Digital pathology has reached the point where if you don’t have digitized slides, you will not be able to do six out of ten things that other pathologists can do today.
Interview with Dr. Anil Parwani
Vice Chair, Anatomical Pathology; Director of Pathology Informatics and Digital Pathology, Wexner Medical Center, Ohio State University.
BIOSKETCH Dr. Anil Parwani is a Professor of Pathology at The Ohio State University. He serves as the Vice Chair and Director of Anatomical Pathology, Director of Pathology Informatics and Director of the Digital Pathology Shared Resource at The James Cancer Hospital. His research is focused on diagnostic and prognostic markers in bladder and prostate cancer and molecular classification of renal cell carcinoma. Dr. Parwani has expertise in the area of surgical pathology, viral immunology and pathology informatics, including designing quality assurance tools, biobanking informatics, clinical and research data integration, applications of whole slide imaging, digital imaging, telepathology, image analysis and lab automation. Dr. Parwani has authored over 300 peer-reviewed articles in major scientific journals and several books and book chapters. Dr. Parwani is the Editor-in-chief of Diagnostic Pathology and one of the Editors of the Journal of Pathology Informatics.
Interview by Jonathon Tunstall – 21, June, 2021
JT – Dr. Parwani, you have embraced digital pathology since around 2005. What brought you into this domain at such an early stage? Did you have a specific use case or was your decision driven purely by interest in the technology?
AP – We had a use case. I was in Pittsburgh at that time. I was trained in Pathology and Pittsburgh was very big in informatics. We had a multi hospital network with around 15 hospitals at that time but even Pathologists who were part of the network still relied on glass slides being shipped and they could be located two or three hours away from Pittsburgh. Then to get a second opinion slides had to be shipped to a second pathologist, so it could take four or five days at that time to get a second opinion. There were also some hospitals in the network that didn’t have enough work for a pathologist to be there full time, so we wanted to look at ways that a pathologist could connect remotely. It didn’t make sense to have a pathologist traveling between hospitals.
For these reasons we started to look for a system that could send slides, enable remote cover, assist with frozen sections etc. There were only handful of vendors making scanners at this time and there was a need for small scanners to which we could connect remotely with full robotic control. There was also a need for remote pathologists, even in other states and other countries, to give second opinion. We needed to connect with Indianapolis for example, which was 5 or 6 hours away and that wasn’t straightforward with anything on the market at that time. So, we built our own system and deployed a small scanner in another state. We were the first site to supply live patient care using telepathology, and we could go directly to the theatre and communicate with the surgeon. In 2005 this was really something new. We could even bill for the service, we made it a revenue opportunity.
After this, we were able to build on that experience to partner with GE and to create a company (Omnyx) as a joint collaboration. This resulted in a scanner and a product which was probably much before its time and certainly before many other companies. Omnyx took part in a quality study at the time and the results were far superior to other products then available.
JT – Did this experience lead to an early realisation that you could use digital pathology for your MDT meetings?
AP – Yes, it did. When we were looking at the potential applications of these products, there was clearly a need there for improving tumor boards and patient outcomes. Also for helping pathologists who in those days were having to take pictures from a microscope and were unable to participate dynamically in a tumor board. We were one of the first hospitals to use this technology in real time patient care. Back in 2005, we didn’t have such a product for tumor boards but pathologists were already really excited about the possibilities. They absolutely wanted to use these systems for their patient care through MDT. Now, if you look at the subsequent products which have come to the market, they all have this type of inbuilt capability for real-time direct collaborations.
JT – Nowadays we can look at digital pathology and we can point to a number of different use cases. It is interesting to hear that back in 2005, you already had a clear use case due to pathologist travel and geography and that digitization was immediately beneficial in terms of time savings. Do you consider that the benefits of digitization have continued to accrue for your own pathology operation since that time?
AP – Yes, we have come a long way from the early days of the first scanning platforms but I think we are at a similar inflexion point now. Today I can press a button and locate the area of cancerous tissue on a digital image. That is diagnosed by a computer and then verified by me. This is no longer in the realms of science fiction. When I went to the first user group meeting in digital pathology, there was so much excitement around the first generation of scanners, and there is a similar excitement happening now. We have come such a long way and it hasn’t happened overnight, it hasn’t all happened without a lot of trials and tribulations, but digital pathology has now become an essential and valuable tool that we wouldn’t want to lose.
JT – What are your thoughts then on how the Covid pandemic has helped to build that value and to reinforce the need for digitization in tissue pathology?
AP – There were pathologists in my hospital who thought they were at risk or had a risk factor and some also had Covid related childcare issues. They could not travel to the hospital and they couldn’t train the residents. Last April we were in a crisis mode, labs were getting shut down, patients were not coming to the hospital, but we still had to do the work, we still had to train the residents. We reached a point where, if you did not have digital pathology systems, you would not be able to provide the service. We were fortunate because we were already signing out digitally and although there were a lot of regulatory issues to overcome, we got an approval to enable sign out from home. That had never happened before. So, in many ways Covid has accelerated the adoption of digital pathology. Actually, if you think about all the benefits and advances that have accrued since the early days, perhaps the most critical one has been the ability to collaborate in this type of pandemic setting. The driving force has been the necessity to understand the patient diagnosis as quickly as possible whilst pathologists are under these types of restrictions.
JT – I guess we cannot give credit for all these advancements solely to improvements in digital pathology systems. There have also been great strides forward in the companion technologies that support the industry. I would guess in the early days going back to 2005, you would have encountered issues with connectivity and the speed of connections, also with storage and storage costs as the digitized files were still very large back then.
AP – Certainly, there have been dramatic improvements. I remember connecting with another country in 2005. We had a scanner transmitting images which were completely pixelated. It was slow and painful. Some pathologists sat in front of the screen for a while, staring at the pixels and just gave up on it. The infrastructure, internet, politics, governments, these have all been barriers we have had to overcome. Now we live in a world of Amazon, Netflix and YouTube, it is a world of instantaneous information. You can watch a live Facebook event in another country or partake in a zoom meeting and we are not worried about sound or picture quality. In those early days it was easy to give up and some pathologists did actually quit because they didn’t want to use this technology. Some of us persevered until it became a reality, and I can now assess a patient in Taiwan as easily as in the United States. All these technologies have progressed and have greatly assisted digital pathology over the years.
JT – Does some of the renewed excitement you feel for this field stem from the potential we have now to use digital pathology image analysis platforms for actual patient diagnosis?
AP- Yes, I think we are getting to the point where we are now seeing the first versions of true diagnostic algorithms. Right now, I can buy for example, an algorithm for prostate cancer detection and its only a matter of time before other algorithms hit the market. This leads to some very important question however. How do we sell these algorithms and where is the market? Who is going to buy this technology? Is it going to be the pathologist who gets one or two prostate biopsies per week? Probably not, its not worth it for him or her. So, unless the vendor companies are willing to implement a pay per click business model or the company says they will analyse the prostate biopsies for you, it is hard to see the uptake in smaller labs. If the software costs $50K for 1 year and I only need it for 10 biopsies, then it makes better commercial sense for me to seek a secondary consultation from a renowned prostate pathologist. Even to get to this point however, we have to have labs who are willing to have their slides digitized, we have to have a business case for digitization. It’s going to be very interesting to watch how the dynamics of the market play out over the next few years.
It’s important to be aware of the difference in digital pathology workflow to that of molecular pathology. If you think what we need to do to get a cancer specimen sequenced. We have to take that specimen, chop it into pieces, put it in a tube, ship it to the sequencing lab. We will get the sequence and we will get the mutations information; we will find out if this patient has a specific mutation which will help a drug to be successful or not and we can then consider what the prognosis might be. But with digital pathology, they cannot do that. They cannot just ship a glass slide to someone to scan it, digitize it and run an algorithm on, so we have to find a way for glass slides to be digitized either commercially or within the lab and for that image to reach the algorithm lab. It’s a process which requires a lot of planning, and this just hasn’t happened on a large scale yet.
JT – That brings me to an interesting point. Ten to fifteen years ago, people were sounding the death knell of traditional histology in favour of molecular techniques and now there has been a resurgence in the need for histology as it provides spatial context; for example, in understanding cellular interactions at tumor boundaries. Do you think the two disciplines will continue to co-exist from this point forward?
AP- I think we will reach a point where both these worlds come together. People think if you have a liquid biopsy, you could find all the answers in that liquid biopsy. It’s much more complicated than that because we are now learning that in many cancers there are differences even between cells. To make a full analysis you need to sequence each cell and there are now single cell sequencing platforms available. In tumors we see that expression levels change between cells even though immunologically they seem very similar, because cells are actually highly influenced by what’s happening in the stroma. So, In my opinion, even if you take away the pathologist, you will still need tissue to make the connection. Even if you get to a point where you have an algorithm which can quantitate the cells in a field, which can predict how many cells are cancerous, which can predict the average size of nuclei in cancer cells. These things are technically possible today with AI, but what is not apparent and not easy to understand is how do you separate those signal levels and intensities to give meaningful information to align with what you get from a molecular pathology report? It’s always going to be important to look at the tissue and look at the levels of expression of cells within a tumor and to consider what is influencing those expression levels.
JT- Clearly the role of the pathologist is changing with assistance from digital pathology and analytical algorithms. I wonder how far you see that process going. Does the pathologist ultimately become subservient to AI and machine learning tools or can we expect some sort of synergy to emerge between pathologist and algorithm?
AP – The answer to that question is very simple. There are going to be two groups of pathologists in the future. Pathologists that have not adopted digital pathology and who continue to use microscopes. They will continue to be happy to come to work and to be satisfied with their diagnoses. There will be another group of pathologists who have now used digitized slides and are used to making morphologically based diagnosis but now because their slides are digitized, they have more tools available to them. They can now quantitate how many nuclei this cancer has expressed. ER, PR, Her2 etc. How many of the cells are cancerous or not and they can be very objective in their interpretation. They can come to their answers in 1/10th of the time of the other group of professional pathologists and perhaps more accurately. They will now have more time than a traditional pathologist, so what can they do with that time? Either they can do more cases, see more patients or they can enter the world beyond pathology and step into oncology, step into areas where the pathologist would be traditionally left out of such as patient care management, patient care decision making.
But we can also envisage a world where there will be no pathology, no radiology, just diagnostic medicine. In this world there will be a bunch of tools available directly to physicians who can then operate between the boundaries of their specialities or beyond. Certainly though if you are going to restrict yourself to just glass slide diagnosis, you will never be able to make use of digital tools. Its like the evolution of cells phones. At one time we didn’t need them but now we do everything on a cell phone. Digital pathology has reached the point where if you don’t have digitized slides, you will not be able to do six out of ten things that other pathologists can do today.
JT – So in that case, what happens to those labs that are not digitized. You can imagine in 20 years’ time that the younger generation who are very computer literate will say “I don’t want to work in the manual system” and it will become increasingly difficult to employ people in a non-digitized environment.
AP – Yes, I think it’s going be interesting, because there is already a shortage of pathologists. Pathologists are getting older, new people are not coming into pathology and you have all these technologies competing for attention. I think the dynamics of all this is going to shift to the point where in order for a pathology practice to survive, they will need to learn new technologies and recruit new pathologists. It’s the same process that has happened in many other professions, it is the decision ‘should I digitize or not’? Should I incorporate digitization into my workflow or not? Its going to be the same thing. People who have digitized will also then have a means to compete and they will need to advertise their services. We will need to use technologies such as social media and pathologists of the future will have to market their services in this way to compete in the open market. It’s also going to be interesting in countries where there is government funded medicine versus private insurances that pay for it. There are different dynamics and there will be different benchmarks where different stakeholders exist. It’s exciting future and at the same time it’s a scary future.
JT – Your response to these last questions are interesting, because many people in our field would point to a future of reduced workload and reduced need for the traditional pathologist as he or she gives up more and more tasks to automization and is gradually reduced to a role signing off cases. What you are saying however is that you see an expanded role for the pathologist in the future because the pathologist will have time to deal with other areas of medicine.
AP – Yes
JT – And do you think the algorithms will ever become as good as a human pathologist if we consider the immense capability of the human brain in terms of pattern recognition, that combined with years of on-the-job training and experience?
AP – I absolutely think they will get very good. Algorithms will have to be refined. There will come a time when they will reach the same level of performance as a pathologist. Even now, today, when I am looking at algorithms some of the new algorithms, I am amazed by their performance levels. If I had these algorithms today, I could improve my ability to make diagnoses more accurately and save time on tasks that take me several minutes to do, such as measuring the size of a tumor and determining the percentage of tumor present. You should also consider that even though a computer system ,ay not be as good as me diagnostically, it can do a lot of things much better than I can do. I see it as two separate lines that need to be crossed. A line of high efficiency and a line of high accuracy and there will be a sweet spot for a pathologist. A good pathologist who is accurate will maybe not become more accurate with his algorithms but he can certainly become more efficient. A less experienced pathologist may benefit from the accuracy of that algorithm, they may gain some speed and they may gain some efficiency. So there will be many factors to consider and each of the stakeholders, experienced, inexperienced, will benefit from the algorithms either in terms of accuracy or efficiency or both. One pathologist may need both accuracy and efficiency whereas another may need only efficiency.
JT – I guess this goes back to our earlier discussion of the synergy between man and machine. computers are better at some things and the human brain is superior in other areas. Computers are particularly good at counting and screening. If you had a lot of prostate samples coming in for example and you asked to see all the Gleason three and above, you could put them through a screening program and serve up just those specific samples; or you could even just ask for all the doubtful cases. Certainly, in cases of huge workload that must be a positive benefit to be derived from AI?
AP – Absolutely. But I think there will be a point where you will need to find balance between all those things. There will be a balancing which occurs in terms of diagnostic needs against the available infrastructure. For example, how do you make so many algorithms clinical grade when they have been built in a research labs and then roll them out on a system so that any pathologist anywhere in the world can use them. That will require a lot of computing power, that will require a lot of infrastructure. There are many variables that need to be addressed. Not all laps have cheap computing, sufficient infrastructure or the bandwidth to make it happen.
JT – Don’t we also need companies to work together more? There are still a lot of file formats out there for example and not all systems are compatible with other systems. If you envisage a world where pathology is completely digital, when all labs can connect together through cloud platforms using shared services, don’t we first need to go through a standardisation phase in digital pathology?
AP – I agree absolutely. There is a need for those labs that have made the decision to digitize and started the journey to full digitization, to have all the right components in place for the future. It’s not just as simple as buying a scanner from a certain company. Some companies saw this in the early days, Aperio for example. They focussed initially on building a high-quality scanner and a good, digitized image but they also had the vision that the same scanner could be used for clinical work in the future, and that connectivity and open formats would be important. UPMC started to partner with GE and build their own scanner. Even back then it was clear to us that it would not be about just a scanned image, but that the scanner would eventually be just a single component in a future connected infrastructure.
JT – Dr Parwani, thank you for your time today.