very detailed calculations on each image file.

Then, we show the results data visually and build a report that shows the different kinds of scoring, such as Gleason scoring with a prostate case, and we provide information on each of the cells. We try to provide all the information the pathologist needs to help make the decision. We try to make it as easy as possible for them; we do the calculations, and they verify. The pathologist remains in the driver’s seat and can modify the results if they wish to do so, and then there is an audit trail behind that. I think that it is really important from a selling perspective to understand that we don’t just provide a black box, but a tool that shows you exactly where the results are coming from and which allows overriding the AI if necessary before confirming the diagnosis.

I would also say that we go deeper than any other competitor in this area. When we won recent contracts at the Mayo Clinic, in the NHS and the Veneto region, the one key factor was that we could also provide the development platform. In bigger hospitals, they always have their own research side, and they can build their own AI models, but on the clinical side, you also need to be able to automate the workflows. If you can combine those two things you can offer a really complete solution for the customer.

Jonathon – What about regulation? How do you think that impacts the industry? I’m thinking about FDA requirements etc.

Jukka – We just announced that we have started an FDA process and we should get the first steps done with the submission this year.

Currently, our customers can use our products in the US with the Lab Developed Test process which means they can qualify and validate themselves.

In Europe we currently have six CE-IVD marked products available and we are constantly bringing in new.

It is also important to remember that the tools that we are offering are not making the diagnosis at this point. We are supporting the diagnosis. We do the calculations, we show the problem areas, and then the pathologist verifies the result. So AI provides efficiency in the diagnostic process. You can avoid some mistakes, and of course, there are areas that you can observe with the AI that the human eye cannot necessarily detect.

Jonathon – What you say here does tell us something about the future direction of the technology and we have to remember there are a lot of pathologists who really embrace AI algorithms as a tool to help manage their daily workload. However, there are still some that are very resistant and feel that this technology is going to take away their jobs. What is the reality behind that from your perspective? how do you see the role of the pathologist changing due to AI?

Jukka – Well, first of all, the total market is huge. There are 3 billion slides that need to be digitized and diagnosed, and there are limited resources. There are about 100, 000 pathologists in the world, so capacity is limited, and at the same time, the volume of cases is increasing.

That means there is a growing need to fix the issue and I think pathologists have recognized the problem and are actively looking for solutions. Of course, there are always people who are against change, but there is then another problem in that a lot of pathologists are quite old, and young medical students have not historically seen pathology as a hot topic. Possibly, the new technology and AI might help to change that and get the pathology back on the radar as something more interesting and relevant.

One thing I’m excited about for the future is that we have been able to demonstrate the real-world application of our platform and our AI models by helping the Mayo Clinic build a prognostic model for colorectal cancer that can give an estimation of patient outcomes based on the AI data. This model is called QuantCRC and Mayo built the model using our platform. It collects 15 different data points from the slides and has tested it in a clinical study across eight hospitals and 7,500 patient cases to build a predictive model that can estimate cancer recurrence based on AI data.

So this model could be used for making decisions as to which patients may need chemotherapy and how much, and which patients have a lower chance of recurrence. There is a potential to save hundreds of millions of dollars by optimizing the use of chemo to those who benefit from it. So that’s where I think it’s going because without digitization and AI, you cannot combine the patient data and make prognostic models.

Jonathon – You also mentioned the attitude of the younger pathologists coming into this domain. They are much more digitally minded and something that is very interesting is the fact that many young pathologists are now being trained digitally. That does make you wonder if the microscope will ultimately become obsolete in the clinical pathology lab? Even now, there are some labs where several of the pathologists have never been trained on microscopes and that means they have never diagnosed or signed off using microscopes. So, do you think that digitization and the use of AI in pathology is inevitable?

Jukka – Absolutely, no question about it. I think currently, the level of digitization is around 15 to 20% of the total market, but that is growing rapidly. There must be a lot of scanner sales initially, and then you can add the AI. It’s a combination of those two elements. Digitization alone does have value as you can share the files, and you don’t need to send physical samples to a colleague in other hospital to get a second opinion, but AI brings efficiency and precision to the process.

I think this digital world is definitely coming, and Aiforia will play a big part in delivering that.

Jonathon – Yeah, of course, and I guess we then have to speculate that laboratories that either can’t or refuse to go digital will cease to exist. Okay, maybe non-digital labs will continue to exist in third world countries or perhaps we will see the development of a two-tier structure, with a lower fee for a non-digital and slower analysis, but it’s certainly going to be a time of big change.

Well personally, if I had a cancer, let’s say prostate cancer, I would definitely want an opinion from several pathologists and also to have an AI system look at my samples. In some disease areas, the human error rates are far too high, and the AI does bring an element of added quality into the process.

Then, to follow up on what you said about third-world countries, I do think this technology allows those countries to attain a similar quality of diagnosis to that in the Western world. As an example. we are participating in some projects in Kenya where they have very low analytical resources. In theory, tissue preparation and scanning could be done locally and then the digital images could be sent to the cloud for AI analysis.You then get the same accuracy and the same quality of results as everyone else, and that has not been possible previously.

Jonathon – Well, we are speculating here about what may be possible in the future, but it sounds as though Aiforia intends to be at the forefront of these developments?

Yeah, yeah. Absolutely. Absolutely.

Jonathon – Well Jukka, I think that is a fitting point at which to end our conversation today. Thank you for your time today and for speaking with pathology News.

Jukka – Thank you Jonathon

Share This Post

Leave a Reply