Conversation with Professor Alexi Baidoshvili, Clinical Pathologist, specialized in digital and computational pathology

In 2015, we became the first completely digital pathology laboratory in the world.

I believe pathology will follow the same process as radiology and will become entirely digital within a decade or so.

Interview with Alexi Baidoshvili

Laboratorium Pathologie Oost Nederland (LabPON)Hengelo, Overijssel, Netherlands.

BIOSKETCH: Alexi Baidoshvili is a physician who graduated from the VU University, Amsterdam and has specialized in clinical pathology at the VU Medical Center in Amsterdam. The subject of his doctoral research was digital pathology and he was the Project Director of the Digital Pathology Team of LabPON between 2009 and 2017. Now, after complete digitization of LabPON he is the Project Director of the Computational Pathology Team of LabPON. Internationally, Alexi is an active board member of several organizations and organizer of numerous international annual conferences.

Interview by Jonathon Tunstall – 16 July 2021

Published – 07 Oct 2021

JT – Dr. Baidoshvili. You are a career pathologist, at what stage in your career did you specifically discover digital pathology?

AB – During my education, I did some practical scientific work at the VU University Medical Center Amsterdam. After one year, I began a specialization in pathology as part of my medical training, and I was doing some research. I was using some computer analytical techniques for quantitative pathology, and I really enjoyed the combination of the computer work and the pathology.

When I finished my PhD and I had finished my education as a pathologist at the VU University Medical Center Amsterdam, I had some possibilities to work in different cities in the Netherlands. The lab I am with now, LabPON, was the one I chose because they asked me to play a leading role in a digitization process. LabPON was, at this time, a non-profit organisation and had the funds available to be able to carry out the digitization work. Due to my knowledge and experience with digital pathology and digital imaging research LabPON asked me to handle that project, which was an extremely interesting one for me. So, I made the decision to move from Amsterdam to Enschede. In 2009, I started work at LabPON and immediately began this digitization project.

It was very interesting and very innovative work, and we had to do everything from scratch. We found a good partner in Philips who was just starting at that time. They had a scanner and an image management system (IMS), but I felt that it was not yet good enough for routine diagnosis. However, I could see that their system was good enough for research, and as our goal was to build a system for routine diagnosis, Philips agreed to become a partner with LabPON for the development of our routine system. I took the role of project leader.

We started work in 2010, and in 2015 we became the first completely digital pathology laboratory in the world. It was a lot of work. We had a lot of challenges, but we did it. Of course, over those five years, we made a lot of mistakes, and we solved a lot of problems. We had to make changes to the IMS, the scanner, everything. It was a unique experience for our laboratory and especially for me. Afterwards, we had a lot of visits from other labs who wanted to look at the way we were working. They shared their experiences; we shared our experiences, and that was a big learning process for me. I worked with some other laboratories, giving them advice, and I also learned a lot myself from those consultations. Of course, digitization of the slides is just the first step in becoming completely digital because there is a second step of adding computational pathology or AI. You cannot just place a couple of scanners and then expect to get the full benefits of image analysis and AI. In order to get the full benefits, you really need to digitize all of your slides.

JT – So, when you say you were the first lab to become completely digital, you mean, the first to be scanning every single slide and to have a completely digital workflow?

AB – It is always difficult to say completely digital, of course, but what I mean by ‘being fully digital’ is that we scan everything. We have a digital workflow, and we provide the pathologists with digital files. We didn’t demand that the pathologists change over from glass to digital slides. We gave them the choice to do that with the understanding that it was possible, from this point forward, to do all their case load using digital files. After two years, we told the pathologists that we would discontinue work with glass slides. Even now, however, I have some colleagues that prefer to use glass slides for some cases; they use a combination of glass and digital. In my opinion we could completely stop using glass slides.

JT – You still keep the glass slides of course?

AB – Yes, due to regulation, we need to keep them for a time, and so we currently keep the digital images for only one year. In the Netherlands, we are changing the regulations, and it is likely that in a couple of years, we won’t need to keep the glass slides. Then the digital slide will become the permanent record, and that will mean increased costs for digital slide storage and maintenance.

JT – I guess that will be offset by the reduced costs of storing the glass slides.

AB – Yes definitely. We currently have a big storage room in our lab, and we rent another building for the storage of glass slides. Digital storage is also getting cheaper all the time.

JT – Are you part of a regional group of hospitals? Can you tell me something about the size and extent of your completely digitized workflow and the associated network?

AB – We work with different hospitals. Currently there are a few big hospitals in the network and some smaller ones, and we have about 60,000 histology cases per year. We use courier services to bring the material in from the various hospitals to our lab. We operate a very modern laboratory using five scanners to digitize the slides, and they are digitized even before the pathologist gets the glass slides. As I said earlier, we continue to use glass for some cases, and I think we need a little time to change that. There are still some challenges, and sometimes there can be issues with the focus or the quality of scanning. I expect that situation to improve in the future, as I have seen dramatic improvements in the scanner technology over the last five years. Technical problems should go away. We do still sometimes have problems with missing parts of the tissue on the scan due to autofocus errors. This situation has improved already, but some of my colleagues insist on having perfect digital slides and that is why we still use a few glass slides in the operation. Some colleagues have worked for 20-30 years with glass slides, and this new way of working is a difficult change for them. I understand that. This is why we need some time before everyone will agree to change to digital.

I think image analysis will play a really important role in the future, because AI will do a lot of the work for the pathologists. The time-consuming work and the tedious work can be done by AI, and if you don’t use the digital slides, you miss these out on these diagnostic tools. I think ultimately, AI will be the stimulus that will encourage colleagues to move to 100% digital pathology.

JT – When you receive your pathology samples, are they in the form of tissue or tissue blocks?

AB – We receive the tissues on formalin, and we do all the preparation, cutting, staining, mounting, etc.

JT – So that removes a lot of preparation inconsistency. It removes variation in the input to the scanning systems.

AB – With the latest AI software you can do a lot of things around quality checking and solving problems with missing parts of the tissue due to focus errors, because you can find the problem after scanning and rescan those specific slides. Then the pathologists only receive high quality slides. Also, you can adjust the images a little, you can change the color and improve the lighting and contrast etc. Of course, we don’t know if this type of image adjustment impacts diagnosis, and we need time to validate these processes.

JT – Which types of image analysis software do you use? Are you able to build your own algorithms or do you rely on the commercially available software packages?

AB – I’m currently writing a vision document and considering three possibilities. The first would be purchasing commercially available packages, Secondly, we could look at developing our own applications in conjunction with third party AI companies. The third option would be a combination of the previous two. I think the combination route will be the best for us, at least for our specific laboratory. We have a lot of experience with digitization, and we have experience in developing AI software. We also have a lot of ideas because of our experience with digital diagnosis. We know how to best use the AI to make our diagnoses more efficient and more objective.

It seems sensible therefore to build some of the software based on our own knowledge but also to introduce high quality software made by commercial companies. From my experience, if we buy software from different companies, and I do mean software with very high sensitivity and specificity, I think in our lab will find that the sensitivity and specificity will be a little bit lower than normal because the commercial AI programs have been trained on different slides. However, we will be able to train those algorithms on our own slides and validate the results, and this will be less work than developing our own software. We can easily do that, we are an innovation lab. Of course, we do recognize that we cannot make every algorithm that we might need, as there are too many possibilities. So a combination of these two approaches would seem to be the best way forward. In fact, we’ve already announced a partnership with Proscia to collaborate on the development of AI, and we can then deploy these solutions into our routine operations.

JT – Do you think image analysis and AI have a really big future in pathology now and that the algorithms will continue to improve in accuracy and become more and more useful.

AB – Yes, of course. We did some research, and we found that when you use AI, you can be much more efficient. You can save time to diagnosis and even improve diagnosis. You can certainly make a good business plan for the inclusion of AI in your operation. You can save money. It is about everything in combination, including efficiency and quality. When we, as pathologists, make a diagnosis, that is subjective. When I make a diagnosis, maybe I grade certain cases slightly higher or lower than another pathologist, and that is my subjective decision. With the application of AI, we can make the diagnosis objective. It is always graded the same way, and this can give a level of confidence to the clinicians.

Personally, I don’t like to search for a small metastatic region across 20 slides. That requires a lot of time and effort, but software can do that for you. I can then open the case, and the software can show me exactly where it found this region, I can take a look, say I agree, and I’ve completed the case really fast. In the case of prostate cancer, you can have 100 slides. It can take me more than an hour to handle the case because I need to look at all the slides. However, if I trust software to screen the slides for me, if it has a high specificity and sensitivity, and it tells me that from the 100 slides, it found five slides with tumor, I can skip the rest of the slides. That is a huge increase in efficiency, and yet we don’t miss the tumor. This type of software can be at least as good as, and maybe even better than, the pathologist at finding the tumor.

It is also particularly important to understand that we are not replacing the pathologist. The software will always be a tool that makes your work easier. It improves the quality of that work and makes you more efficient. Then, we pathologists can do other things with our time.

JT – Clearly, we need to have a tolerance of false negatives to use these types of screening algorithms, but we also need a tolerance of pathologist error. How should we balance these two approaches?

AB – There is some research on that, and there are publications that show there is really no significant difference between using only software or only a pathologist. Also, if you use a combination of the two approaches, then the quality of your work can be much higher. You will likely miss much less than when only using a pathologist.

JT – So what happens to laboratories that don’t digitize? Will there be two tracks in the future, digital labs and traditional labs still using microscopy?

AB – I think within ten years, we will have improved the algorithms further. You also have to consider that the grading of tumors is currently very subjective. If I grade a prostate tumor in the morning and again in the afternoon, I may give it a different score. There are known inter- and even intra-observer variabilities. There are a lot of publications about that. You can also improve the software through training on three different groups of images: patients with good, intermediate, and poor outcomes. The software will be able to work out the differences between these three groups by finding features that may not even be visible to the pathologist. So, I think in the future, there could be computer-based grading systems that will change our existing grading systems. If that happens, then you will no longer be able to do grading in the traditional way. I don’t expect this in five years, but I think within ten years it is very possible that we may be operating new computer-based grading systems, which will be accepted by the whole world.

JT – So, there could be laboratories in the future which cannot access this digital world and they will need to send slides to the larger labs. It may also become difficult to recruit younger pathologists into the non-digital microscope-based system.

AB – I think all the academic medical centers will become digital, and as you say, the traditional labs will not be able to find young pathologists. It will be much more difficult in the future to operate laboratories only with microscopes. Also, there are real financial benefits from becoming digital since you don’t need to ship slides around. We have a project now to expand our network and allow everyone in the network to make digital consultations. This also saves a lot of money, and these types of savings will make it difficult to make a financial case in the future for non-digital laboratories. I think everyone will go digital ultimately. Radiology is a good case in point. Nowadays you cannot find a radiology lab still using the old chemical systems. I believe pathology will follow the same process and will become entirely digital within a decade or so.

JT – Dr. Baidoshvili, Thank you for your time today.