Conversation with Dr. Giovanni Lujan, Director of Digital and Computational Pathology at Ohio State University Associate Professor, Gastrointestinal Pathology

Conversation with Dr. Giovanni Lujan

Director of Digital and Computational Pathology at Ohio State University

Associate Professor, Gastrointestinal Pathology.

“The future pathologist will analyse the reports generated by Artificial Intelligence from digital images; but that will not be the end of their participation in the diagnostic process, but the beginning, as AI gets more robust, it will also provide us with feeds from other Pathology subspecialties which are currently siloed, as well as from other diagnostic and clinical modalities. This all-inclusive data will enable us to be the ultimate diagnostician and reclaiming our position as the doctor’s doctor. In my opinion, our future is very bright!”

BIOSKETCH: Dr Lujan holds dual board certification in Anatomic and Clinical Pathology and is also certified by The American Board of Artificial Intelligence in Medicine. He has 20 years of experience as a specialist in Gastrointestinal Pathology. During his training, he completed a Surgical Pathology fellowship at The Johns Hopkins Medical Institutions after graduating from the Pathology residency program at The University of Texas Southwestern Medical Center. Dr Lujan has held previous academic appointments at The Johns Hopkins University School of Medicine and The University of Texas Southwestern Medical School. He was director of a GI fellowship program for several years. Since early 2020, Dr. Lujan has been Associate Professor of Pathology and Director of Digital and Computational Pathology at The Ohio State University Wexner Medical Center. He is passionate about the new era of Pathology, being digital and amenable to be analyzed by many computational tools, including AI.


Interview by Jonathon Tunstall – 25 Apr 2023

Published – 12 Feb 2024


JT – Today, my distinguished guest is Dr. Giovanni Lujan, the Director of Digital and Computational Pathology at The Ohio State University Wexner Medical Center, Department of Pathology, which can be considered one of the world’s pioneering departments in digitization, being one of the first labs in The United States to fully digitize its pathology workflow. Dr. Lujan, welcome to Pathology News.

GL – Thank you for giving me the opportunity to be here and to share my experience in the field.

JT – It’s a pleasure to speak with you. Perhaps I could start by asking you what led you towards the domain of pathology and specifically an interest in the digitization of pathology images?

I had been working in private practice for many years when I first heard about the concepts of digital pathology and computational pathology, at a CAP conference in 2017. The entire concept was fascinating to me, and in my excitement joined the different organizations that were studying the field, I educated myself with all the resources available, even motivating me to take the initiative to tour some pathology laboratories in Europe that were already fully digital back in 2018. That took me to LabPon in the Netherlands and to Granada, Spain where I saw with my own eyes, pathologists signing out digitally and I knew it was not the future but the present.

As I gained knowledge, the excitement kept growing, to the point that I decided to leave my private practice group and upon a nationwide search, I had the good fortune of landing at The Ohio State University, where I learned under Dr. Anil Parwani who is one of the pioneers in the field; under his mentorship, I later became Director of the newly created division of Digital and Computational Pathology.

Dr. Parwani was the first pathologist to sign out digitally in the United States, slowly some other pathologists at OSU started joining him, when I joined, after two years of having initiated the digitization process, there were only a handful of pathologists signing out digitally. My first role was as a cheerleader to try to encourage and motivate pathologists to make the transition from analog to digital.

As soon as I arrived, I started working along with our fantastic digital team in training all pathologists on the Image Management System, this training was mandatory even if they were not prepared to sign out digitally, we also validated them, so they were ready whenever they felt like it was their time. I also had talks with all the lagging pathologists to understand their reasons for not going digital where possible. The main ones were a lack of trust in the monitors and fear of not being able to be as efficient using digital technology as they were using the microscope. We let them all take their time and transition at their own pace.

In March 2020, the COVID pandemic started hitting hard, everything was shutting down, including healthcare institutions, pathology departments around the country were shutting down too, pathologists at academic centers had to halt their teaching, tumor boards were cancelled, and the bare minimum were kept to serve the population seeking urgent care.

At this point at OSU, as well as in most hospitals, a call from upper management was made to keep all personnel able to work from home to do so, which at the time included physicians including radiologists. In pathology, we were ready to be remote as well.

So, the university provided us with VPN equipped computers to take home, and that meant we could work from home with the infrastructure we had in place, which most of the pathologists had been trained on. So that was the big moment, what should I call it? Maybe ‘accelerant’ is the word I’m looking for to describe the adoption of digital pathology at our institution. That was a big moment of change because at that point all of our pathologists jumped totally on the bandwagon, and if they were not, they started using digital because we could do it from home. We were one of the few departments able to do that and we were the lucky ones. In other pathology departments, everybody still had to come in and do their work. They were exposing themselves, exposing their families to something that was still very unknown at that point. This was around March or April of 2020, so it was more the fear of the unknown at that point. We didn’t know what was going on necessarily. We just knew that people were getting sick, they were contagious, and they were dying.

Also, of course, pathology is not a very young specialty in terms of the people who practice it. We are a very old crowd in general. I think the average age for pathologists here in the United States is 55 or something along those lines. So, admitting that you are at that age where you need to take care of yourself and now having the possibility of signing out from home whilst still being available, that of course was something that nobody could say no to.

So that was my purpose in coming to Ohio State. I was the facilitator, perhaps I should call myself the ‘cheerleader’, for digitization. When the pandemic came along with all the bad things that it brought with it, that was one of the silver linings for us. We were finally able to deploy the technology and prove to the skeptics that it worked. Ever since we’ve been one hundred percent digitally deployed with all the slides being digitized ahead of time, except that is for a few use cases in some subspecialties where the technology was not ready to benefit them, examples were hematopathology, renal pathology, and cytopathology. They were not included in the initial deployment, but we are working on that now, there are tools available for them and we are working diligently to bring them up to speed.

As for me, I am now the Director of digital and computational pathology with a focus on clinical deployment. We are currently working with several AI companies and our goal is to deploy several algorithms for clinical use during 2024.

JT – Well let me ask another question here before we go onto the topic of AI. I’m interested to hear you say that you are one hundred percent digitized, does the microscope still have a role in the laboratory?

GL – Yes, I can definitely say that we are not ready to get rid of our microscopes one hundred percent, although we probably could. In my case, I don’t look at anything that isn’t digitally produced, and I have lost touch completely with the microscope. There are only very rare occasions that I still want to look at something under the microscope. For example, most recently I was using Congo Red. Congo Red is a stain that you use to identify amyloid, and it’s not only that you need the microscope for this stain, but you also need to insert a polarized lens to detect whether or not the material that you are looking at meet the characteristics you are looking for. Other occasion is when you are looking for small microorganisms, sometimes it is better to use the microscope.  All these shortcomings however are being addressed and now there is immunohistochemistry for amyloid, as well as AI and there is also an algorithm for almost any micoorganisms you may want to find. So, in my opinion is only a matter of time before the microscope will be only of historic value.

JT – Do you still find that there is some resistance, say perhaps from older pathologists to becoming completely digital?

GL – Initially there were some people who said that they were never going to do it, they were never going to try it, and that they were going to quit if they were forced to. Then the pandemic proved that none of that was true. In fact, some of the early opponents are now some of the strongest advocates. Nevertheless, some pathologists still require the glass slides on a daily basis, they just feel more comfortable that way, even though they are doing most of the job digitally. This dual delivery is costly, nevertheless, our leadership has decided to keep it that way.

JT – You briefly mentioned image analysis and AI earlier and of course that is very much a second-level technology, isn’t it? So, tell me something about that topic and how far you have gone with your AI projects.

GL – Well, we’ve been testing algorithms, individual algorithms for those with quite straightforward targets, such as counting mitosis, counting different types of cells like lymphocytes or eosinophils, and to some immunohistochemical markers like ER, PR and Ki-67. We have tested and validated most of them and we are now ready for clinical deployment after a final official validation. Our plan is to start deploying them in early to mid 2024.

When we talk about going digital, there are a lot of things that come into play beyond the digitization process itself. For example, we need to reconfigure systems that are not specifically designed to talk to each other. For example, the electronic medical records (EMR) of the hospital are integrated with the LIS of the pathology lab, the image management system, and the scanners themselves. Now we have AI, and that is another layer of complexity that also needs to be integrated into the other systems. The perfect setup where everything can be integrated into everything else does not exist right now, but we are trying to get all the pieces together so that we can work in the most harmonious way possible. This means that initially, we are not going to attempt to integrate these AI algorithms into our workflow, but instead, use them in parallel.

As we are using them in parallel, we don’t need a complete digital workflow to deploy these algorithms, we can get the glass slide, obtain the image, and then separately add the algorithm.

Our goal is to eventually have all this AI integrated into our platform and into our LIS where pathologists can just open the case and the AI is automatically running in the background.

JT – That’s interesting and of course, because you are talking about clinical deployment here, we can really see this as a synergy between algorithm and human. The algorithms will run, they will give their standard interpretation, but then the pathologist still needs to review those findings and make his or her own decision. The algorithms are acting in this case really as pre-screening tools.

GL – Correct, and that is the beauty of it, how machine and human prowess work together.

JT – Are you finding that the quality of algorithms is increasing because there are a lot of new ones coming on the market right now, at least for certain tissues? Prostate is a good example.

GL – Yeah. The first algorithms that were coming out were very simple and were designed for one particular function. For example, there were algorithms that were just able to find tumors or count mitoses, which are very simple tasks.

Then they started to become more complex. For example, we had a series of algorithms running in one mini program which acted together as one longer algorithm, if you will. They could now count Ki-67, for instance in neuroendocrine tumors, where the algorithm first identifies the tumor, then defines the cells that are positive, finds the hot spots, counts them, counts the positive cells, and finally gives you a percentage.

Each of these steps require different algorithms that are trained to different levels and then combined to create this kind of more complex algorithm. You mentioned prostate algorithms and certainly the newest prostate algorithms that are out there are very much diagnostic tools. They not only find the tumor but also grade it and then give you further information.

The latest algorithms will give you a lot of information. When you open the image, the AI is already working there in the background, and it will give you additional information about pretty much all data that you may or may not find in that biopsy. For example, when you examine a gastric biopsy you are looking for several things. Under the microscope, you are looking for Helicobacter pylori, then you’re looking for intestinal metaplasia for atrophy, and for inflammation and for malignancies of different kinds, epithelial, neuroendocrine, lymphoid etc. Now, an algorithm can test the biopsy for everything. It looks for H. pylori, atrophy, looks for inflammation, looks for malignancies of different types, it looks for dysplasia or premalignant conditions and creates a heat map for all possible findings. One of the scariest things for gastrointestinal pathologists is the fear of missing goblet cell carcinoma, which can be found in tiny non-discrete foci and it is easy to overlook. AI tools will make sure that all those non-discrete foci of cells will be looked at.

I’m talking of course about very complex algorithms that go beyond one single response or one single target and look at pretty much the entire tissue and the individual components. There is a new generation of AI algorithms that use unsupervised learning or computer vision so that we are not having to train the computers anymore. The computers are training themselves, correlating morphology with provided data, creating prediction models based on morphology. So, you just have to feed the computer information to correlate with the morphology and then the algorithm will find the correlation and predict the data subsequently from the morphologic features.

So, we are testing some of these algorithms right now. We are very excited about doing all these different types of testing and looking into validation, approval, and deployment. Here at OSU, we have the big advantage of having a digital workflow. If a developer or researcher wants to test their algorithm, we are one of the most advanced institutions that can do that.

JT – You know, I really concur with your thoughts here, for example, there are new algorithms coming out, which can operate just on H&E and of course, the whole world of unsupervised learning is very exciting. Things are moving very quickly, so what does the future look like? If we think 10 or 15 years ahead to a time when we may have a lot of fully digitised pathology labs using image analysis and AI and using that for clinical reporting. How does that change the pathology lab? Will young pathologists still be able to operate in a microscope driven environment, and what happens to labs that don’t digitize?

GL – I think the specialty is going to evolve, it’s going to move away from morphology only, because that’s what we do right now, morphological analysis. That’s not going to disappear however, it’s just going to get enhanced by AI and other computarized tools, and we will have to learn more and more about how AI operates and find the sweet spot of interaction, where machines do what they do best and human do what we do best, which is our abstract thinking and our ability to see through potential AI errors. We need to make judgements and to find out what works and what does not, but there always has to be a human at the end of the line to determine if the diagnosis is correct or not. If we think about supervised learning, we know what it is doing because we have set the ground truth. However, in the case of unsupervised learning, we’ll have to learn how to assess if the computer is telling us the truth. We’ll need to understand that clearly, before we can let it go on and we act upon any diagnosis.

It’s very interesting how computer intelligence, that is, artificial intelligence, and human intelligence differ. The computer can follow linear patterns and make calculations based on those linear calculations, but human intelligence goes beyond this and deploys abstract thought, which doesn’t exist in the computer world. A human can understand something that doesn’t follow an expected linear pattern can still be correct or could be correct, or vice versa. Computers are good at the mathematical thinking and that is what they do. All these AI algorithms are numerical and are making big mathematical calculations that we either cannot do or would take too long to do. The computer moves very fast and very quickly, but if you remove that linear trail, then the computer doesn’t understand what’s going on. It’s like, for instance, when you’re reading DNA or RNA, the computer calculates the sequence very quickly, but if there is suddenly something missing, then the computer will crash or stop. The human brain can understand what’s going on in those circumstances and can find a solution.

I’m using a very simple analogy here, but to go back to your question, I think that is going to be the human role. The future pathologist will supervise what the AI is doing to be able to come up with the ultimate diagnosis which encompasses all the different ‘omics’ or data originating from each diagnostic field.

I think pathology right now is like a big set of buckets with laboratory information coming from pathology, molecular, morphology, or surgical pathology, cytogenetics, immunology, etc. In the future, you will extract the information from all these different sources in one whole package. Then you will be able to analyse the report and provide a diagnosis. The function of the pathologist has become so subspecialized and divided now, that no one knows what we do exactly. What I envision for the future is that you will have a pathologist who would be able to solve complex problems using all the tools available in all the different diagnostic buckets and will use AI to link and analyse all those different buckets and will produce a ‘wholesome diagnosis” encompassing all pathology subspecialites coming out from different laboratories.

JT – A truly holistic approach

GL – Yes, everything together and I think that we will a kind of “super pathologist”, that is, the new pathologist of the next generation. I also think this will be very attractive to students, and we have seen it, we have an elective rotation on Digital and Computational Pathology here at OSU that keeps growing. Every time a student comes in, they say ‘Oh my gosh, I haven’t thought of pathology as a specialty, and now I think I’m going to consider it.’

JT – And so there is a great future for clinical pathology?

GL – I don’t see us disappearing. I don’t see us being swept under the rug. I see us reclaiming our position as the doctor’s doctor and the ultimate diagnostician as we will provide the ultimate all-encompassing diagnosis as well as tools to implement treatment and measure response and determine follow-up. I think our future is bright, very bright.

JT – So the pathologist can look forward to a bright and exciting future. That’s a nice conclusion and also probably a good note to end on.

Dr. Lujan, it’s been a pleasure speaking with you. Thank you for your time today.

Share This Post

Leave a Reply