Pathology Meets AI: Combining Veterinary and Medical Pathology for Digital Advancements w/ Richard Doughty, DVM, MD, MSc AIFORIA

In this episode, I spoke with Dr. Richard Doughty, Senior Medical Advisor at Aiforia Technologies, who brings a rare combination of expertise as both a veterinary and medical pathologist. His perspective highlights the unique intersections of these specialties and how they shape advancements in digital pathology.

We discussed:

• A Unique Career Path: How dual training in veterinary and medical pathology unlocks insights that benefit both fields.
• The Role of AI in Pathology: How tools like Aiforia enhance diagnostic workflows and enable more precise, efficient work.
• Pathologists as AI Innovators: Why pathologists are crucial to the development of practical, clinician-centric AI solutions.
• Educating the Next Generation: The importance of integrating AI into residency programs to prepare pathologists for the future.
• Practical Advice for AI Integration: Tips for selecting the right tools, managing implementation challenges, and achieving meaningful results.

This discussion offers actionable insights and practical takeaways for pathologists looking to embrace AI and advance their expertise. Whether you’re already using AI or considering its possibilities, this episode provides valuable guidance.

• Learn More About Aiforia
• Download Digital Pathology 101
• Contact Dr. Richard Doughty: richard.doughty@aiforia.com
• Watch MORE VIDEOS created together with Aiforia: 

Be Part of the Pathology Evolution: Stay informed on the latest in digital pathology innovations. Subscribe for more insights, become a member of the Digital Pathology Club, and get your complimentary copy of “Digital Pathology 101“. Embark on your path to discovery and progress in the fascinating world of pathology.

Richard: [00:00:00] So we have a pathologist shortage. We have delayed cancer times, errors. So like data’s not reproducible between pathologists and things like that. I would first start at a different point. Really. I would track back and say, pathologists are amazing at what they do and amazing how they do it and how quick they do it and how effectively they do it.

AI companies have a real. Sort of steep road to become better than a pathologist when you look at an effective pathologist work They make very little errors amazingly few errors for the amount of glass they push through a microscope Yeah, and the fast and their evaluations which sometimes, they come out semi constant But when they just look at it and they say yes, this is 50 percent for example Yeah, it’s amazing and all these decisions are having Quite considerable effects on the patient’s journey.

Intro: Learn about the newest digital pathology trends in science and industry. Meet the most interesting people in the niche and gain insights relevant to your own projects. [00:01:00] Here is where pathology meets computer science. You are listening to the Digital Pathology Podcast with your host, Dr. Aleksandra Zuraw.

Aleks: Welcome my digital pathology trailblazers. Today’s episode is brought to you by Aiforia.

Aleks: And my guest is Dr. Richard Doughty and Richard, you are a person with a very rare combination of specialties. Dr. Doughty is a veterinary pathologist and an MD pathologist. And he also supports Aiforia with their AI efforts in the domain of I don’t know, veterinary or MD or general pathology. Richard, welcome to the show.

Aleks: Tell the listeners about your background and how did you become an MD and veterinary pathologist? Because I just want to give a little bit of context. It’s basically, more or less, doing medical training twice with the residency in two different domains.

[00:02:00] And you also worked in both areas. So yeah, let’s talk about that first.

Richard: Great. Thanks for the introduction. Yeah.

Richard: I started off originally wanting to be a veterinarian, a clinical veterinarian, and I went to vet school. That was exposed to so many different species and and I began to be very fascinated by pathology and veterinary pathology is amazing.

You could be looking at a slide from an elephant and a slide from a a dog or a cat in the next time you look under the microscope. And so I just started to decide, okay, I want to be a veterinary pathologist. And then I, when I left vet school, I went to the Royal Vet College.

Aleks: So did you know already in vet school that you can become a pathologist as like a full time job? Because I didn’t know. I did my vet school in Poland and only when I went to Germany for a kind of externship, I learned that you actually can be a veterinary pathologist full time.

Richard: Yeah. Yeah, I guess I did. I was always looking [00:03:00] midway into the vet school, I started to think, what else can I do with my vet degree? The clinic is the obvious place, but what other opportunities? Because in reality, it’s one of the broadest biomedical degrees you can ever do. And so it offers so many opportunities.

So I was looking at things from. Environmental sciences one health and veterinary pathology can be a big deal.

Aleks: I was looking into one at some point, quite as well.

Richard: Yeah. And so what I did as I took a year out of vet school and study the masters of science in toxicology, and that made me aware of the role of pathologists in regulatory toxicology and specifically toxicology pathology.

So that’s sort… 

Aleks: Shout out to ToxPath. 

Richard: Absolutely. And I had some really good Tom Williams from GSK, Ian Pyre, I think he’s now maybe Amgen. They really wanted pathologists to join the field or wanted vets to join the field. So they were so helpful. And then I so that was really my plan after the sort of third year of vet school was to do pathology [00:04:00] and actually specifically really to do toxpath.

But as most careers go, you meander, you find opportunities, you hop to another thing. But when I left vet school, I went to the Royal Vet College and did a year there. But at that time, they were desperate for toxicopathologists, so I was recruited into toxicopathology after a year. And went to AstraZeneca where they had an incredible training system for me where I learned lots of different toxicological studies lots of different species.

And then after that, I jumped over to, again, to another company, General Electric Healthcare, which, who was developing contrast media for radiology. And that’s when I started to think. I, we, I’ve learned many species, but the one species, maybe I haven’t learned anything on is the target species for most of these things that I’m actually looking into toxicology for it almost felt like completing the circle by doing the, or [00:05:00] completing the jigsaw puzzle by studying Human medicine.

Aleks: They should have some kind of a combined degree, at least for pathologists, but you’re just a pathologist and you can do anything like animals and people, because in, I don’t think they do it in, in veterinary pathology, but in in veterinary medicine, but in MD, in human medicine, you can, for example, enter residencies from different backgrounds. You’re still an MD, but like you can do dermatopathology as a dermatologist and as a pathologist. And there are a couple of these combinations I don’t really see in veterinary medicine. I think they should do something like that for pathology. You’re a pathologist, you’re a pathologist. 

Richard: Yeah, exactly. And I think the most important criteria is your interest in your willingness to learn. . And if you have both of those, then you will then you’ll do it well in any subject. 

Richard: Yeah, so [00:06:00]  in the end, I coming back to the story, I went to human medical school, worked as a clinical vet a lot of the time while doing medicine.

And it was quite strange. I’d be doing a cesarean on a dog in the evening and then going to the human hospital and seeing exactly the same operation. So it was quite amusing. When I was finished, I started to think, okay, what am I gonna do? And then I decided eventually to. specialize in, in a human pathology or anatomic pathology as you stay in the States.

And…

Aleks:  And then you’re sub specialized as well, right?

Richard: Yeah, exactly. So my, initially. 

Aleks: Cause it was so boring. You’re like, why would you just be a pathologist? You can’t even sub specialize.

Richard: Yeah. So I eventually did mainly for a while scientific pathology, but now I’m mainly focused on neuropathology at the moment.

So that’s occupying my mind. Plus, of course, the work I do with Aiforia, which is something a little different and gives me a bit of variety [00:07:00]  in my daily practice. 

Aleks: Your life is not interesting.

Richard: We have to find something else to do. I’m not sure what it’s going to be, but

Aleks: I think AI is a super cool vehicle to just contributing to different areas of interest, different areas of biomedical sciences.

Aleks: So you being an MD and veterinary pathologist and a DVM pathologist, what unique perspective does this bring to your role at Aiforia, to the AI work that you’re doing?

Richard: Yeah of course, as pathology is pathology. It doesn’t matter whether it’s a fungal disease in a mole or whether it’s abnormal carcinoma in a lung of a human.

The basic morphological principles of the same. I think so. So in many ways, Aiforia is very unique and this is what I was attracted to, it has probably one of the broadest portfolios in the AI pathology [00:08:00]  industry and that it covers preclinical, it covers, it’s also got an interest in veterinary pathology, but it also has the human pathology side, the clinical side, and I think that’s what really attracted me to, for example, Aiforia was that I liked that idea that I would be working with maybe some clinical projects, veterinary projects and, and human projects and also interacting with both medical and veterinary pathologists. So it was that sort of that really attracted what I can, what I think you bring as a veterinary and human pathologist is that you bring that variety. You can probably laterally think you’ve seen a lot of things.

You may think a little bit out of the box. And that can be often an advantage when you’re trying to think up strategies for setting up layers and things like that. So I think it’s that breath of knowledge that is really important. Also it’s that breath of knowledge of workflows because veterinary pathologists and hemo pathologists do have subtly different workflows. [00:09:00]

And so I think that can often be utilized in developing clinician centric AI, what’s going to augment us. And that’s what all pathologists, and I think that’s why it’s an advantage to have that background in both veterinary and human. Yeah, this isn’t just for medics.

It’s for the whole breath of pathology.

Aleks: I want to highlight two things that you said. So one is that the morphological principles of tissue are the same, especially in mammals it’s basically the same. And I think people are not aware of that because in the pharmaceutical industry, most of the pathologists are veterinary pathologists, and they often work on human tissues, like all the discovery things, biomarker, quantification, a lot of things happening on human tissues are being handled by veterinary pathologists. And I think like the general public doesn’t know that, [00:10:00] that you, even though as a veterinary pathologist, you don’t have the the authority to diagnose, you still can recognize and use all the knowledge that you learned in veterinary pathology residency.

And the second thing is workflow, not only veterinary versus MD, but very much diagnostic versus tox path work. I think because ToxPath is so niche, few people and few companies outside of the space are aware how this workflow affects how you then later interact with your slides. Definitely a super cool combination.

Aleks: So having pathologists in AI, involved in AI development, contributing to creating effective and trustworthy solutions. Like, why is this important?

Richard: Yeah, I think I can take this in two ways or two levels. Firstly, why should there be.  [00:11:00] Physicians or veterinarians involved in med tech in general.

And then also more specifically, why should pathologists be involved in AI pathology? I think that bigger picture, I think, instead of having vets or physician pathologists in, in, in med tech, it’s really important because they bring. Clinical expertise and they bring credibility to that company.

Vets like to talk to vets, doctors like to talk to doctors. They’re familiar with the training, they’re familiar with their terminology, they’re familiar with the culture of medicine. So they like that sort of connection. So that’s a very good way Of how a physician or a veterinary can get leverage in a company.

Also, it’s a bridge between the technology and the clinical practice. It’s yeah, the clinician can essentially act as crucial bridge between tech developers and the clinical community, and they can also translate that complex medical requirements into what essentially become technical [00:12:00] specifications for developing the model.

So I think that’s where they can assist very much. The other things, I think, like regulating compliance guidance, that’s a very central role, for example, for for a physician or a veterinarian, also developing user centric design in the in the interface that the pathologist is going to have with the system.

Other things are very much broader. And if the clinician has listed, developed that role is so market insights and strategic directions for the company, where are they going to go? What are the problems? You’re living it daily when you’re a pathologist, the irritation of having to the irritation, but the challenge of having to go through 15 meningioma slides to find those hotspots, there’s mitosis, it takes a long time, and it is intellectual, but it is also volume, so if we can get things to help us, and you will only know that if you’re a pathologist, screening through these things, knowing where these sort of [00:13:00] pressure points for us are.

The other things are, I think, where clinicians can add value to a, to an AI company is through education advocacy to the profession in general, teaching them what this technology, you’re really central in understanding how this technology works, how it’s been intended to work and what maybe the pathology profession needs to equip itself within the future to be able to really utilize this to its maximum.

So I think that’s one level as being a clinician, I would say specifically as a pathologist, I would say defining relevant use cases. So which models shall we focus on? What do I see most of each day? Because it’s a market we’re trying to sell. So we need, we can’t, for example, pick a very rare tumor or something that, that nobody would want.

So that’s picking the models and finding where is the volume. I would say also, of course, Data annotation and model training. That’s really important. If it’s not going to be exclusively [00:14:00] pathologists doing that, at least the pathologist needs to have some control over the ground truth and quality controlling what is being done.

I would say validation and testing. That’s also really important for the pathologists. That really gives them an insight. I’ve felt it. I pour it being involved in the validation and testing gives me an insight. How is this model working? Almost? How is it thinking? To try and understand it and also to try and understand how to make it better.

I think also feedback on workflow integration, things like that. Pathologists like to communicate with pathologists. And I think as I’ve said before and I think that flow of understanding of how we do it, how we do it. And how we might do it in the future is really important. I think also developing trust because not everybody is bought into the idea of AI coming in, or they maybe don’t understand what AI is going to do for them.

So I think we also have a role.

Aleks: Yeah, I think many people don’t know [00:15:00] how those tools are used and extrapolate things from, I called the consumer AI tool. To, to what would happen in the medical domain. And are often afraid, but they are, there are also two ends of two sides of that coin, either they’re super afraid and say, it’s not going to fly in the medical space at all, or they’re overly enthusiastic saying like, how I have it in my phone, how come I cannot have it while accessing electronic health electronic health records or like, because.

Yeah it’s either fear or like disappointment that it’s not there yet and when it’s going to happen.

Richard: Yeah. And I think it’s, I think this is one of the roles a pathologist can have in the company is that if you think about technology companies in general, they’re well known for hype. [00:16:00]

And that’s the way the industry is built. It’s built to hide something up. It relies on that sort of investment to keep going. But as a physician, as a veterinarian, or a pathologist, we know we, we, the hype can be dangerous for our patients. In some respects, so we can also ground it a little bit, make pathologists understand what is this technology, how do we expect it to work, or how are we hoping it’s going to work and guide them through it.

So I think it really is a partnership between the AI companies and the profession in general. And it’s not for one, it’s not for the companies to dictate how this is going to happen. It’s almost holding hands together and leading our way through it. A lot of this is new for both the companies and the profession.

So we really need to find a way through the see in the dark. And that’s, I think that’s best done with us together. Yeah, absolutely. 

Aleks: I’m gonna tell you a sad thing. And most of the AI companies, they do not have pathologists on staff. [00:17:00] And I kind a lived through the difference.

Maybe they might, they have somebody like in a more managerial position, but not really hands on working with the teams. And they rely on collaboration with customers or collaboration with some networks. It is not the same. And I lived through this, like you don’t have so much skill in the game.

The moment you don’t have good communications, you’re like, okay, I don’t care anymore. I’m going to do my job. That’s like within the scope of my responsibilities, I don’t have to keep communicating with this partner or that partner if they don’t get me or don’t want to understand, don’t want to listen to me.

And on the other hand, that the customers are never part of your organization. It’s somebody you want to deliver a product for and sell something to. And if you don’t have a representative of the [00:18:00] customer within your organization from the, from the partnerships that I have observed, read about and lived through, if you don’t have this kind of representation, in this case, in form of a pathologist who knows what pathologists do.

You, the collaboration suffers. So…

Richard: Yeah, I couldn’t agree with you more. I think, and I think this has been a maybe a suffering of technology companies that they haven’t maybe acknowledged that. It’s really important to have not just external consultants, but also physicians, veterinarians within the company integrated at every level, from deciding maybe how the model is going to be initiated, how it’s developed, not least ground roots, for example and how and up to the levels of sales and marketing, really, I think I think that there should be [00:19:00] an element of pathologist input in all of those.

It’s a business. So of course it has another set of elements to it that are independent of the medicine. But medicine, you can’t be a company claiming you’re making a product that will augment somebody that you don’t even have in the company. That’s not good practice in general. It’s a little bit like.

Maybe being a, selling cards, but you’ve never driven a car, how can you truly sell it? It’s, it doesn’t make any sense at all. So very much in some respects, the person selling the car doesn’t need to know the technical components of the coding, the engine in the car, but they need to know how it runs, how it functions when they’re, how it feels to drive it.

And I, that’s the same with AI, I think. And I think… 

Aleks:Yeah, what is it good for? What terrain is it good for? Like…

Richard: Exactly…

Aleks: Like all the use cases and, everybody has a car. So that’s [00:20:00] nobody questions that anymore, but it’s that, it’s very much the same principle. And one thing I want to add is it has to be somebody who’s going to get upset on the inside of the organization if things are not done properly and will invest energy to change it because if you’re working with external consultants, it’s not going to happen. The consultant, and every now and then I consult, I’m like, I give you the information you’re asking for, you can do whatever you want with this information. Most of the time, like you don’t use them the way they should be used, but that’s okay, that’s the role of a consultant, whereas somebody who’s in the organization has skin in the game once they accompany to succeed and as a pathologist has a totally different influence. I would say on the whole industry. 

Richard: Yeah. I would agree.  [00:21:00]

Aleks: So that was my experience. 

Richard: Yeah. I think you’re right. And I think it’s very different being within the company and the, as you say, with an external consultant, the balance of power maybe there is a little bit different.

You don’t want to necessarily highlight disadvantages and things like that. Whereas when you’re… 

Aleks: Even if you do, it’s like you’re not gonna highlight it five times and go to the right people and say, Hey, and figure out like a different way to talk to them, figure out a way to get along. No, it’s like information here.

You have it. Do whatever you want with it. You hired me to get you this information, please. The next step is on to you. But obviously. This is not without challenges. So what challenges have you encountered in bridging this gap between technical and medical aspects of AI development?

Richard: [00:22:00] I think. On one level, it’s interesting because I think some respects that many companies, of course, when you see the typical advertising from any companies is that we have a pathologist shortage.

We have delayed cancer times, errors, like data is not reproducible between pathologists and things like that. And I would first start at a different point, really, I would track back and say, Hey, Pathologists are amazing at what they do and amazing how they do it and how quick they do it and how effectively they do it.

AI companies have a real, sort of a steep road to become better than a pathologist. When you look at an effective pathologist work, it is, they make very little errors, amazingly few errors for the amount of glass they push through a microscope. 

Aleks: And fast. 

Richard: Yeah. And the fast and their evaluations, which sometimes, they come out semi quantitative, but when they just look at it and they can say, yes, this is 50%, for example.

Yeah. [00:23:00] It’s amazing. And all these decisions they’re having. quite considerable effects on the patient’s journey, if it’s behind the sample. So I think I would first say, don’t approach it that pathologists aren’t. Are these people constantly making this face

Aleks: Good enough, if you leave something, because that’s what you have, right?

That’s what the whole medicine relies on pathological diagnosis. Most the most public facing aspect is going to be tumor diagnostics, cancer diagnostic oncology, but there is, that’s like a fraction of pathology. And they are doing a great job being those who like determine what’s going to happen next because this is the step that determines everything.

Richard: Yeah, and I think when I think about what my role has been in Aiforia, it’s maybe given that sort of, reality of what the daily life of being a pathologist is. We have incredibly competent [00:24:00] software engineers and biomedical scientists, but they haven’t sat down and picked up a glass and put it down and know you have to write a diagnosis.

It’s going to have a quite considerable effect on the patient. So I think giving them how we do it, why we do it the way we do it, because it’s not really obvious sometimes why we do the things, if you’re from the outside looking in, it can look bizarre. So it’s for example, with prostate…

Aleks: And you know what, I realized recently, not only is it not obvious between pathologists and computer scientists or like pathologists and non pathology and non medical people. It’s also not that obvious between toxpath workflow and diagnostic people and between specialties like Radiology is a pretty closely related specialty. It’s image based. When I talk to radiologists on my podcast, neither I am like super familiar with their [00:25:00] workflow, nor do they know what’s happening in pathology.

Like they don’t know and neither they could, they can, they, you, there is a lot of things in parallel, but you can feel the discrepancy. So that kind of highlights, okay, how much. explanation and like showing should go into training somebody who’s going to build tools for pathologists.

Richard: Yeah, exactly. Like you say and some of these, for example, if you think about radiology and the example you used in pathology, they’re both image based specialties, but they’re very, when you they’re not the same, they do very different things and they the way they, their workflow is different.

And those nuances are really important. So certainly I think a lot of. We have thought or there’s been a lot of discussion in the literature about, what can we learn from radiology? And we can learn a lot. But also we have to find our own path in how to do this because it we’re not radiology [00:26:00] and so our, needs are unique to our speciality and that always has to be remembered.

I think I do believe now

Aleks: So when you just started what was the biggest surprise to you? I’m going to give you my like surprise. I shouldn’t have been surprised, but basically when I joined and was working with image analysis scientists. My first like realization was, Hey, they don’t really understand the tissue at sometimes at the level necessary to conduct the project.

So my first thing was like, okay, let me let me prepare materials to teach people how to recognize tissue. That was like my, not really a surprise, but first like a action item when I joined an image [00:27:00] analysis company. Did you have something like that? That you said, okay, that would be the first thing I should working on.

Richard: Yeah, I think as pathologists, we work with tissues every day we histology, histopathology is a central thing, a central skill set. And I think often we don’t, it’s very difficult for us to understand…

Aleks:  How it is taken for granted

Richard: …little, for example, most biomedical science courses will have little to no histology.

If you think about a vet coming out or a medic coming out of vet or medical school, they in fact have very little histology. Many of them maybe wouldn’t even be able to, they’d know it’s a tissue but they might, they might wonder where it’s from. So I think I agree with you that sort of stepping back and saying, okay, this means that this is this almost giving a crash 101 on, on, on histology, I think is important because, I remember I was talking to one project owner [00:28:00] about, for example, desmoplasia -desmoplastic response to tumors. And I was just talking about it and I didn’t really realize, but they were wondering, is that a subtype of tumor? And of course it’s a response and it, to me, it just seems so obvious, I was talking about it and then, okay.

Then put that on a wall. Is it a subtype, a morphological subtype? And then you realize you have to step back and talk about, make, educate them I guess, same as they’re educating us about technology aspects, which I don’t understand as well as them. 

Aleks: Exactly. And I think this is the bridge part that the pathologists is the bridge between the technical and the users.

And it comes with realizing, okay, what’s everybody’s else’s expertise? How can we convey my expertise? Because it’s not pathology specific that you being highly trained, like straight of training or like doing something [00:29:00] every day, you assume that other people understand you. They don’t, neither do we understand a lot of things that we would like to have explained to us.

Richard: Yeah, and I think it’s interesting because I always thought to probably be in an AI company, you should be a pathologist or a clinician with a lot of high tech knowledge. But in fact, now I backtrack and think probably the most useful for a company is a clinician with no technology knowledge, because if they can teach that person or that person’s coming with a perspective that they don’t know anything, and there will be unfortunately some pathologists out there who are very, should we say, technology averse, and we need to win those over as well if we’re getting to that.

So we need to understand what a non tech savvy pathologist would how they learn technology in some respects.

Aleks: So how much, like technology, I think I will disagree here because I think [00:30:00] if you don’t and you can learn it on the job if you are open minded and, agile and see what’s needed.

Because to be able to convey the pathology information needed for other people, you need to speak their language. So if you’re talking to developers, if you’re talking to image analysis scientists, you need to at least have a basic understanding of those concepts. Because then what I have seen is you can translate their concepts into what is pathologically relevant.

So I have this, this term of computer vision to pathology, vision translation, where I explain, okay, if you just need to detect cells, that’s going to be object detection. If you need to, for example distinguished epithelial tumor region from the stromal region, that’s going to be semantic segmentation.

And if you want to measure everything that’s going to be a [00:31:00] segmentation, like I have this glossary that, you use a technical term and then you show on the slide, what does this mean and when should it be used? So in that regard, I think to support the company you should have that understanding.

And, but to use a product, you don’t need any of this. Maybe you do. Because they use determining the use case and because at some point it’s going to be prevalent enough that people are going to be shopping for different things, they need to know what they want from this algorithm. They would ask for support from the company, but basically you know what you’re looking for.

So you, you need to understand the limitations of the method that you’re going to be using. 

Richard: Of course. And I think what I more meant was when you come into the company, it should be. The technology side, much like we are teaching [00:32:00] them the pathology, they should be able to teach a pathologist. What these technical aspects are and so they can become and they should basically be able to teach you in a way We can teach them pathology so they can understand it they can use it and they know the limitations of or the complexity of the things they’re dealing with so I would agree with you and I but I would say that I meant more 

Aleks: You can also disagree, it’s fine to disagree with me. 

Richard: I’m not disagreeing, but I’m just nuancing it. 

Aleks: That’s okay. So do you guys have a good process at Aiforia, like for both sides? Do you guys have, how do you deal with that when you get new pathologists, when you get new computer scientists that should work together? Do you have some solution for that? 

Richard: Yeah, with Aiforia, …

Aleks: How to speed fast?

Richard: Yeah. I mean, as with any sort of young company that’s scaling up and there’s [00:33:00] always changes in the way things are organized and the way things are done. But at the moment we have project owners and they own that particular project and then a pathologist or pathologists will come in and assist them.

And that will be things like setting up ground truths, teaching the project specialists how to annotate, maybe quality controlling those annotations. But also, for example, with our plus date project, it would be the project donor would gather the three or four pathologists once a week. And we go through problem annotations, how best maybe to annotate something to get the best out of it.

So we’re all fairly consistent. And also we can discuss problems or, or discuss, for example, borderline cases, how are we going to deal with these, how will we deal with something that. We’re trying to essentially annotate certainty when two of the pathologists think it’s this and the other two think it’s that.

So things like that. So I think we have a very, I would say a [00:34:00] prostate project was incredibly successful with that and that weekly process, I think, got us to know the project owner, project specialists better, and they probably also got to know how as pathologists think as well. So I think that’s a very good idea.

One thing we’ve always discussed or have discussed since I’ve been there is this concept of project pathologists who they own the pathology side of it and they run hand in hand with the project owner and it’s not necessarily they will do all the work but they will maybe have a little bit better understanding of the model of how it works and what its weaknesses, what its strengths are.

So I think we have a very good process. We also have quite regular meetings where the whole science team will meet up and say, “How can we do things better? What are the challenges we’ve had and things like that?” So I think I think you have to actively work, not to develop a technology biology divide.

I think you really have to work at it. It’s [00:35:00] so important because they are two different types of people sometimes and they can very easily deviate from each other. So that’s always a challenge, I think for management to hold those two together. But I think at Aiforia we have a good process for that. Absolutely… 

Aleks: I love this concept of project pathologists and I like very much like the regular meetings and reviewing what’s going on because it’s so easy to just like. Get so absorbed with a project and then just keep doing your thing without checking back with with your teammates of different expertise and in this space, I don’t think we can afford it at the moment, at this stage, at the maturity stage of that technology, it has to be constant feedback.

And having I’m, of course, I’m biased because I’m a pathologist and I think this is how it should be [00:36]:0] but I’m very happy that this is how it is, at least for Aiforia.

Richard: Yeah.

Richard: And I think it’s and it’s, what’s been very, really interesting is for example, I’m based in Oslo and some of the pathologists are based in Sweden and Helsinki, that you can interestingly develop a diagnostic culture in each place.

Where maybe my grade three is their grade four, for example, it’s very interesting how you can develop these sort of localized and I think one of the advantages, for example, having these group pathologists is that you can, to a degree, try and even those out a little bit and get a model that’s you know, what we’re not trying to do is develop a, a sort of Helsinki model that goes around the world.

We want to model the works around the world. And that we are very conscious of. So I think having these meetings, understanding why somebody has come to that conclusion, pathology is a discussion, a speciality. You should always be willing to defend what you see and explain why you see it and explain [00:37:00] why you think you’ve come to that conclusion and and those, that’s what those meetings serve to do.

And an enjoyable process, pathologist like this kind of thing.

Aleks: Definitely.

Aleks: So for new people coming, like for pathology residents and practitioners, and how do you think they should be prepared to take advantage of AI? Do you think there is a specific, do they need a specific curriculum or ou think they’re like already tech savvy because they use this AI probably since it came out and their smartphones, like what’s your.

What are your thoughts on that?

Richard: Yeah, I think there does need to be a certain amount of integration gradually into in, into residency education. I think that’s going to has [00:38:00] to come because this is going to be the pathology residents today are going to be the people who are leading the profession in the future.

And so they have to have that input, much the same as they need to have training in molecular pathology, which, maybe 30, 40 years ago, it wasn’t even part of it. It was pure morphology. With elements of immune histochemistry. So I think we need to, and I think at the level you were discussing earlier, about what is segmentation, how do we, how are models developed?

Maybe you don’t need to understand the coding, but just understand where has the model come from, and how was it developed and how do you use it? Model? To understand what was, how to understand what was the model intended to do? Because there’s an incredible diversity amongst first of all, what models are intended to do, but also what pathologists thinks, think models should do. For example, some think I think it should almost serve them the diagnosis and they just look and say, yes, Sign into now other think, I think it should be picking out the rare cases [00:39:00] that we never see, so we can highlight those and others think that it should be more of an augmentation that we should be interacting with it, it should be showing us thing, highlighting things and we excluding them, including things and a more interactive.

So I think we need to educate residents in what this process could be and what the diagnosis, how the diagnosis process will change, because what I will say is. AI will change that diagnostic process. It is different from looking under looking under a a microscope and thinking, okay, what are my criteria for a grade three Gleason score?

For example ding. Okay. That’s what it is. This is going to be served to you in a way, and then you have the cognitive process is a little bit different and that they have to get used to, I think, and they have to understand that as well and also how to get the most out of it. Because We need to get them to sort of know how to get AI on its own [00:40:00] is okay.

Odds on our own are very good, but I think together we can be much better. And I think. But we have to learn how to be much better. It doesn’t come intuitively. I don’t think.

Aleks: What I have experienced recently in the last two years, a lot of discussions from different organizations, different societies, both MD and veterinary that, oh, it should be included, but it’s not yet included in the curriculum.

So recently I’ve came across a friend shared the Digital Pathology Academy or it by the European Society of Digital and Integrative Pathology. They have a course. So that’s something official. It’s called the E-S-D-I-P Academy. I didn’t go through it yet. There was a course by national histology.

Richard: Yeah. National Society for Histotechnology [00:41:00] has a digital… 

Aleks: Exactly. National Society for Histo technology. They had a course. I don’t know if they updated it. I have to check it because that was like already also like in 2019 and there was something. So there’s like bursts of activity. I have a YouTube channel where I tried to provide this information and I have a few courses.

So I think there is there is information and quality information out there, but it’s not like you have to actively put effort into acquiring this knowledge. It’s not that, oh I’m doing a pathology residency and it’s going to be included and I’m good to go after finishing the residency. It’s more like okay, I need to actively search for this information and get a little bit more involved.

Which is fine, but I don’t think everybody is aware of it.

Richard: No. [00:42:00] And I, I think to a degree, I think if you look at the society, radio society of North America, they have a very nice courses on AI at different levels that people can take. But like you say, you have to be actively interested to look for it and to do it in some respect.

Aleks: Or you need to be a member of a society, when your early in your…

Richard: Yeah. So you, I always think that there should be almost a sort of every resident, for example we could take an example, in Denmark, in Norway, in, in Finland should have to have at least a modular course or online course in it.

Just so they know it not ideally, with things like AI Aiforia, with the creators, why can’t they have a go at making a little model during the residency, just understand the process and things like that. I think you can also come much earlier as well in medical school.

So I think one thing that’s a little bit interesting about Aiforia is they also… [00:43:00]

Aleks: How cool would it be to teach histology like that?

Richard: Exactly. And so that’s what that’s what they have a capability with Aiforia to use it as a teaching tool. And then you get used to using that interface. We’re taking the slide, zooming in, having a look at something.

So it’s almost getting that culture, that sort of muscle memory and brain memory of using this. This is what we do a bit like as when you’re in the clinic, you have your stethoscope around your neck. And then the first thing you do is put it in and have a listen to the chest. So it’s, it should be a bit like that in pathology.

And I think, again, I think Aiforia has been a little bit unique and it has had that educational elements to it. So there is a cohort of medics. Maybe not a pathologist, but someone will be, that are used to using the interface with it. But I think all residency should have, there should be almost a standardized European or USA, such as AI driver’s license sort of thing for residents.

I think [00:44:00] what’s interesting about pathology contra radiology is digitalization and AI are all coming at once. Whereas radiology has had 20, 30 years of digitalization or 20 years of digitalization and now they got very comfortable with that and now they’re thinking about AI, whereas with us, it’s all coming at once.

And that’s quite a big task in itself, but represents challenges, but also represents opportunities for better take up of using AI, AI, I think, because it will be still integrated in the digitalization.

Aleks: Which organization I’m super proud of. The JPC, Joint Pathology Center, and their Wednesday slide conference rounds.

This is something for veterinary pathologists during their res, their residency. They have these slide conferences every week on Wednesday, and they are being scanned and they have been scanned since [00:45:00] I did my residency in 2012. In the digital pathology years, it’s a lot of years. So I was actually.

Studying for my boards with those slides, I was like traveling, my microscope broke and I was like but, and the exam was on a microscope. I took my boards in 2017 and it was on a microscope and there was always this like, “Oh I want to have my own microscope for the exam because I’m so comfortable with it.”

And at some point I like my microscope was so bad and I was traveling that I started hating looking at the microscope and I’m like, and that was my second attempt and I was like, okay, I need to do something that will prevent me from failing again. I have to start looking at slides. And we started doing virtual conferences with a study buddy of mine.

And I basically said, it’s about the content, not the content. about the form or [00:46:00] the way you’re looking at those images. If you understand what’s, let me first learn everything what’s in the image and then worry about the way I’m actually evaluating those images later,

Richard: Yes.

Aleks: I passed the boards. I was learning exclusively my second attempt exclusively on digital slides and sharing them through Skype.

And in terms of like high resolution, high quality, everything…

Richard: Impressive. 

Aleks:…like people would think maybe not the best way to do it, but it was enough for me. To learn the entities that I had to know to pass the board.

Richard: Exactly. And look at you now, you are a board certified veterinary pathologist

Aleks: Yeah, I am a board certified pathologist. Learned on digital slides from JPC.

So shout out to JPC for making those slides available digitally. And another story I wanted to tell you, because when I said, Oh, how cool would it be to learn histology [00:47:00] like that? Because when I was learning histology in vet school, we had this book. Where we had one it had a circle. It was like an exercise book, like a coloring book for students and half of the circle had the correct structure of some, and that was for pathology as well, for histology, we used to draw what we saw under the microscope and for pathology, we have this exercise book.

I have to get this exercise book and basically show people, but basically we were doing what we’re doing now for annotations. We were like delineating everything, but on paper, so it’s basically like the same thing. But digital, and I think that would be super cool, annotating, doing models and having this be part of the pathology course.

Richard: Yeah, exactly.

Aleks: We need to do it. If you guys are not doing it, I’m volunteering to help you.

Richard: It has to happen. We have to be ready for this and we have to make, [00:48:00] we’re doing our residents a disservice if we’re not making them comfortable for what they’re going to be getting in the future. And I think that’s a role for us in industry and also a role for the academic institutes to really address this and think how are they going to be dealing with this?

Exciting. 

Aleks: Yes. So is there anything else that I should have asked but I didn’t? Anything?

Richard: I don’t know. I think that was it. I was going to cover a little bit about the cognitive things, but I don’t, I think we’ve covered, had a good good good element to it.

Aleks: Cognitive, like in which way? Cognitive things. Let’s talk about it.

Richard: A little bit like human factors. So a little bit like, automation bias…

Aleks:  Oh yes.  Let’s talk about it. And let me tell you it’s a perfect perfect spot to talk about it because I recently interviewed Dr. Famke Esner and she wrote the paper the gold standard paradox.

Where whenever I like hear about, oh, pathologist cannot agree, the annotations are not consistent. [00:49:00] I think about her paper where she like mentions a lot of those biases. So yeah, let’s talk about it. What are your, what is your take on that?

Richard: I think, of course. With AI pathology, it essentially what we’re developing is at the moment, at least is a augmenting system.

So it’s a human AI interaction. That’s what we want. We’re not trying to just deliver the diagnosis and pathology, except if we want them to look at it and maybe say, okay, I want to remove this bit because I don’t agree or something. And then, or this bit isn’t how I like it. And then they’re getting the answers.

So they’re interacting with that. It’s highlighting things. And then, Presenting things in a way that enables a human to cognitively think that the highest they can. So I think in some respects, the human factor is the sort of, there’s an X-factor that, we can develop models, but how do we interact with those models is the interesting thing.

And I think probably the most, obvious one that has been spoken about is this automation bias. If you become [00:50:00] so complacent or the model, let’s say works very well, you start not really critically assessing it anymore. You just start thinking yes.

The, so these diagnosis, so you almost cognitive offload yourself. It’s interesting because really, research from radiology actually shows quite varied results. It’s really interesting. There was a work done. I think it was with a group in Harvard who looked at how does AI make radiologists better?

And the answer is it makes some better and it makes some worse. So if you are somebody who doesn’t critically assess a model and you just agree with it all the time, they found that those sorts of people, they actually do worse. They become worse radiologists and they would have done without the AI.

If you’re a radiologist on the top of your game, then you’re probably always going to be on the top of your game. It doesn’t really matter so much.

Aleks: Regardless of the tool you’re…

Richard: Yeah. But if you’re that sort of cadre of excellent radiologists, [00:51:00] which should I say, average radiologists, I wouldn’t like to offend them, but you know, it’s a, they can really benefit from it because they might be the group that might be able to put a step up and, they are critical, they are looking at things but they have a busy day and they have, 30 things on the list. So this may really augment them to be much better radiologists.

So I think it’s not going to be one thing to everybody because humans are individuals, pathologists, they’re all different. And so the way. The way we interact with it will be different as well. And I think that’s a really interesting thing. For example, I think one thing I’ve noticed, especially amongst maybe the less, and we’ve touched on a little bit, the less technology savvy is this algorithm aversion.

And that’s if they find one critical thing, but it doesn’t do as they think it should have. For example, it’s graded again, I’ll take these and it’s great. Or maybe breast. Grading, it’s graded something grade two when they thought it was a grade one, then that’s it for them. They don’t want to[00:52:00] touch that that algorithm again, and the whole of AI is destroyed for them.

We did, so we, again, it’s for the company and good people like yourself, who are really educated in the community to make them understand what is the model meant to be. It might not be right all the time. But that’s not the point. You’re meant to interact with it. So if you see it is doing something you’re not sure about, then you need to interact with it, take it out, remove it, see what the result comes up again.

See, are you in agreement with this? It’s not a one way path. It’s not information being thrown at you. It’s information being given to you in a way that allows you then to assess, do I agree with this? If I take something or I do something here, would it change it? And would I agree with it better than I can write it out?

So I think that’s the elements that I’m really interested in. How do we make interfaces that lower both algorithm aversion, also lower automation bias. The other thing [00:53:00] I’m really interested in is the idea of skill degradation, if AI becomes well, how do we maintain our skills as pathologists?

Do we, do our skill just go down or we how do we maintain that? First of all, will it de skill us or will it up skill us or will it change our skills? It’s an interesting discussion.

Aleks: I think it’s going to change our skills. 

Richard: Yeah, exactly. 

Aleks: I recently had a discussion that was about language models and automated reporting.

And the concern was, hey, but then like you learned. It took you so long to learn this report writing and now like a tool is gonna just write it for you. And I compared it to manual transmission versus automatic car. Like people in the US most of the cars are an automatic transmission. They don’t know how to drive stick. [00:54:00]

Do they need to? No. They can just buy the automatic cars. This is the most prevalent car and they can get wherever they want to get. It’s so much easier. And you are still in the driver’s seat. So the skill of driving a stick is like, why would you need it? And I’m from Poland, so I learned with the stick.

But if I was to buy a car I would buy an automatic, please. Why should I? Choose my cognitive power to figuring out which gear to put it in. Now I’ve been driving so long that it’s automatic. So it’s not really a problem for me, but traffic it is. And why not capitalize on this thing? So I think it’s going to change us for large language models.

It’s going to be, okay, how do I communicate with the AI, exactly what you’re talking about, the interface. How, what are the limitations of the method? And I think people [00:55:00] like you say, the algorithm aversion, it’s like, it’s funny because every method has limitations, like PCR has limitations. You always have the control, negative and positive control.

You have to interpret whether something was contamination. You just don’t like blindly take it and it’s Oh, these are my results here. You interpret it and it’s the same with whatever the output of the pathology algorithms are at the moment. Yeah, the concept of method limitation is…

Richard: I think the challenge for us with AI pathology, it’s so near to what we do.

So PCR, we can almost abstract it that it’s, we can almost be, but I think some respects with algorithm aversion, we, it’s so near to what we are doing ourselves that we can fear it to us, or some pathologists at least can fear it to a certain degree. But I think, [00:56:00] 

Aleks: Yes. 

Richard: I think it’s going to be really interesting the way we maintain the pathologist skill and at the same time, optimize automation and I’ve been thinking about what can you do?

And I think, for example, in radiology, they talk about having AI free periods where you switch off the AI and you have to use your own knowledge, efficiency. The other one is that I’ve learned that… 

Aleks: Maybe should be integrated into the software, like automatically sometimes you just don’t get the markup and you’ll have to do it yourself.

Richard: Exactly. And the other one I’ve seen to make sure, or heard of, I haven’t actually seen a product, is that when you’re doing quality control a pathologist, that you actually put in a system, which is wrong, it is opposite to what you and check the pathologist doesn’t just agree with it anyway, so something that, that is really obviously it’s benign, but you’ve highlighted as a grade five decent, and then just check that the pathologist is not maybe put four or five of those in a daily routine for one day in six months [00:57:00] and see what happens.

Aleks:  Oh, interesting.  But then what do you do with all the other results? It’s always Oh, if the pathologist failed, then are you like taking away everything? 

Richard: Yeah, it’s going to come up with a lot of questions, but I think it’s going to be I think…

Aleks: Maybe it’s questions we have to ask each other. 

Richards: Yeah, exactly. For me, the fun question is that how, how do we make sure, because you can have, if you think about errors of omission, you can have, Errors of omission where you’re missing critical findings, not flagged by the AI, for example.

Then you can also have errors of commission where you’re accepting incorrect things that the AI has done. And how do we have interfaces that highlight the possibility of these? Highlight, okay, I’m focusing on the I’m focused on this cancer, but maybe there’s an inflammatory component going on elsewhere that I haven’t looked.

So focused because the AI is highlighting it on eMap, this, but I haven’t seen this. So how do, and the [00;[58:00] solution is in the AI itself. So this doesn’t mean that AI is useless. This means that we can solve this problem somehow by the interface and how we highlight things, how we make maybe checklists, have you done this, okay.

Yes, I have. I can sign it out. So really exciting. And it shouldn’t be used to negate the potential of AI. It should be used to understand it. And develop things to make it even better.

Aleks: Yeah, I totally agree.

Aleks: So if someone was interested in starting working with the AI tools in their practice, so in diagnostics, for example, what would you recommend as the first?

Somebody has a diagnostic practice and they want to do AI. What’s they do?

Richard: First of all, I think they would see,

Aleks: You call Aiforia, right?

Richard: Yeah, exactly. Can we sort it all out? No, I think, first of all, it is really important to choose a company really carefully because you’re not [00:59:00] buying software, you’re buying a relationship, you’re buying a system, you’re buying something where you will have quite a close relationship with that company for a period.

So you have to know that it’s not like you just deliver, you buy it on PlayStation, you buy a game and that’s it. There’s an incredible process of sort of integration, optimization that I think most people, most pathologists maybe don’t understand, but that is part of the process of implementing AI is it, if you get an AI model and you put it into your system, we don’t expect it to work perfectly when we just use it as the raw model, we expect the period of optimization and with those, and that’s holding, us being regular meetings, discussing problems taking images from them and annotating.

So we optimize the model into that system because. All sort of systems are different. All stains [01:00:00] are different or, a stain in in Seville is different to a stain in thing that’s subtly different. And AI really picks up on those. So I would say really do pick the company that you’re going to choose that has…

Aleks: That you like… 

Richard: First be a pathologist available because that’s who you want to talk to.

That’s who you want to be able to to have a discussion with. But also who’s going to be there and he’s going to be understanding the problems you’re having and allow you to go through it. Secondly, think, what do you want to get out of that model? What is it that you’re trying to do?

Are you trying to save time? Are you going to try to reduce inter and intra observer variation? Probably all of the above in many respects. But what is it that you want that model to do? And what organs do you want them to do? Cause not, I think breast and prostate are the most prominent in AI AI human pathology.

So do you have the caseloads for those, for example, to [01:01:00] justify taking it? It’s a cost and you need to have a number of, quite a number of cases to make it worth doing. So I think if I was looking to implement. I would first of all think, what do I want to achieve from this? I think you need to have really clear goals about what you want to get out of AI.

It sounds really sexy and amazing and technological. . And I think a lot of people maybe want that. But what do you want to get out of it? Is the first question you should ask. And also discuss that with pathologists. Do you have a team that are engaged and also want to take on this technology and are ready to take it on?

Aleks: Very important that you want to, because there’s going to be troubleshooting in like both aspects, like a company that you like, the people that you like that has the expertise in pathology expertise in house and people who want to do it, who want to keep troubleshooting because there is going to be a period of troubleshooting, like with any IT [01:02:00] integration.

Richard: Exactly. And this is expectation management… 

Aleks: That as well. 

Richard: The companies have to be clear about, What is the reality? What is the process? This process is something that might take six months, a year maybe shorter but what the process is so that you don’t get a bunch of pathologists who you, they start with an algorithm, they think of man, this doesn’t work, or this isn’t working how I thought it would work.

And then they stop using it. That’s not a, that’s a disservice to pathology and it’s a disservice to AI. It’s and you have to find a way of bringing those two together. So they can start, and I have to say, people who start with AI pathologist, now they’re real trailblazers. They are the ones who are staffing the future. 

Aleks: Yes, my digital pathology trailblazers. 

Richard: Exactly. 

Aleks: Those are the people who are listening to this podcast.

Richard: Yeah. Yeah. And these people these hospital, [01:03:00] these health regions that are going in for it. They’re the ones that are setting the ground for the future.

They’re amazing. It’s, it’s amazing that they are doing it. That seeing the future, they’re seeing that the prospects of it and the future capabilities and they’re going for it. And I think they should really get a lot of praise for that because it’s it’s, it is a little bit jumping in the dark for them.

So they don’t need to jump alone. They can jump with Aiforia.

Aleks: I’m going to link to that like a demo page in the show notes so that anybody who wants to learn more about the system can do that. What’s the best way to contact you if people wanted to talk to you? Is there a way?

Richard: Yeah, of course I can give my email and people can just contact me and send if they have questions, if they have.

Any interest, any ideas, just fire them to me. I love that sort of thing.

Aleks: What’s your Aiforia email? Let’s leave your Aiforia email. [01:04:00]

Richard: It’s richard.doughty@aiforia

Aleks: Okay. I’m going to leave this in the show notes as well.

Richard: Exactly.

Aleks: And If somebody wants to learn more. 

Richard: Yeah, and if they disagree with what I’ve said, question me.

That’s absolutely fine. We can have a discussion about it because it, nobody’s right in this particular situation. 

Aleks: So it’s figuring out still. 

Richard: Yeah. And the way we get better is by discussing it by maybe disagreeing and finding out what is, common ground and things like that. So I think please contact them.

And and if I can help, I will.

Aleks: Thank you so much for joining me. It was a pleasure having you on the show, and I wish you a fantastic rest of your day. 

Thank you so much for staying till the end. It means you are a true digital pathology trailblazer. And thank you so much to Aiforia for sponsoring this episode.

If you’re interested in developing AI for your practice, please for your research, [01:05:00] feel free to contact them. There’s gonna be a link in the show notes or in the description, wherever you’re listening or watching this podcast. There is also a lot of content that we created together with them. So I’m gonna link to the Aiforia playlist, so if there’s anything there that sparks your interest, feel free to reach out on LinkedIn or via email if you’re on my mailing list.

And if you’re not on the mailing list. The best way is to get this book, Digital Pathology 101, which you can download from digital pathology place.com. And this is perfect for people who are starting their digital pathology journey, or for those who have expertise in a specific area like I used to have in image analysis.

And then I wanted to learn all the other aspects. So the PDF is for free on www.digitalpathology place.com, and there’s a copy on Amazon as well. 

So grab your book and I talk to you in the next episode.