238: How Do We Know AI Is Ready for Pathology

Do you really need a scanner, whole slide images, and AI infrastructure before you can start in digital pathology?

In this episode, I argue that you do not.

I’m Dr. Aleksandra Zuraw, veterinary pathologist and digital pathology educator, and this talk is about a belief I hear all the time: I don’t have the tools yet, so there is no point learning digital pathology. I used to think that too. When I was training in Berlin, there was one Leica 6-slide scanner, and it felt like digital pathology was only for a small group of chosen people. That experience made the field feel distant, exclusive, and not really available to beginners. 

What changed for me was not a new scanner. It was a small project.

I needed a more consistent way to quantify a senescence marker in archived skin samples, so I used a microscope camera, captured images, opened them in Microsoft Paint, and manually marked cells with colored dots. It was scrappy. Very low tech. But it was also digital, consistent, and verifiable. That project became my first real step into digital pathology and helped me get my first job in the field, where I worked between pathologists and image analysis scientists on biomarker quantification and patient stratification problems. 

That is the core point of this episode: knowledge unlocks technology.

Scanners matter. AI tools matter. But the deeper bottleneck is whether enough people understand how to use these tools, ask good questions, and connect pathology expertise with digital workflows. That is why this episode is really about readiness. Not readiness of the hardware. Readiness of the people.

I also talk about Dr. Taladzer from Pakistan, whose story makes this point even more clearly. At the time, Pakistan had around 220 million people, about 500 pathologists, and zero scanners. She still started learning digital pathology during COVID using a microscope and camera, joined the Digital Pathology Association, taught herself from papers and online resources, and kept going even after multiple AI vendors rejected her because she did not have whole slide images. Eventually, she found a DIY image analysis platform, learned to annotate and train models on static images, completed projects quickly, and went on to publish more than 10 digital pathology papers without ever using WSI.

Why should you listen?

Because this episode is for pathologists and lab leaders who are interested in digital pathology but still feel stuck at the beginning. It is for people waiting for permission, perfect infrastructure, or a formal roadmap. And it is for trailblazers who came back from a meeting or conference energized, but need a practical way to turn that energy into action before it fades.

I also address an important AI question near the end: How do we know an AI model is good enough for pathology? I talk about why models are only as good as the pathologist annotations used to train them, why concordance between pathologists matters, how orthogonal labels like IHC can improve model quality, and why pathologists still need to stay in the loop as these systems develop and get deployed.

If you are trying to figure out where to start, this episode gives you a practical answer: start where you are. Start with what you have. Start learning now.

Episode Highlights

00:00 – Why the real barrier to digital pathology is usually not the hardware
00:33 – What it feels like to be at the beginning of the digital pathology journey
02:50 – My first practical digital pathology project using a microscope camera and Microsoft Paint
05:37 – How that low-tech project led to my first digital pathology job
08:52 – Why knowledge, not infrastructure, is the real unlock
09:57 – Dr. Taladzer’s story: starting digital pathology in Pakistan with zero scanners
12:03 – What happened after repeated vendor rejection and why persistence mattered
14:39 – The “forgetting loop” vs the “commitment loop” after conferences
16:48 – Practical next steps: book, PubMed alerts, journal clubs, webinars, vendor resources
18:52 – Why I believe digital pathology is the gateway to faster diagnosis
20:00 – How to think about whether an AI model is really ready for pathology

Resources Mentioned

• Digital Pathology 101 – free book recommended as a starting point for learning digital pathology. 
• Digital Pathology Association – mentioned as a learning resource and professional community. 
• PubMed alerts for AI and digital pathology. 
• Journal clubs – mentioned as one way to keep learning consistently. 
• Webinars and vendor resources – suggested as practical ways to keep building knowledge. 
• A4A – the DIY image analysis platform that supported Dr. Taladzer’s early work with static image annotation and model training. 

Transcript

00:00:00
Thank you so much for coming. And I love the analogy uh that we just got about how boring it is to look at normal stuff at the airport. That kind of reminds me of my job as toxicologic pathologists because 90% of what I look at is boring and I do sometimes find it therapeutic. I listen to podcasts. But today I want to talk to you about something that unlocks digital pathology, AI and all these technologies that uh we now learn about or or have access to. And it's not what you think because how many um of

00:00:33
you thought well I don't have a scanner so what's my point what's the point of learning about digital pathology I don't have access to AI tool what's the point of of doing this like anybody thought that ever that was uh me at some point and today I'm going to change that belief so we often think I need so much before I can start but This is a myth actually because the deep barrier is not the infrastructure, it's the hesitation of people and uh those of you who are here who maybe are like at the beginning

00:01:11
of the digital pathology journey and I recognize that we are all at different station of the digital pathology journey and it's not my analogy it's Dr. and Neil perwan is this train station train ride of digital pathology >> I can look at images of which are diagnostic quality and I I feel confident in my diagnosis as a pathologist >> everybody is at a different station and some of us are at the very beginning and we are all motivated you are motivated you're not skeptical but I hear from my

00:01:40
people that send me emails that they are paralyzed because they don't know uh where to start they're asking asking, "Do you have any projects for me?" They are waiting for permission. They're waiting for like somebody to take their hand and put them in the digital pathology uh universe, digital pathology ecosystem. And they are waiting for resources. They're waiting for clarity. That was me in 2013. This is me over 10 years ago. Uh younger and a lot more naive. And uh this is uh I was training

00:02:17
to become a veterary board certified veterary pathologist in Germany in Berlin and that institution had one scanner and my impression was that digital pathology was just for the elites for the chosen. Why did I think that? Because there was this one scanner. It was one six slide Leica scanner that was only used for special occasions. And by special occasions I mean maybe some mystery cases, some conference scanning. And there was only one person, one very special person, PhD researcher who was already like four

00:02:50
years ahead of me that was allowed to work with the scanner. So to me coming from Poland, just starting being like at the very beginning of my pathology, not even like digital pathology journey, I was like this is for the elites, this is for the techsavvy people like not really for me. Anybody had these thoughts before? I see a few hands up. until I had a research project, one research project where I had to stain skin samples. It was a scinessence project. So I had to find some scinessence marker

00:03:23
for I don't know fibroblast some I don't know which which cells in the skin but basically I h had to quantify the scinessence like how old these samples were from archive and I needed to be consistent and when I looked at these slides I'm like there is no way I can consistently go back to these slides and estimate that one has more and the other has less. So it was totally a guesstimation and I had no scanner access right access scanner was for the elites. So I had this project right I

00:03:56
look at these at the IHC and I'm like not it's not possible for me to consistently evaluate it. So and scanner was not an option right but microscope camera was an option so I took digital images of these samples. I just like figured out which what was my criteria, how like far away from wherever. I basically described it to have some kind of consistency and then I did these pictures and I opened them in Microsoft Paint and I just like with a mouse was crossing the ones that I wanted to count

00:04:32
and uh yeah, I was putting colorful dots. I didn't even open it in Apirio image scope that has an automated counter. I was just like drawing it and later counting it. But I could show these pictures to somebody. I could ask them, "Hey, am I counting the right ones?" And like they could see that I actually was counting uh the ones that I was counting. So it seems very scrappy, very manual, very like DIY version of digital pathology. But the key thing here was that it was consistent, that I

00:05:05
had data that I could show somebody and somebody could verify, okay, is this good data or is this bad data? you're maybe counting the good ones or the bad ones. If that was visual, it wouldn't be possible. And that kind of shaped one belief that I have that every time I hear about the different pathology scorings that they're very much visual guesstimation, I get triggered because I would rather calculate every day of the week and twice on Sunday. I would like do these dots if necessary rather than

00:05:37
guesstimate. If you can measure and now with digital pathology you can then let's do it. And this scrappy project led me to my first digital pathology job. My job my first job after this residency was at an image analysis company. This company is now part of a pharma company Astroenica. And I went there to this interview. They were asking me about all different kinds of immunology markers that I had no idea about. I think I could maybe mention CD3 and CD8 and two components to to getting this job. One was this one project where

00:06:15
I said well I decided that for this project I had to have something digital quantifiable and you guys have this software and you have the power to basically unlock a lot of quantification for tissue. And the second part was he looked at the list of languages that they speak and it was like five at the time or six and he said oh you speak six languages you can learn anything. So these two components got me the job. I'd like to believe that my experience in digital pathology was the main driver

00:06:45
and then I started working. This door opened when I took the first step. So I wouldn't without doing this project on uh in Microsoft Paint I wouldn't be even able to talk about the value of quantification to a company that was doing quantification. I probably would have no idea what they were doing and like what is the value in in what they're doing. So um I went there I was it was pretty nice feeling to get straight out out of a residency and go to a environment of non-pathologists

00:07:17
where you are suddenly the expert on tissue feels pretty well. Anybody who just finished residency or is in residency can relate to that or in any type of training where you're just like barely finished. there is so many of people who are a lot more experienced and then you change environment you suddenly have some magic knowledge that people don't have. So that was my magic knowledge that knowledge about tissue and I was supposed to work with image analysis scientists who knew the tech

00:07:45
who were able to detect plenty of different things in the tissue and my job was to make sure that the what I'm what they are detecting actually matches the tissue. It wasn't always the case and it took us it took me some time to acquire the knowledge. Okay, how do I communicate with them? What does this software actually do? They knew everything about computer vision and I was teaching them tissue. So at some point we had enough common understanding of each other's expertise that me

00:08:16
knowing the tissue and them knowing the tech was not just 1 plus one is two but it was exponential unlocking of whatever was whatever were the questions of our customers and that was pharmaceutical industry. They wanted to do biomarker quantification, patient stratification, different signature discovery. So, this helped us get a lot more out of this than me doing my dots and them trying to uh do image analysis on something that they didn't really know what it was before. And that brings me to the point

00:08:52
that knowledge is key. Knowledge unlocks the power of technology. Not everybody is going to ever have the access to the same level of technology, but technology exists. Scanners are getting cheaper hopefully and technology is getting cheaper. I know the chips the Nvidia chips are getting cheaper every year. So let's uh hope it translates to other tech as well. So so this tech is out there and the bottleneck is actually people who understand like how how this all works. My example was from image

00:09:24
analysis from computer vision. I had to understand enough of computer vision to be able to guide people who do that at a really high level. They had to understand enough tissue but the same counts for scanners for different types of AI for image management system for you name it anything in digital pathology. So people who understand is what we need. This unlocks the power of technology. But I am just one story and uh you can you know say hey you're not like I got introduced as the celebrity

00:09:57
which is very nice but like you may have the impression okay you're the celebrity you've been doing this for for 10 years. You are in Europe you are privileged. Uh and that is right. I I was lucky. I am still lucky to be able to to be to be working in this space. But I want to tell you about a different digital pathology trailblazers. Somebody who who had it like not as fantastic as me, which I didn't think I had it so fantastic at the beginning. Let's go to Pakistan. Anybody from Pakistan? So

00:10:27
Pakistan, feel free to correct me if I'm like mis saying anything about your country, but we have 220 million people there. And according to the person I'm going to be citing, it is around 500 pathologists there and exactly zero scanners. Maybe now you have a scanner. So just to give you some context, I'm from Poland. We also have 500 pathologists, but we only have 38 million people in the country. So until I learned about this situation, I thought, "Oh my goodness, only 500 in

00:11:00
Poland." I'm like, "500 in Poland is a lot compared to Pakistan." And the trailblazer I want to talk to you about is Dr. Taladzer. She so she started with these constraints and I had the honor to interview her on my podcast last year or or two years ago. So yeah, zero scanners, microscope, camera only. And she inspired me because she started learning anyway. So her story with digital pathology started during COVID when everything was shut down and she decided okay let me figure out this digital

00:11:32
pathology thing and she didn't have anything digital pathology she had glass slides and microscope and a camera she joined the digital pathology association and started learning and they have fantastic resources she just started learning about it and then she was reading papers research on digital pathology and then she conducted a survey in Pakistan about the awareness of digital pathology. So she didn't like even touch any digital pathology tools and she started doing work in the digital pathology space. So no

00:12:03
infrastructure and she just started with knowledge and then she wanted to do some more right AI is so cool. I like AI as well. So she started reaching out to vendors and everybody said no especially because most of the models that are being created are on whole slide images. When somebody hears that, oh, you you don't even have a single whole slide image, then obviously it's easier. Uh, well, they say no because it's not really applicable. So, it was a rejection after rejection after

00:12:33
rejection, but she was pretty determined. Uh, I encourage you to listen to that episode. And she persisted. She kept asking. And then one company said yes, A4A. And they have a booth in our exhibit hall. So whoever is interested in this and what they have can visit. Kudos to them for saying yes to Dr. Taladzera. And she got the access to the platform. Granted, this is a if you don't know what this platform is, it's a kind of do-it-yourself image man image analysis software where you can

00:13:04
train a model with your annotations. So if you have the tissue expertise as a pathologist, you can start basically building your models without too much further knowledge. And she said herself that she was learning from zero. She had zero idea how to annotate. She didn't know like what these epochs are for training like nothing. And you know who helped her? Customer success representative from Aphoria. And then in just one and a half month that she had access to this software, she um completed three

00:13:35
projects and she published on these projects. Then she changed her career from whatever her previous track was to digital pathology, computational pathology. As of now, this is Google Scholar from couple of days ago. It's definitely over 10 publications and most of these publications are on um static images. They're not on whole slide images. Different models, different options on how to do digital pathology and oh, she's pretty creative, I have to say, when I looked at the like what all

00:14:06
you can do with static images. So, no need to be disheartened if you don't have a scanner. No need to be disheartened if you don't have access to AI. There are ways around it. And what what our stories show is that you start with knowledge. You start with figuring out what that is. How can you leverage it? How can you leverage it for where you are? H because then technology follows or you follow somewhere the technology. This is the pattern of transformation. And now we're here. Everybody is here to learn to network.

00:14:39
This conference is a little bit like company Christmas party where where they fly you to a to a place like a company that has multiple locations. They fly you to a cool place. You have not seen your colleagues from other sites for a long time and it's so much fun and you're going to do so much because you're so motivated and it's exciting, energizing. And what happens after? Well, we go back to normal life. The energy drops. We have to like read slides. If you're a pathologist, manage

00:15:07
projects and basically do like whatever is expected of you to keep um earning money. So this kind of fades. So we can do two things. The first path I call it the forgetting loop. You just you came here, you learned, you forgot and next year you maybe are going to come again because you are still interested in this topic. It's a fascinating topic. That's why we paid money to be here. and next year you're starting over. The loop repeats. We don't want to be there next year, right? We want to do something

00:15:39
else. The alternative is that you come, you attend the conference, you learn, you absorb the insights, new insights, and then you commit. You commit to keeping learning beyond today. So in one year, you're not going to be starting from scratch. You're going to be asking sharper questions. You're going to know where to to which vendor to go, which questions to ask. You're going to know exactly like what could be leveraged in your institution and you will be able to figure out your road map. I'm sorry to

00:16:14
disappoint you that there is no one road map fits all and every country, every institution, every like geography is going to have their own way of doing digital pathology. But by committing to learning about it, you will know what your way is going to be. You're going to identify opportunities that were invisible before and you're going to lead them. You may become the go-to person or the celebrity in your institution for digital pathology. You can lead projects, start pilots, and you are already starting here at the

00:16:48
conference. I want to encourage you to start with a book that I published. This is a first step that you can take. You can scan the QR code and a free copy is going to be available to you. And most of you who are already have seen me online or are getting emails from me probably have the book. But I'm just suggesting this as the first step, a step that I provided for you. But it doesn't have to be me. If you like are annoyed by me touching my glasses or whatever, then doesn't matter. It

00:17:19
doesn't matter. You just pick something else. pick something and then keep going. You can just set up your PubMet alerts for AI and digital pathology, which is what I do. And some people join me at 6:00 a.m. every Friday to discuss abstracts from PubMed. Kudos to them. You are amazing. I have a few regulars that join me for Digipath Digest Journal Club. And in another resource that I want to point you to is the seven-part series that was co-authored by Dr. for Lon Pantanoit who is here. We discussed

00:17:52
that at the journal club but it can be pummet alerts it can be webinars by a vendor if you have a vendor like A4A customer success representative helped Dr. Taladzer with understanding something that led to like so many of her publications and the the vendors I have to say they have pretty good resources. So you can do that. You can do anything. You can read books whatever. The only wrong choice is that you don't choose anything. The only wrong choice is to just go back to your normal stuff and not do anything. Not

00:18:25
like do this consistent learning commitment because you choose your path. I would be honored if it's going to be what I put out there on the internet, but it doesn't have to be. It doesn't matter. I strongly believe that knowledge advances digital pathology. And when you commit to learning about it in whichever shape or form, you commit to progress and you commit to advancing the field and by that because I believe that uh every patient has the right to fast diagnosis and that digital

00:18:52
pathology is the gateway to fast diagnosis. you by learning about digital pathology whichever aspect you're now interested in are committing to providing them a faster diagnosis and better treatment because I believe that knowledge drives adoption and you have everything you need to start today. Well, maybe you do need a microscope camera. If you don't have one, just let me know. We'll figure out the way to to get you one and you can do it. You can start today wherever you are either at

00:19:21
the very beginning of the digital pathology train ride or wherever you are to get to the next station. Thank you so much for your attention and I think we have some time for questions. We're talking here about is our model good enough? So most of the well not most but a lot of the models at the beginning uh were trained by uh pathologist annotations. So then like how good can the model be only as good as the pathologist that is providing the annotations and there are publications on like whether how to annotate to be

00:20:03
more consistent whether annotating and this being called ground truth is actually the ground truth but like let's say real world you would put three pathologists to do the same thing. You would check how concordant they are and that is the level of your concordance of that you're expecting that we are currently getting as standard of care and then you have a model if the model or or the algorithm achieves more or less what they achieve then it's as good as them do we want the model that can do

00:20:34
better be as good as them maybe we want it to be better so there are different ways of providing better annotations and I think we heard it in the previous talk but One thing um that I like is no it was in the in the talk in the different room h where you can provide orthogonal method like a different method of uh labeling structures for example IHC for epithelial cells in the lymph nodes that reading lymph nodes may be therapeutic but uh making annotations may be late less therapeutic. So you want to do IHC

00:21:09
and see and use this as labels and then you can have a model that is better than the pathologist. I I I can give you an my perspective and perspective of people who are kind of involved. So not those who like you have to bring on board but basically being somebody who is in the loop who is part of the team. the algorithm development team, the software development team, the deployment team. And that kind of ties into okay, don't take this special power, special knowledge away from me because first of all, I'm going to keep

00:21:48
feeling like needed special and that's for every professional, but pathology is like pretty unique, right? So everything that touches pathology just keeping a pathologist in the loop in whatever and the role is going to be changing, right? Maybe it's going to be supervision of a model of helping software developers develop a software that is easy to use. But basically having them as part of the digital pathology team, digital pathology efforts. I'm going to tell you about something that pathologists like

00:22:21
did not like. There was a a diagnostic company in the veterary space that started deploying an AI model directly to clinicians and you could get pathology readouts from this AI algorithm like in five minutes without pathologist supervision or you could pay extra to get pathologist expertise. That was kind of terrifying for the people in the audience who were pathologists. They're like then not only are you pro and it was in the venary space. Not only are you providing like totally unsupervised thing that the clinician

00:22:57
may not know what to do with it and you're not even asking the expert. So it's like okay for chat GPD you can read through the text and verify okay is it BS or not but for something that you don't as a clinician don't have an expertise in you don't know if it is reasonable or not. So they did not want to take part in that. they did want to take part in being in the loop in whichever shape or form. I hope that answers the question.