Conversation with Daniel Plews, Product Manager for Digital Pathology at Roche Diagnostics UK & Ireland

“Digital images will always have value, not just for algorithm developers, but perhaps more broadly for the patient in terms of the value that can be extracted over time…

Algorithms of the future put to work on an H&E image of today are going to be able to tell you things that we don’t know today.”

“Digital images will always have value, not just for algorithm developers, but perhaps more broadly for the patient in terms of the value that can be extracted over time…

Algorithms of the future put to work on an H&E image of today are going to be able to tell you things that we don’t know today.”

Conversation with Daniel Plews

Product Manager for Digital Pathology at Roche Diagnostics UK & Ireland

BIOSKETCH: Daniel Plews moved to Roche Diagnostics as an Applications
Specialist in 2016 following seven years’ experience in the histology lab at the
East Kent Hospital Foundation Trust. He became Product Manager for Digital
pathology in April 2021 and now manages the portfolio and lifecycle for Roche
digital pathology products across the U.K. and Ireland.

DISCLAIMER: The views shared in this piece represent the views and opinions of
those involved and do not necessarily represent the views of Roche Diagnostics
UK & Ireland.

 


Interview by Jonathon Tunstall – 18 Sep 2023

Published – 18 Mar 2024


 

Jonathon Tunstall (JT) – Welcome to Pathology News. Today I have the pleasure of speaking with Daniel Plews, Product Manager with Roche Diagnostics.

Daniel, welcome to Pathology News!

Daniel Plews (DP) – Thank you, it’s a pleasure to be here.

 

JT – Well Daniel, maybe we can start by you telling us something about your background and your early career, perhaps specifically, about what brought you into the domain of digital pathology.

 

DP – Yes, sure. I started off in histology about 14 years ago. That was in the NHS, at the East Kent Hospital Foundation Trust in the Southeast of England, working as a biomedical scientist in the cellular pathology lab. I spent seven years working with the NHS in histology. So that gave me a good overview of the entire workflow within a cellular pathology lab; all the way from specimen reception to the report being signed out.

After seven years in the NHS, I felt like I needed a change. I was looking to get into the industry space, and thankfully there was an opportunity that came up at Roche. So, seven years ago, I joined Roche Diagnostics UK as a field-based application specialist covering the Southeast of England. That role covered the entire tissue portfolio from primary staining on platforms such as the HE 600 to advanced staining on the BenchMark ULTRA systems. I was also doing digital pathology with the DP 200, and DP 600 scanners, so really working on the whole tissue process. I had to go into hospitals, make sure that the instruments were set up and validated, and that the staff were correctly trained. After three years in that role, I was promoted to senior application specialist. At that point I started focusing more on digital pathology at a national level, covering the UK and Ireland. Then about two years ago, I stepped into the digital pathology product manager role where I’m now managing the lifecycle of digital pathology solutions here in the UK.

Roche was a very attractive company to work for. Being in the NHS, I was well aware of Roche, obviously, they’re well known in the healthcare setting, and it seemed like a nice, prestigious, and safe role to move into. It’s always a bit daunting leaving the security of the NHS.

 

JT – I can understand that. I have a similar background myself, working for the NHS and then making the move at some point into a commercial company. You said that you have been at Roche seven years now and I’m guessing that you have already seen a few changes because tissue pathology is a very fast-moving market and simultaneously Roche products are constantly evolving. Tell us something about that.

 

 DP – Well yes, Roche has had quite a long history in digital pathology and that stretches back before my time with the company. Obviously, tissue diagnostics at Roche is Ventana and Roche acquired the Ventana brand back in 2008.

Ventana however was founded back in 1985 by Dr. Tom Grogan. At that time, they were selling the early BenchMark instruments for automating immunohistochemistry and it wasn’t until around 2005 when they started to dabble with digital pathology.

Ventana began that journey with an image analysis system which wasn’t much more than a camera bolted onto a microscope, but it did at least allow them to launch some early algorithms for HER2, ER, and PR. Then in 2010, Ventana (Roche) acquired another digital pathology company called BioImagine which was based in California. That’s when they acquired the iScan Coreo, and shortly after that they launched their own scanner, the iScan HT. That’s about the time I joined, and the iScan HT was really our first generation of true digital pathology scanner where you could load a batch of slides for automated capture of the whole slide images. At that time we also had an image management software called Virtuoso and these were the platform I was using when I started to support digital pathology as an applications specialist.

More recently we moved on to our second generation of scanners, the first of which was the VENTANA DP 200 slide scanner launched in 2018. This instrument was a complete departure from the previous scanning methodology due to the learning process with the early scanners. Then in 2020, we launched the VENTANA DP 600 slide scanner which is our much higher capacity slide scanner based on the same optical pathway as the DP 200. We also brought out a second generation of our image management software during this time. That was initially known as uPath and is now known as either uPath, for the on-premise solution, or navify Digital Pathology, which is the name we use for our cloud-based software as a service offering.

We learned a lot in those years moving from the first generation to the second generation of scanners. The issues we saw with the first generation were in part caused by a rack-based system. On paper that seems like a good idea, being able to use a slide rack that’s common with third-party staining and cover slipping instruments. It means that you can take your slides off your cover slipper and click them straight into your digital pathology scanner. That minimizes hands-on time and helps to ensure a clean process, and that’s the way our old scanners used to work.

However, there is an issue with this rack-based system approach, because you need to get the slide from the rack positioned correctly under an objective. The common way to do this is using some sort of robotic arm to manipulate the slide within the scanner, but it turns out that this is quite hard to do.

One important prerequisite to adopting digital pathology is having some sort of slide tracking within the laboratory. Typically, that requires barcodes to be on all the slides and that is usually done with labels. However, if you have 20 different people in a laboratory putting a label on every single slide there is, eventually you’re going to have a label that’s not perfectly straight.

When that slide with the imperfect label is fed into a machine that requires manipulation with the robotic arms, be it a cover slipper or a digital pathology scanner, there is a good chance it is going to be mishandled. The slide could be dropped, and at best, that is going to abort the scanning run, which could be a few hundred slides that have been loaded for overnight scanning. At worst, it could result in the slide being destroyed.

We have to remember that in histology, the creation of slides and the running of the required tests, is a destructive process. You can only do so many tests on a biopsy from a patient until that biopsy is exhausted. Then you have to re-biopsy the patient, which is obviously wasting their time and could even be detrimental to their health.

We now use a tray-based system. That means that once the slides are inserted into a tray by a user, they don’t need to be touched while they’re in the scanner. By making this change, by moving away from a rack-based system, we have removed the issues we were experiencing with automated slide handling. That means wonky labels and not perfectly straight coverslips have become problems of the past. All in all, this new approach has improved the reliability of the scanners and also the safety for the slide, because it no longer needs to be directly manipulated within the scanner.

Another benefit that comes from looking at the slide prior to scanning, is that a skilled biomedical scientist is reviewing the slide. They can make sure that the correct patient sample is on the correct slide, because even in a lab with specimen tracking where there are barcodes, human error does still occur. You still can find the wrong patient sample on the wrong slide on rare occasions, so it’s good to do have a double-check step for patient safety.

It’s also good to do a quality check. Not all slides are equal. Sometimes when a slide reaches a pathologist, they are unhappy with the quality. Either the section itself is poorly cut, so there are a few chatters and scores present, or it could be that the section is not a full-face representation of the tissue that was within the block. Perhaps the microtomist didn’t cut deep enough into the tissue to get that full overview of the section, but by having a quality control check between cover slipping and scanning, you’re making sure that the image captured is always of diagnostic quality. It means that the slide can be removed from the tray, and it saves you from having to rescan. We find now that when we sit down with a customer and have a conversation about their histopathology workflow, the benefits of the tray-based system outweigh the negatives of a slight addition in manual handling.

 

JT – I worked with a scanner company myself from around 2008 and at that time we used a robotic arm which would frequently mishandle and often break slides. There has been a huge evolution right across the scanner industry over the last 10 to 15 years in image quality, in the manipulation and handling of the slide and as you just mentioned, in the labelling barcoding, etc, in the entire process. It’s good to hear that Roche is taking these significant steps forward and to some extent leading this revolution.

 However, you make me think a little here about the market as a whole and the need for partnerships. Roche is a huge company that works in digital pathology as well as many other fields in the life sciences, but it’s still the case that even the biggest companies can’t do it all by themselves. There are so many aspects to this industry; there’s storage, there are algorithms, and some companies specialize, some companies are quite generalist, but I imagine that Roche has also developed some mutual relationships in the industry as well. Perhaps you could speak a little about how you work with other organizations.

 

DP – Yes, indeed. There are two sides to our partnerships. One being the infrastructure side. Roche has, for example, partnered with Amazon Web Services (AWS) for the provision of all of our cloud-based software as a service solution. The reason being that at Roche, we have a lot of specialism and expertise in the platform itself, in clinical diagnosis pathways and in the patient treatment. AWS has greater experience when it comes to the web infrastructure, the scaling of the service and the cyber security and deployment. So, by making that partnership, we can bring our solutions and their concurrent patient benefits to market more quickly. It saves us having to reinvent the wheel.

Also, because we are part of the AWS partner network, we have frequent communication with AWS. We get visibility on their future plans and their road map and we have input on that. By having visibility of their plans, we can use that knowledge to shape our own digital products going forward.

Another aspect to our partnership approach can be found within our navify digital pathology solution. This is our own cloud-based image management software and we use an open Application Programming Interface (API) to encourage third-party algorithm developers to bring their algorithms to our platform. This means that our customers have access to the best tools available in the market and are not limited to only using Roche algorithms. The algorithms that we have publicly announced so far are PathAI and their prostate algorithm, which will be available in North America. We will also have Ibex’s Galen breast and Galen prostate algorithms, which will be available worldwide. These will be launched by the end of this year and we have many other discussions ongoing at this time to bring more partnerships into our solution.

Our single biggest partner is another Roche business, Roche Pharmaceuticals, and they are one of our largest digital pathology customers. Roche Pharmaceuticals have a lot of our scanners and software and they are working on developing new algorithms and new drugs tied to those algorithms.

The idea is that when they come to market with a new drug, that will be marketed alongside some sort of companion diagnostic assay that has to be done first to see if the patient will benefit from using that specific drug. An algorithm will be needed to assist the pathologist when it comes to scoring the test. This is because pathologist workload is increasing, diagnoses are becoming more complex and, from a global perspective, populations are growing and there are not enough pathologists to look at all the slides.

 

JT – Well this topic of companion diagnostics and personalized medicine is a huge conversation in our industry at the moment, and from what you say, Roche is well positioned to be at the forefront of these types of developments. Personally though, I do wonder how regulation will impact the personalized healthcare market. Particularly when a complex algorithm becomes essential to the decision-making process.

 

DP – I think that regulation of artificial intelligence (AI) and artificial analysis tools is necessary. We have to be very careful when it comes to AI bias. We’ve seen this for a long time and since AI started being used for everyday needs. A basic example is a soap dispenser that doesn’t recognize all colours of skin when you put your hand under it.

So, when it comes to these AI tools that have been trained on large data sets of slides, we will need to be careful about where those slides originated from and which populations they are sampled from. Diseases are not equally spread across the globe and the genetics of one population is not the same as another.

So, I do think there needs to be strong regulation to ensure that the new AI tools are fair, will work for all people and, most importantly, that they are safe. I do also think however that histopathologists are going to be quite a good judge of that for themselves. At the end of the day, the reporting pathologist is the one who signs off the report. It is their diagnosis, whatever the methods they use to reach that diagnosis, and they need to be happy that they trust the AI tools.

So, I think there will be some kind of self-regulation, but it is also appropriate to have some more formal rules in place to ensure that standards are equal across the world.

 

JT – Well, we are in a fast-moving market at the moment, and it seems to me that if you look at the requirement for regulation, although regulation is, as you say, entirely positive, it can also put the brakes on development. If we look at the current pace of development of algorithms and of AI in pathology, then regulation, not just in the US, but also in Europe, Japan, the entire developed world, is both very time-consuming and very expensive.

 

DP – It is indeed. Even at Roche, we’ve got our ER/PR and Ki67, and it feels like we’ve been waiting a long time to get those CE-IVD and FDA marked, and as you said, it’s time and money.

That being said, due to our size and the scope of our business, Roche can do a lot of the CE-IVD validation internally and have end-to-end ownership. For example, with our existing CE-IVD algorithms such as HER2 4B5 and HER2 dual ISH, we have been able to do all the validation in house and that is because we create the antibodies, we create the probes, we create the instrument on which those probes are used. We also make the scanner and the software which are used for the analysis. This end to end approach does result in an algorithm that is not as open as some of the others on the market, because you need to use each piece of Roche equipment for it to be ‘on label’, but it does streamline the development process and makes it easier for the customer by removing some of the validation from the lab.

 

JT – That sounds like a very sound approach to minimize the regulatory bottle-necks. Let’s move the conversation on now a little to discuss Roche’s position in the digital pathology AI market as a whole. I go to a lot of conferences and earlier this year at a European meeting, I counted about 18 companies in the room that had some kind of AI algorithm offering.

Clearly this is a very crowded field right now and we can speculate that a number of these products, and perhaps companies, won’t survive in the long term. So, tell me about Roche. How do Roche products remain relevant in such a competitive market?

 

DP – I think our long history helps. Roche has been in healthcare for over 125 years now and hopefully we will still be in healthcare 125 years from now. I think it helps to know that whoever you’re procuring your solution from has a good chance of still being around for the period of the contract.

At Roche, we can also leverage our strong clinical and workflow expertise. There are trusts around the UK that procured a digital pathology solution two years ago, maybe even five years ago, and they’ve never really been able to embed that solution within their clinical workflow. Instruments get left gathering dust on a bench, or in a room down the corridor. We have healthcare consultants at Roche that can assist from both a workflow and a people change management perspective and they can help with both the support and with the building of trust in large projects like this. That ensures that the solution is fit for purpose, that it can be embedded into the routine workflow, can be moved into routine use, and can benefit the patient. We are also not just a digital pathology provider. We provide a lot of other diagnostic solutions across all of pathology, including biochemistry, hematology, microbiology, etc. that means we are very well placed to assist and advise at every point in the workflow.

We now also have a new division within Roche diagnostics, called Roche Information Solutions, which is solely focused on software as a service and information technology (IT) solutions. These IT solutions are concerned with extracting the value in the data that our instruments generate because as more and more tests are made there’s more and more information produced. We are creating digital solutions that can consolidate all of that information in one place. For example, we’re working on an oncology hub so that all the oncology information can be in one place for an oncologist to review, and this is about pulling in all the data, not just from digital scans. I think this is a real strength to be able to consolidate the scan and AI data with sequencing and all the other tests. It gives us the possibility to bring some non-routine methodologies into the clinical workflow.

We are doing these things in collaboration with pharma companies. I mentioned earlier, about developing companion diagnostic tests with algorithms focused on a specific drug. We’re also working on bringing in new testing, so bringing multiplexing into the clinical environment and launching a multiplex clinical assay.

Obviously, a pathologist is used to looking at one, two, maybe three biomarkers on a slide, but that’s usually looking at what’s present or not present. Here we are working on the whole spatial omics side of digital pathology, being able to have highly multiplexed slides, having more information on a single slide, and then starting to look at how populations of cells are interacting with each other. This is not just a question of whether a marker is there or not and this work is difficult and time-consuming for a human pathologist. We are starting to cross the line of what a human could be expected to do in a reasonable time versus something that might take them a week.

 

JT – Other companies are also working in this area and this kind of data and information consolidation is surely going to require an increasing reliance on IT and digitization. How do you think all this changes the pathology lab in the future? Have you got any views on that? I wonder myself, does the microscope disappear at some point?

 

DP – I think the microscope will be around for a little while. Even labs that are 100 percent digital, are not truly 100 percent digital. They still have a few slides and sometimes they need to jump on the microscope, but I do think it’s going to change the role of pathologist. I think that will change in quite a big way.

Currently, pathologists spend most of their time screening, almost triaging cases. Skin pathologists for example have to go through so many benign skin tags etc. to finally get the odd malignant melanoma. So, I think these changes to a digital way of working are going to act more like a force multiplier for pathologists and enable them to spend less time screening and more time diagnosing.

I think in the short term that digitization is going to remove some of the postcode lottery of where your biopsy is done and whether you’ll get the best diagnosis possible. More long term, I think it’s going to be interesting from a pathologist’s point of view, because AI will start doing things that can’t be done with a microscope. As an example, there are already algorithms today that are gauging protein expression, which would normally be done by immunohistochemistry, solely on an Hematoxylin and Eosin (H&E) slide. There are even algorithms that will do this on an unstained slide which a human couldn’t even look at, as the tissue would just be transparent. So long term, I think we’ll see a reduction in the number of slides generated within histology labs because more data will be able to be extracted from a simple H&E. There will also be less need for all the immunohistochemistry work.

 

JT – There is also another aspect to this discussion, because we are to a large extent training young pathologists on digital systems now. This is a new trend appearing in the last ten years or so, but you can already see that pathologists trained on digital systems don’t particularly want to use microscopes and prefer to work for labs which have digitized their workflow. It could be that those labs relying purely on microscopy become a rare entity in pathology purely because of difficulties in recruitment.

 

DP – Yes indeed. Our largest deployment of scanners within the UK and Ireland is in Northern Ireland, where they went fully digital using the VENTANA DP slide scanners. One of their key reasons for doing so, was to make that area more attractive to young pathologists as they were struggling with recruitment.

I was at a conference earlier this year in London, I think it was the Royal College of Pathologists. There was a doctor there from one of the Nordic countries and and as part of her training, she did a cycle over in the U.K. She said that up until then, she had never used a microscope. Her whole training up to that point had been digital. In her training she had reached the stage where she could report on her own without being checked and now she didn’t know how to report using a microscope. That brings up an interesting point, because we are currently creating validation pathways to make sure that a pathologist is safe to report digitally. However, here we have young doctors that don’t know how to diagnose using a microscope and we have no validation in place to check that they can actually diagnose cases with a microscope as well as they can digitally.

I do think it is inevitable that the majority of slides are going to be scanned and screened by AI, and reported digitally. That means that digital is going to become an important part of pathologist training from now on. The sooner all sites can have some sort of digital pathology in place, even if they don’t want to go fully digital today, then at least their trainees can start to learn and adopt the digital workflow.

 

JT – I agree with that, and perhaps we will see a two-tier structure developing in the future, with labs which report only from microscopes providing a reduced service and charging less than a higher-level fully digital laboratory. I think we will also likely see a separation between first and third-world countries. There will be countries in Africa and various parts of the world, which will continue to use traditional microscopy for a long time to come. Consequently, we’ll see a huge difference in the standard of healthcare provided.

 

DP – Interestingly, I’ve heard a few stories from Africa and digital pathology can have the biggest impact in that region, because it gives clinicians there easier access to expert second opinions from Europe or America. This is something that was previously practically impossible. Personally, I think in the third-world countries, we will see much more adoption of very small single slide scanners.

They will be used just for those referral cases, because they most likely won’t have the means to be able to adopt a fully digital workflow.

 

JT – I think that is a very interesting point. Will Roche then be a central part of a global digital future for pathology?

 

DP – Yes, I believe we will. We have a high focus on digital pathology and we’re getting a good Research and Development (R&D) budget allocation into digital pathology at the moment. Roche has always been one of the highest global spenders on research and I think we put between 20 and 25 percent of revenue back into R&D every year. That puts us up there with the Googles and the Apples of the world. As part of that, I would point again to the work of Roche Information Solutions which is really interesting at the moment, because data is the most valuable thing that comes out of a laboratory and we are working to make that data available to the right people at the right time.

 

JT – And of course data storage and having access to historic data to run your current analysis over at some point.

 

DP – Yes, data storage is a hugely important part of digital pathology and it’s probably one of the biggest hurdles. The last time I checked, we still didn’t have firm guidance on data storage. I know most labs here are looking at eight years of storage which is two audit cycles from the Royal College of Pathologists, but I would like to see the NHS come up with a centralized storage solution for their digital pathology whole slide images. I believe the NHS is the world’s largest healthcare organization, so if they were to centralize the storage of whole slide images and standardize slide annotations and slide labeling with information on diagnosis and ongoing prognosis, those images would be highly valuable sources of information for industrial partners who want to develop new algorithms.

Roche Pharmaceuticals already pays for access to images on which to train new algorithms at NHS sites, so if the NHS could centralize the storage, it could be a revenue generator, rather than a cost. Digital images will always have value, not just for algorithm developers, but perhaps more broadly for the patient in terms of the value that that can be extracted over time. Algorithms of the future put to work on an H&E image of today are going to be able to tell you things that we don’t know today. The digital image also doesn’t fade, it will still be there in 30 years waiting to be analyzed. Physical slides do degrade over time, they are not as pink and purple when you come back to them a few years later.

I know there are some organizations working on centralized digital storage, but I would be happy to see an NHS data lake which is all on one standard, perhaps something like DICOM. I do think the NHS has a lot of untapped value in those images.

 

JT – Daniel, that seems like a good note to end on, so we’ll leave it there.

 On behalf of Pathology News, thank you your time today and for a most interesting discussion.

Explore the potential of digital pathology with Roche. Book a demo here.

Share This Post

Leave a Reply