Conversation with Donal O’Shea Chief Executive at Deciphex

“A very careful strategic positioning of your businesses, is so important. That’s not to just simply introduce yet another X or yet another Y, but for trying to find a clear differentiation for the service or product you’re introducing to the market and ultimately executing, That is so you can do those things better than anybody else in the market. I think that these two things in combination are what has put us where we are today.”

“A very careful strategic positioning of your businesses, is so important. That’s not to just simply introduce yet another X or yet another Y, but for trying to find a clear differentiation for the service or product you’re introducing to the market and ultimately executing, That is so you can do those things better than anybody else in the market. I think that these two things in combination are what has put us where we are today.”

Conversation with Donal O’Shea

Chief Executive at Deciphex

BIOSKETCH: Donal is an experienced corporate officer in the technology and diagnostics sectors, having 20 years of business leadership experience from start-up through to large corporate enterprise. Donal has considerable M&A experience. He has founded three companies and has senior management experience in Slidepath, Genetix, Leica Microsystems and Leica Biosystems.

 


Interview by Jonathon Tunstall – 19 Sep 2023

Published – 19 Feb 2024


 

JT – Hello all and welcome to Pathology News. Today, I’m interviewing a leading figure in our industry, Donal O’Shea. Donal can be described as a serial entrepreneur having founded three separate businesses in this domain and he is currently the chief executive of Deciphex.

Donal, welcome to Pathology News.

DO – Thank you. Great to be here.

JT – So let’s start by maybe talking about your background, how you came into pathology, into the domain of pathology specifically and how you developed an interest in the digitization of pathology images.

DO – Yeah, that dates back quite a bit. It goes back to when I was an academic, around 1999 or so and I started working initially as a postdoctoral student with a professor of pathology here in Dublin using quantitative pathology – or I suppose as we called it then, image analysis – to measure features in breast cancer biopsies.

I suppose I first got bitten by the bug then. At that point, I think we all realized very quickly that a single image frame or a photograph of a specimen wasn’t adequate, and we needed to look at wider fields of view. So, we began looking into things like stitching and tiling images into a wider field pattern, that is, creating whole slide images.

I think shortly thereafter, we saw the advent of the first digital pathology scanners, such as one from Aperio and so on. And I think the rest is somewhat history from there, right?

JT – Yeah, okay, and then SlidePath came along. I think I am right in saying you founded SlidePath in 2003?

DO – Correct. In fact one of my graduate students was coming to the end of his PhD and we were looking for opportunities for him. Quite naively at the time, we said what a great idea it would be to spin up a company in this very nascent area of virtual pathology, as it was referred to back then. That’s how we founded slide path in 2003 and really with no specific kind of agenda.

It was more a case of saying, well, ultimately, this is an emerging space, emerging technology. Where do we see virtual or digital pathology being deployed? We ended up focusing quite heavily on educational use cases, research use cases and ultimately we had quite a wide scale adoption of our platforms and technologies in both of those markets during the following five to six years.

JT – Well, as you said, it was really quite early days then. So that was quite a brave decision and I guess you were trying to identify the gaps in the market there with the education and teaching modalities. Is that how would you sum up SlidePath? A company very much focused on education and training in pathology?

DO – Well, I think that from my point of view, it was very much directed by the market. I think we focused on concepts like best of breed software for specific use cases. That is instead of trying to build a one size fits all type of product.

Platforms and technologies that were quite commonplace at the time and those were largely associated with hardware vendors who would provide a kind of an entry-level platform for accessing digital pathology materials. We were very much focused on how we take that to the next level in specific niche segments. From our point of view, education was one of those and we did a pretty good job there in terms of scaling and growing the use of digital pathology for education. That was particularly so in the UK universities, and that’s something we’re very proud of.

And also, as we got into the latter half of SlidePath’s existence, we also focused more and more on the research side where we again built what I would call next-generational solutions for pharma life sciences; particularly in relation to the appropriate management of data in a research setting.

We even see this to this very day. Research users trying to shoehorn data into systems that are largely designed for clinical use cases and that is never a good strategy, because their domain is different, their use cases are different, and their workflows are different. On that basis, I think this was probably one of the really important reasons why we gained traction at the time with some of the offerings that we developed.

JT – Yeah. And it’s really about market development and moving with the market isn’t it, because this was a market which was evolving very quickly at the time?

DO – I think it was a market that relied on niche adoption in the first decade of its existence and it doesn’t matter where you look, I think most vendors were kind of drifting with the opportunities that prevailed at the time. As I said, the early adopters and those in digital pathology largely came from the educational research communities. I think again particularly considering the reticence that existed around the adoption into clinical along with the regulatory challenges and hurdles, vendors were left with limited choice and had to to follow that pathway, if that makes sense.

JT – Yeah. And there were those who tried and failed in the early clinical market at a time when Aperio, for example, was selling hundreds of slide scanners to pharmaceutical companies for drug discovery.

DO – Correct, and I think that was a very prudent strategy. There’s value in niche adoption or following the market. I also think it creates some challenges, because ultimately how your platform is conceived and developed can lead to difficulties in applying that product in another particular segment in the future. Another good example is Philips’ pure-play focus on the clinical sector that we saw the early 2010s, particularly as we got into that period where they quickly moved from being a new market entrant to having their first regulated product. They were able to do that because of a very singular and pure-play focus on clinical market adoption.

JT – Yeah. Well, it’s never a good idea in business to fight the flow. You look and say, where’s the niche? What’s the need?

DO – Absolutely and you could also argue that Philips being a late market entrant, has circumnavigated a lot of those early challenges that companies like Aperio or Slide Path had to deal with. There were headwinds that would have existed in terms of pure clinical adoption because the market was just not mature enough for clinical adoption at that moment in time.

JT – Yes quite. Anyway, returning to SidePath, you eventually sold the company to Leica?

DO – Correct. Yeah and I think that was quite a step change for the team. Two years prior to that, I was still teaching here in Dublin and kind of running a company part-time, then two years later, I found myself in a very large corporation.

JT – That must have been a huge change for you.

DO – Absolutely, and I think, there are always those who cannot adapt to that and you see a lot of people struggle with these types of significant changes, In the end I did quite enjoy it.

I was in academia for seven or eight years with having graduate students teaching courses and doing all that kind of thing. Well, I suppose for me, it had begun to feel a little stale, a little political and a little slow for my energy levels, so to speak, right? Do you know what I mean? I found myself frequently bumping off things in academia because of that. I think that the rate of change in academia can be quite glacial and slow. So, in some ways, Leica was a breath of fresh air, and it opened my eyes to a new world and a new set of opportunities. So, for me, I took to it like a duck to water, so to speak.

JT – Well, it sounds like you adapted to that world, but you also did other things at Leica, didn’t you? Didn’t I read that you were head of cytogenetics for a while as well?

DO – Yeah, that’s correct. My initial foray into Leica went from running SlidePath within Leica to running digital pathology within Leica. I was a member of the acquisition team that brought in Aperio technologies back in 2012 and thereafter I switched to cytogenetics to kind of breath some new life into what I considered at the time was a bit of a sleeping giant within the portfolio. You know, we did some great things over a short period of time to grow and scale that business and to bring it to a new level, but coming back to your question on the transition to the large corporate entity, I did find Danaher as an organization to be very well structured, very process led, with a process for everything.

I mean this in a good way, right? Where there were clearly defined goals, ways, and means to measure, quantify, and assess things, with standardization across many areas. I was coming from a startup, which as you say, is largely chaotic one month to the next with a lot of cashflow management, to something that was a lot more deliberate and thoughtful and structured and strategic. Developing those kinds of muscles does very much changes your outlook. You also see the value of M&A in terms of breathing new life and energy into a larger organization where core growths or core growth rates are maybe in the five to ten percent range.

And so you’re trying to layer in high-growth startups, new technologies, new innovations through an M&A strategy. I think that for me was a really, really interesting period, and I learned a lot. Going forward from there, I was able to bring a lot of that learning together and apply corporate level management structure to smaller, more agile organizations. I think for me, trying to hybridize those two approaches has been a very interesting and significant part of the ongoing development of our new startup, well, it’s not so new at this point, our existing startup, Deciphex.

JT – well, being agile is your asset as a small business, isn’t it?

DO – It is and you don’t want to destroy that with process, but you still need some process in order to function.

I think for me, the one thing I learned through my period with Danaher is that view of setting a north star and just continuously going for that north star consistently. So, for instance, we want to be number one in the market or number two in the market in a particular segment. It’s very much then about what’s the strategy that gets you there.

I think the startup environment can be very, very tactical where people are kind of moving left, moving right and I think the key ability is to create that North Star for the team and ultimately build an organization and a structure and a culture that gets you there. That for me is the key takeaway from the corporate period of my career, being able to fuse those two things. The energy of a startup combined with a very strong strategic framework is the winning formula from my point of view.

JT – I think there’s a lot of sense in that. Now, we’ve been talking about the big corporate environment, but obviously you also missed the startup environment because you then went on to found your two new companies Deciphex and Diagnexia

DO – Correct. We almost look at Deciphex as a pillar brand now, with two businesses almost inside of Deciphex. One being our Patholytix research business, which supports drug safety assessment in pharma and clinical research organizations. Then on the clinical side, Diagnexia, which is what I suppose you would call a digitally empowered diagnostics company Diagnexia does thousands of diagnoses a week now on behalf of the NHS and other international clients.

JT – So through the Diagnexia business you’re providing a service? You’re providing pathology resources, analytical resources, are you?

DO – Correct. When we sat down to define a strategy for clinical market entry, we deliberated quite long and hard. Ultimately, we felt that maybe one of the challenges that have beset the industry in terms of driving adoption is the fact that in the clinical environment today, you have multiple information systems and multiple different hardware platforms.

Trying to marry all of that together to get what I would call, an optimal workflow and an optimal view of the data, is very, very challenging. Particularly in the public healthcare sector, you have some archaic LIS systems. I was with a client recently who had a 40-year-old LIS, and then we’re trying to marry that to digital pathology, which is very 21st century in order to generate what I would call a cohesive optimal workflow. That is very challenging. There are also a number of vendors in that space who do, what I would call, enterprise digital pathology (PACS) quite well and these platforms support a lot of the market’s needs in relation to adoption.

So, I think we made a quite deliberate decision that we didn’t want to be any of those things. We wanted to use our technology, married with a community of pathologists, to become a provider of pathology services. That is rather than a provider of digital pathology technology, if that makes sense?

So, it’s a quite deliberate strategy not to be a PACS player, but one of the interesting things that comes about because of that decision, is that we had to build our own combined laboratory information system. As a result of building our own PACS, we find our reporting pathologists really love working on the platform because effectively it covers all of the use cases that the pathologist would expect in their daily workflow, in the most optimal way we can possibly do it.

I think the other real value that we get every day of the week is the fact that we now have 200 pathologists on our network. They literally work in-house and that means they get involved in figuring out the best strategy for the product, where to take it and the evolution of the capabilities of the product. In reality, we simply have a massive team of users. We can talk to these people every day, and they are very frank and forthcoming in their feedback when they need to be. That means we have the ability to evolve the technology quickly and efficiently on the back of the user feedback and the fact we are operating a service.

In reality, we are our own best customers, maybe you could even say we’re our own only customer and that makes sense for the technology we build. On that basis, we can look daily at efficiency, productivity, ergonomics. All the factors that impact data flow through our service. Ultimately, that means we can technically enable the most efficient and most effective service in the marketplace. As an example, right now we’re operating on a two day turnaround time basis for cases received by our accessioning centers from UK clinical labs. That compares to two weeks, which is probably the kind of time frame our next nearest competitor offers. That is down to us owning our own technology stack, and our ability to optimize our processes, our workflows, our ergonomics to such an extent that we’re not reliant on third parties to build things for us and we’re not waiting for the next release of somebody else’s software.

So you can see that the kind of strategic approach we have taken is to be differentiated in the context of NOT being yet another clinical PACS player.

JT – Yeah, absolutely and I’m also thinking that you’re doing this within the context of an increasing number of pathology samples and a decreasing number of available pathologists.

DO – Correct and I would argue that our pathologists are reading cases more efficiently in our system than they are within the NHS or within other healthcare providers on the simple premise that we just have a better technological workflow. A lot of our pathologists would say they are 30 or 40 percent more efficient just through reading on our platform. So if you can think about that, that means our pathologists can be 40 percent more productive than other people’s pathologists.

This is one aspect of it, but then of course, we also now see the growing advent and use of artificial intelligence in the industry and I think we would be clearly naive not to be assuming an AI enhanced future for our service as well. Again, we’re in a position to be able to bring best practice approaches, technologies and tools to the fore and ultimately we can help pathologists be safer and more productive and to meet the supply-demand dynamics that exist in the marketplace. It has to be a combination of both of these things, you can’t just simply introduce technologies that drive productivity alone. If we are going to go faster, we must also consider safety and risk management. We do see AI initially being introduced in this context and only thereafter in the context of productivity. You need to do enough surveillance studies and adequate real world assessment of the capabilities of the tools before you start to start to leverage them.

JT – I want to come on to AI in a moment, but before that, I want to hear something about Patholytix. Now, from what you said, Patholytix is a business name, not a product name. Is that correct?

DO – That’s correct, and Patholytix serves a very different community, it serves drug safety assessment in pharma. Again, similar to the Diagnexia experience, where we sat down to conceive how we could position a product in the marketplace, we thought about the enterprise solutions and what problems currently exist. Then by networking and AI enhancing the pathologists, we can build additional productivity or efficiency into the service. The conventional process in drug safety is where a contract research organization does a preliminary study and that study is then peer reviewed by, let’s say, the pharma organization. We had scenarios where people were shipping slides around or people were flying from one continent to another to sit in a room to read a peer review on a trial study. So, a pharmacologist might fly from Basel in Switzerland to the Midwest of the United States to a CRO site, just to sit there in a windowless room for a week on a microscope, and get on the plane home on a Friday.

They would have had four days or whatever to review that study, or maybe even longer in the case of larger studies. The alternative to that is the shipping slides around which come with more issues, such as slides being checked in and out, questions as to “have I received all the slides?” or “are they stuck in customs?”

We’ve seen scenarios where slide sets have been stuck in customs for a number of months awaiting clearance, and you can imagine this is effectively impacting a drug development pipeline. The safety of that compound can’t be assessed and the pharma companies can’t move forward with their human clinical trials until that box of slides comes out of customs, right? These are very bizarre scenarios, but they actually happen every day and it’s these inefficiencies that are slowing down the drug development pipeline.

So, quite early on in our process, we sat down with key industry stakeholders such as AstraZeneca, Janssen, Charles River Labs, and asked them, “What are the big problems we need to solve here?” It turned out to be about the interconnectivity between these organizations, not necessarily about an enterprise PACS for this or an enterprise PACS for that. It was about how do we bridge organizations safely and effectively and securely without compromising anybody’s data. That’s the major concern in the pharma sector. If a CRO is sharing a study with a pharma company, could that study get inadvertently shared with the wrong pharma or something to that effect.

So, we had to be very confident about our information security and risk management in that respect. Looking back, we’ve taken an industry that was entirely analogue back in 2017 or 2018 and we’ve been the first company to get a peer review status for a study digitally with a regulator and the first company to do primary reviews digitally with a regulator. So effectively, in conjunction with our collaborators and the community, we’ve managed to bring an industry from being completely analogue to a digital state in three to four years.

I would argue however that COVID was a catalyst for this and I think our timing was somewhat bizarrely impeccable in relation to accelerating that transformation. Regulators were issuing emergency use authorizations here, there and everywhere for all sorts of different scenarios at the time simply to enable clinical service provision and the maintenance of drug development pipelines. So, on that basis, the timing was extremely good. We might have been two or three, four years behind the curve had there been no pandemic, but ultimately, we were still in the right place, saw the right opportunity and were the only people who were investing heavily in the opportunity at that point in time.

Ultimately, we lived to reap the reward at the end of the process. The way I would look at it is that we now work actively with the top two global CROs and, considering drug safety assessment is a very consolidated marketplace, we also work with nine of the top ten pharma companies and almost 50 organizational customers worldwide. I think we’re as close to being ‘the market’ as anybody has ever been on anything. Do you know what I mean?

I keep coming back to this idea, but the concept of a very careful strategic positioning of your businesses, is so important. That’s not to just simply introduce yet another X or yet another Y, but for trying to find a clear differentiation for the service or product you’re introducing to the market and ultimately executing, That is so you can do those things better than anybody else in the market. I think that these two things in combination are what has put us where we are today.

To give some sense of the scale and growth of the business, I think Deciphex is collectively heading towards 150 people, which is a fantastic growth since our start date in late 2017. We’re hiring another 70 people in the next 12 months.

JT – That’s fantastic, and as you said as you said, there’s been some luck there, it was a fortuitous time, when you started, with COVID.

DO – I think there’s always an element of luck in any process, any journey, any story. There’s a first adopter, there’s chance meetings, there are pandemics, and all sorts of weird and wonderful things that you could never predict would be part of the journey. Ultimately though, it still requires that North Star, that kind of view of knowing where you are going to go. This is the problem we’re going to solve, and we’re going to stick religiously to solving that problem and not get distracted by peripheral opportunities that pop up along the way.

JT – I think you’ve done a wonderful job and you’re also addressing a real need in the market here. Let’s go back to topic of artificial intelligence because we talked a little about that earlier and you said you expect to enhance your service in the future with AI,

Our team was at the Budapest meeting in March and we counted up 18 exhibiting companies that had algorithms for digital pathology. Some of these are small startups and others are big players in the industry. So, with this in mind, how is Deciphex different? How does your own AI offering stay relevant and different in this crowded market?

DO – It is a fascinating market and we’ve seen probably three-quarters of a billion or so of venture capital deployed in the digital pathology AI market since around 2015. Most of that has been targeted at development of AI tools and approaches and we have to understand what is driving that. I think it all comes back to what you mentioned earlier, there not being enough global pathologists and the global supply and demand gaps which are probably in the range of five to ten billion over the next ten years. I think these factors will continue to catalyze the adoption of AI and ultimately people will come to see that AI is probably the only logical solution to these problems.

There are some regions where the demand and supply gaps are very acute. I think there are parts of Africa where you have less than 3 pathologists per million people, and that is completely unsustainable from a quality of healthcare point of view. Many other large parts of the world are not that far behind. Even very well-developed economies like China struggle with the number of active pathologists per capita that are available to support diagnostics.

I would say that those are the markets that need the most rapid help. I think a ‘good enough’ solution could be adopted at the low end of the market first – and I don’t mean ‘the low end of the market’ in a disparaging way. I think if you’ve got three pathologists per million people and an AI that performs at 95 percent, the patient is probably getting a better quality of care on the premise that AI would at least review biopsies and review materials rather than nothing being done at all.

I think that over the next decade, we will see adoption in those emerging markets to a greater extent than other markets, and that is for the very good reason that there is an absolute need. We have already seen that regulators are capable of responding to this type of need.

COVID which we discussed earlier, is a good example where we had an environment where there was an incredible need to change regulatory positions quite rapidly. Consequently, regulators responded to that with emergency use authorizations. I do think in some of these emerging economies, we will see very similar trends in the context of AI adoption as a primary screening methodology or as a primary tool for case triage. That will help at least to get the right cases in front of the three pathologists per million people that exist, right?

Similarly in China, I’ve heard some really interesting stories. For instance, AI-based cervical screening is being used routinely now and is seen as the standard of care. That is because, there just aren’t enough cytologists available to screen all of the biopsies that are taken in China.

So, there are all these markets that ultimately have a real imperative to use AI, but coming back to your question about the pure-play AI organizations, I do think many of them are going to struggle. I don’t personally see a landscape where all will exist in five years’ time and there are multiple reasons for that. One reason is that AI will become commoditized over time. What I mean by that, is that the problems that look very difficult to solve today, will be easily solved in five years’ time. Data will become the key driver and ultimately the organizations with the biggest data collections will stand the best chance of building long-term confidence and capabilities. Often those organizations are technical institutions that can provide access to that data.

I also think that all of these organizations, that is, the pure-play AI guys, are stuck with the challenge of having to compromise their own deployments. Similar to the conversation we had earlier about PACS and LIS. They will generally need to work through a PACS and they’re going to have to somehow sit behind that PACS and provide a visualization of their AI data through that PACS. That is not what they want to give their customers either, because it’s going to be suboptimal the PACS will ultimately control how their data is visualized. I think we will see a number of these companies starting to pivot a and build their own AI-empowered PACS, but again, the adoption curve on that is going to be another long one.

So, I do see a lot of these companies having significant challenges as we move forward. Our own strategy again is “to be our own best customer”, and we work very closely with one of our CRO partners, Charles River. That is in terms of co-developing AI for safety assessment and through that we have access to incredibly large data sets. In the next six months, we’ll start to productionize the first outputs from those collaborations.

The ambition and vision are quite significant. If you think about safety studies, a typical project generally involves around 40 tissues across a single or multiple species. If we think about looking at and grading a specific type of cancer in a specific tissue. Then consider extrapolating that to maybe 30 to 40 lesions by 30 to 40 tissues. You can see the scale and magnitude of the problem in a safety arena, over that for, let’s say, a clinical setting. As an example, our initial foray is a 70-location, 10 legion model across seven tissues in rodents.

We’ve learned a huge amount through that exercise. We’ve worked collaboratively with Charles River in terms of building and validating the models and in assessing their utility to the pathologists. We plan to extend that by maybe another seven tissues and another 70 locations in the next iteration.

Similarly, on the clinical side, through collaborations with several of our customers, we have started to build anonymized consented data sets for AI model development and validation purposes. What we see in the clinical realm is the antithesis of what we see in the pure play AI world, where the problem is in the volume end of the business, the low complexity end of the business.

Our goal is to have 80 percent of all samples that go through our service being screened with AI by the end of next year. What we mean by screened is screened for high-grade disease. That’s a safety check and we feel that safety is the key aspect of how we will deploy and build confidence around AI in the marketplace. It’s not about augmenting the pathologist’s view, it’s just about making sure that there’s no discordance in diagnosis. Does that make sense?

JT – Yeah, absolutely.

DO – If we find a discordance, we can then go back and verify or put that into an EQA program or whatever. The QA program is our own service, so there are lots of ways in which we see ourselves using AI that might be very different to the way some of these pure play vendors conceive AI on the basis that they’re not running a service. They are not trying to manage the challenges that come with a service day in and day out.

JT – Yeah, exactly. You’ve made a few really key points there and you have highlighted how this is a disruptive technology which is moving very rapidly. You’ve told us the reasons for that and how Deciphex is at the forefront of those changes, but you also mentioned the commoditizing of AI. In that respect, I wonder if you see a threat from generative AI programs, such as chat GPT and its successors, for instance? How do general developments in AI impact this market in the next ten years?

DO – I’ll put it this way, our business model isn’t threatened by generative AI, but I would consider that other people’s business models are entirely threatened by that.

JT – Yeah. I mean, it must impact the industry at some point somewhere.

DO – In my view, yes, but in a positive way. I think it’s those organizations who have thought through how more and more AI is integrated correctly into a workflow or how AI is integrated into a service paradigm who will ultimately be the winners here. As much as AI performance will increase and be enhanced and will improve on specific indications, the important thing is the context of use. We need to consider how particularly mature economies with strong regulators and the rigorous regulation of both services and products in this space will create an appropriate pathway for these organizations to allow judicious application of AI in the context of service provision. Again, I think both having a service and running a service gives us an unparalleled view on the kind of optimal strategies and approaches for how that gets done.

We will continue to develop really good data sets, continue to develop best-of-breed models but, as I keep coming back to, the science and the technology will in my view, be largely commoditized over a 15 to 20 year period. That is because, as you say, generalized AI will come to the fore and we even see that today with image recognition programmes. If you go on to Google’s image recognition application and put in a colorectal biopsy or give it a particular lesion feature, like an inflammatory process of the colon, it will actually find similar images.

So, we’re not too far away from what I would call both generalized or multipurpose AI that can be specified for a given use case. As I said earlier, I think an organization’s ability to maintain this kind of vertical niche will largely depend on their understanding of the regulatory process, the AI workflow and other aspects.

JT – Donal, that’s probably a good point on which to bring this to a close, so we’ll leave it there. Thank you for speaking to me and thank you for your time today.

Share This Post

Leave a Reply