No More Microscopes. How close are we to glassless pathology? w/ Dr. Richard Levenson, UC Davis Health

Dr. Richard Levenson joins the Digital Pathology Podcast for an insightful discussion on innovations in digital pathology. With an eclectic background spanning the humanities, pathology research, and experience in startups, Richard offers a unique perspective. He is currently pioneering groundbreaking research in glassless pathology techniques and AI applications at UC Davis Health.

Richard first developed an interest in research during an internship in Dr. Judah Folkman’s renowned lab studying angiogenesis. He went on to medical school at the University of Michigan and completed his pathology residency at Duke University before entering industry. At Duke, Richard also began pursuing his own lab work around the plasminogen activator and its role in cancer malignancy.

But as Richard details, he struggled to progress his research there given conflicting experimental data and results. This experience highlighted key lessons around embracing anomalies and paradigm-shifting findings. While it may have stalled Richard’s early research ambitions, his openness to new directions would prove formative.

After a period of challenges, Richard went on to help pioneer multispectral imaging platforms at Cambridge Research & Instrumentation. Their work laid the foundation for influential tools still widely used across digital pathology today. This early foray into industry kindled an appreciation of emerging technologies and their potential clinical value.

Now at UC Davis, Richard is pushing the boundaries of real-time tissue characterization with microscopy advances. His lab conceptualized MUSE (Microscopy with Ultraviolet Surface Excitation) and FIBI (Fluorescence-Imitating Brightfield Imaging) microscopy frameworks that image thick fresh or fixed tissue sections without traditional histology workflows.

MUSE utilizes deep ultraviolet light at 275-280 nm to excite autofluorescence for subsequent standard camera detection. FIBI leverages blue light to similarly generate intrinsic color patterns reminiscent of H&E staining. Both facilitate direct on-site imaging without frozen section processing.

Commercialization is also underway to bring these technologies to market. Richard’s startup Histolix has studied over 100 specimens using FIBI across diseases like breast cancer. Their validation achieved a 97% concordance between glassless FIBI and standard pathologist H&E reads—a remarkable result suggesting clinical utility.

Such solutions could soon transform procedures relying on frozen section analysis. Richard believes slide-free telepathology will enable rapid turnarounds critical for real-time decision making in settings from operating rooms to global health clinics. It exemplifies the move toward fully digital, computational histology analysis.

Richard meanwhile contributes strongly on the responsible AI front through his research and educational outreach. His lab has pioneered AI techniques to convert non-standard histology images from modalities like MUSE and FIBI into H&E-like digital slides. This facilitates pathologist interpretation and fuels the development of computational analysis tools.

However, as detailed in Richard’s recent paper on “AI Pathology – What Could Possibly Go Wrong?”, risks around bias, cost, access, and error mitigation remain. Addressing these through proactive policies and oversight will prove critical as AI sees increased clinical integration across specialties like digital pathology.

By driving multiple initiatives around ethics, global health outreach, instrumentation, and education, Richard’s influential body of work has significantly shaped digital pathology’s ongoing responsible evolution. His approach synthesizing humanistic and technological perspectives makes him an especially compelling leader in this transformative age for diagnostic medicine.

Support the show

• Histolix website
• A Pilot Validation Study Comparing Fluorescence-Imitating Brightfield Imaging, A Slide-Free Imaging Method, With Standard Formalin-Fixed, Paraffin-Embedded Hematoxylin-Eosin-Stained Tissue Section Histology for Primary Surgical Pathology Diagnosis
• Pocket MUSE: an affordable, versatile and high-performance fluorescence microscope using smartphone.
• AI in Pathology: What could possibly go wrong?
• Pigeons (Columba livia) as Trainable Observers of Pathology and Radiology Breast Cancer Images.

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

• Digital Pathology Club Membership
• Free Digital Pathology 101 eBook

• The Evolution of Digital Pathology – from Improved Histology Quality to Fair Use of Pathology Data w/ Matthew O. Leavitt, DDx Foundation
• What Does the FDA Say About Digital Pathology for Nonclinical Toxicology Studies?
• HistoWiz: fast histology and an image database for mining from Brooklyn NY w/ Ke Cheng
• Podcast
• What is the current role of Artificial Intelligence in digital pathology? Part 2 – state of art.

Aleks: [00:00:00] Welcome digital pathology trailblazers. Today, my guest is Dr. Richard Levinson. And let me start with a story. How I met and met in the quotation marks because we actually never met in person.

But I got introduced to Richard, which is work. I didn’t know it was his work. I mean, I heard the author of the paper, but it was still during my residency at the veterinary conference in full that Germany,  where it was the first time when his paper about training pigeons was introduced.

And the title of this paper is Pigeons as Trainable observers of pathology in Radiology Breast Cancer Images.  And this was presented as like kind of a funny thing, funny research.

And I didn’t know it was Richard’s paper, but I believe any pathology will recognize the picture from this paper, a pigeon sitting inside a box with some grain and disposing device [00:01:00]that gives the pigeon a reward for a collect a correct classification of the malignancy of the tumor.

So that was the first thing. And the second thing was quite recently, and then we were already connected on LinkedIn and we kind of started the conversation was in our paper called AM Pathology. What could possibly go wrong where there is a list of what could possibly go wrong. And that’s what we’re going to be talking about, though.

Richard, welcome very much to the podcast. How are you today?

Richard: Thank you so much. I’m reasonably well. It’s sunny and gorgeous here in Davis, California, for those of you who don’t know, and that was probably 98% of you is is just 12 miles away from Sacramento,

where the UC Davis Medical Center is.

Aleks: Fantastic! I didn’t know it. I mean, I knew you were in California, UC Davis. I didn’t know exactly why. So, Richard, let’s start with you. Let’s I would love you to introduce yourself to the listeners

and talk about yourself. Talk about your research. Talk about the focus of your research. And then we’re going to dive in into the episode.

Richard: Thank you so much. It’s a real pleasure to be [00:02:00] here or in Davis. And my particular story is very, very unusual. And in fact, I was asked to give a talk on it or talk about me to I think it was at

Dartmouth. And the title of my talk was My Path In and Out of Path, because I literally in and out of  pathology over my many years, and I actually started as an undergraduate studying early modern English,

French and German history and literature.

Aleks: Okay.

Richard: Yeah!

Aleks: I think that explains parts of one of the papers

Richard: That entirely possible and the title that I should have had for my thesis, but I didn’t have the courage back there, back then was it was about the French restoration. And the title should have been  Charles the 10th Bourbon on the Rocks. But I didn’t. I didn’t dare to do it.

Aleks: You didn’t do it?

Richard: With deep regrets. So managed to extract myself [00:03:00]from history literature and went to medical school. And then…

Aleks: Did you actually, like, practice the literature? Did you?

Richard: Well, actually, I’m a post-M.A./B.A. at the time that University of Cambridge or Cambridge  University. To this day, I don’t know whether it’s University of Cambridge, Cambridge University,  but one of the two and I was doing more history and literature there. And interestingly enough, with the  same tutor that my mother had had when she was in Cambridge, it’s bizarre, but fortunately for me,

I had spent the summer before that in a wonderful thing. Judah Folkman laboratory to have before many of you will recognize as the sort of founder of of modern angiogenesis research the words how tumors largely how tumors get there.

Aleks: How is literature? Researcher Did you spend the summer in angiogenesis lab?
Richard: I have one word for you. Nepotism.

Aleks: Okay. Like and literally the connections is everything.

Richard: My nepotism literally [00:04:00] means sort of doing your nephew a favor. And it turns out that my uncle had been a neighbor of Judah Folkman, and so this was literal, classic nepotism.

And I had I had to be fair, I had done enough of sort of pre-med courses to qualify to go to medical school.  So I wasn’t completely naive. I knew what organic chemistry was, etc., etc.. And so I got put into

Judah Folkman slap that summer and it was such a fantastic experience that history sort of paled in comparison. My last one term at Cambridge and only managed to learn and forget how to do the London  Times crossword puzzle that I came home and rejoined the lab and then went to medical school the next year.

University of Michigan.

Yeah, there of the other sort of fortunate thing was that in my years, all smart kids went into internal medicine. That was sort of the track, you know, the the not so smart ones went into orthopedic surgery and the smart ones all went into internal medicine.

So I was [00:05:00] tracking myself there was no, you know, no no real internal sense of destination. And fortunately, one of my friends said, have you considered pathology? And I hadn’t. In fact, just to the extent that I had actually failed the final in pathology, I passed the course because of quizzes. But I feel 69 is failure.

I got a 69 on the final and he said you should go into pathology because he knew I was interested in research because it means that you can study anything you want. If you go into Durham, you have to do in for the rest of your life and pathology that study anything.

And secondly, if you go to a good academic center, you can be on the autopsy service,  which is like only a 20% commitment. And the rest of the time you can have four for your self-lab. And then also you don’t have nights and weekends if you’re on the autopsy service because they just put the body in the chiller over the weekend. And it sounded great to me. And that’s exactly what I did. I went to, you know, I went after Michigan Medical school. I went to Duke University for to be on faculty.[00:06:00]

And along the way, I also managed to help one of my friends who was also one of the smart kids who were going into internal medicine. And but because she was smart, but she was so under-committed to that,

that she was in horrible despair and was crying every night. And I said to her, Have you considered pathology? And she had not. And she looked into it and she loved it. And she became a career.

Aleks: What is that like? But actually I had a similar experience. Like I didn’t even know when I was studying in Poland and studying veterinary medicine. I didn’t even know the pathology was as a veterinarian,

you could be a full-time pathologist. I like I liked it. I was lucky enough to have a board certified pathologist teach me pathology because I did it in Spain during my exchange year and we had like a private class

for those exchange students because we were like they were transitioning the curriculum and they didn’t really know where [00:07:00] to put them. So we had this like nine-people pathology class with a board-certified

pathologist, and I’m like, Wow, this is so cool. But after I went back to Poland to finish vet school, I had no idea I could do this full time because in Poland, your at the time you had four veterinary schools and

the only veterinary pathologists were working in those schools. So maybe you could like 20 veterinary pathologists and we didn’t we still don’t have any accredited residency centers. So yeah, people don’t know  how good

pathology is and like how life-compatible it is.

Richard: Exactly. And let me just finish the story about my friend and then I’ll call a little bit about the state of pathology training for medical students. So anyway, my friend switched out of the internal medicine

destination to pathology and she later that year or later, a few years later, she told me that I had saved her life. So she was the only, only person whose life I saved in medical school. I was once a fellow student…

Aleks: As a pathologist. But you know what? The physician burnout is a real, [00:08:00] real thing. So,  yeah, it’s a real life-saving. And you know something?

Richard: I’m of a certain era. We had a whole pathology course. As for medical students, like a whole semester, every every several times a week and the pathology lab and everything. And now days…

Aleks: Are you telling me that that its not the case?

Richard: Not the case anymore? You get histology, but pathology just shows up as sort of an afterthought for case presentations…

Aleks: What?

Richard: You know, what…

Aleks: I’m sorry, I’m like in shock and hyperventilating in American med schools. You don’t have a  pathology course.

Richard: I don’t know and I can’t generalize, but it’s certainly changed dramatically in the while. Just show up is sort of part of team teaching where you get the clinician and then you get the surgeon and then you

get the pathologists, as I was really ex and they just show up as sort of the answer giver rather than, you know, you know, very sad.

Aleks: This is bad, I hope like I get a bunch of comments, this is not the case in the US school we have pathology [00:09:00] is of course I hope I get this.

Richard: Please do. This is just my impression. But I think it’s it’s sort of generic. There’s so much demand for time, you know, teaching time, so many so many things the school has to get through.

The pathology is taken a…

Aleks: But that explains why people just don’t know what know if they can move, of course.  Richard: So that’s why we need air pathology because there any pathologists,

Aleks: Maybe then it’s going to be also cooler because it’s going to be AI-powered and then more people would be interested in that. But the why path in and out of Pathology,

Richard: Actually, I was at Duke for nine years, assistant professor. I had my own lab and I started to get  into technology. Back in the day, confocal microscopy was a wild and crazy thing and started

confocal microscopy for pathology. And my other lab was doing trying to understand the function of plasminogen activator inhibitor type one in cancer.And it was a I kept getting very conflicting [00:10:00]  results,

so much so that I never my research just never gel. And sometimes I would find it in the cytoplasm and sometimes I would find it in the nucleus. I thought this is just artifacts. And essentially my research cratered.

And so my career at Duke cratered and come to find out sort of ten years, 15 years later, that Pit Pi one is  a really complicated and interesting molecule. And my results were actually accurate and it can be intra-nuclear as well as cytoplasmic. And I just didn’t have the courage to go after these anomalies as if they were true and just or potentially true and not just artifact.

Aleks: Your research was kind of inconclusive, but then it turned out that it was inconclusive for a reason.

Richard: Yeah, I mean, that’s just sort of a moral, which I think has been, you know, the people who who  really are successful are those who embrace results that make no sense and pursue them rather than throw them out.

And they’re the ones who break the paradigms. Right. [00:11:00] And I didn’t have the courage to do that.  So in fact, my career is broken and it’s just another story of how I got out of pathology.

SO my research service done, although I was board certified, I really had no clinical skills left. And it occurred to me that, you know, I really needed to go back and just be a practicing pathologist and the  person

who sort of told me I was no no longer have a job at Duke said, But we’re going to give you, you know, a  year’s salary and you can do whatever you want. And I said, Maybe I just need to go back and be a resident again.

And I think I’ve done something that no one in the history of the world has done. I went from being the professor to being a first year resident in the same department.

Aleks: That is definitely nonstandard career step. Oh my goodness.

Richard: And I lasted about six months there.Not because it was so terrible, but a friend of mine said, you should really look at this. And she showed me an article in I mean, an advertisement in science [00:12:00]  from

Applied Spectral Imaging, which showed the use of spectral imaging to understand histology examples.  So in other words, instead of just looking at sort of clear images of of, you know, red, green and blue of  HA,

if you use a device which they had developed, which was a spectroscopic camera, you could get a color spectra from, from the extinct slides and actually distinguish things that you can’t distinguish by eye.

I thought that was really cool. And I had had this sort of confocal microscopy experience and I called up the company said, Hey, you want you want a pathologist to help? And after some hemming and hawing,  they said,

Sure, and I actually got a job with the company. But at Carnegie Mellon University,  Aleks: okay, oh my goodness!

Richard: Which is a good range and which it doesn’t have a medical school, but it was right next door to  University of Pittsburgh.

And I started developing applied spectral imaging tools for Multi-Spectral imaging and pathology. And  that went well for for about three years. And then things changed as they do. [00:13:00]

And then I took a job in industry in full time, full bore industry, Cambridge Research and Instrumentation to develop a different kind, different technology, Multi-Spectral imaging, which has turned out to have  real legs because as your audience knows about omics, that’s all path omics or multi multi omics stuff.

One of the technologies use that’s used for that is to do fluorescence imaging and use a spectroscopic tools  to actually distinguish more than four or five or six different fronts. So the company that I work for,  eventually their technology ended up with another company called Akoya Biosciences…

Aleks: Which is very familiar to my listeners.

Richard: That’s right. So it’s I mean, that’s really I’m happy because most people who develop, you know,  work on technologies from 15, 20 years ago don’t usually see it come to fruition. But the same thing is still

cooking and is out in the market and making real science. So I’m fortunate to have that sort of part of the background. And so I actually work for [00:14:00] a company, so this is my path out of path like ten years  I became

vice president for the research firm, if I recall correctly, in this technology company, which again is unusual  for a history and literature major.

Aleks: Indeed

Richard: Indeed. And that ended sadly. my, my, my. Sort of interesting joints or, you know, adventures have a a non-classic ending. And one day I was asked to leave because of various reasons and as a but here’s the

thing. As I was driving home…

Aleks: Did you do something

Richard: No it was actually I was looking for another potentially another job and mistakenly pressed the reply all on my email. (Laughs) You know, stuff happens.

Aleks: I already loved this episode. We’re like, still in the introduction part. I love it. I love

Richard: Listening that, you know, there’s a lot more than classic, [00:15:00] you know, career paths out there. But the point I was driving home from being, you know, asked to leave and my car was headed direct into

not just one, but a double rainbow. So that was a sign from the heavens that there was  Aleks: It was that you’re supposed to leave.

Richard: Exactly. And then I did something that, you know, because it was sort of sudden, I did what everyone who suddenly deprived of a regular job, I became a consultant.

Aleks: Yes. Yes. That develop…

Richard: And I was really, that was that went actually very, very well. My my income actually didn’t go down. And my first major job was this is where I get into digipath with OMNYX, if you recall, which is the  joint

venture of G.E. and the University of Pittsburgh, if I remember correctly,

Aleks: Did you get some approval at some point like clearance or something

Richard: Cratered as a company. But I think their IT sort of lived on it may have been absorbed by somebody else. I was working for a year and a half [00:16:00] there as a consultant on the some of the  first, you know,

serious digital pathology company.

Aleks: Which year was that?

Richard: A while ago, probably 80 or 19. I can’t even tell you.

Aleks: I’m like, I’m, I’m, I have this digital pathology timeline which is pre to 2000 pre whole slide scanner and post hosted like counter and then they’re like sub sub parts of this timeline like pre FDA clearance of the Philips and both.

Richard: Exactly! Actually when I was with a client spectral imaging I worked with Dirk Soenksen who was the founder of Aperio.

Aleks: Oh my goodness!

Richard: Yeah. So these are all these chance encounters with, you know, major, major players, which has been very, very interesting. So about 29 is when I when I left CRI and probably around then is when

I started working for Omnyx and then that my my consulting gig [00:17:00] lasted for three years and here’s the last thing that never, ever happens. I’m just sitting at home working from one one of our  bedrooms in

Brighton, Massachusetts, and I got a phone call and the person on the other end of the call said, “Hi,  Richard!” She’s from Romania, I think… “Hi Richard, would you like to be a professor at UC Davis?”

And I said,

Aleks: A professor from Romania calling you from…

Richard: No, I mean, she’s a faculty member at Oxford.

Aleks: Ok

Richard: Yeah. Not no, no. That would be very, very strange. Yeah, I was just trying. Was putting on a slight accent for accents. No, no specificity.

Aleks: So like, out of out, she took out of consultancy with the phone call to become a professor at UC  Davis.

Richard: Right. Full professor with tenure

Aleks: Like without, like out of nowhere…

Richard: I didn’t apply it just said, Hey, you want to be a professor? And I said, “Sure!”.  Aleks: You know what, anything can happen. S

Richard: So I’ve been here at UC Davis for [00:18:00] coming on. It’s something like 12 years and it’s been  really a wonderful run.

Aleks: So didn’t get bored at UC Davis, You didn’t get kicked out. And I thought, Right. So yeah. Tell me about your research at UC Davis.

Richard: Well, sure. Well, I mean, it was at UC Davis that it managed to do the pigeon work. Aleks: Yeah, I want to hear about like, how did you come up with training pigeons?

Richard: It was easy. I’m driving to work and there’s somebody on the radio who’s describing a UC Davis researcher who was talking about his work on pigeons and and visual recall…

Aleks: pattern recall?

Richard: …visual recall. Can they recognize, you know, sort of how many…

Aleks: And you heard this on the radio?

Richard: Yeah, I heard it on the radio. And I thought, wait a second, visual recall. That’s what pathologists do. And so I called up this guy and said, “Hey, do you think it would be a good

idea to look at…” [00:19:00]

Aleks: borrow a few pigeons…

Richard: …the pigeons can do what pathologists do? And he said, That’s a great idea, but don’t talk to me.  You have to talk to Ed Wasserman, who’s the guy in Iowa who actually is the major player in in the in the  field

and all pigeons. And he thought it was a great idea. And then we we just did the research.  Aleks: So who like you build this cage for the pegeons?

Richard: No, that’s his stuff.

Aleks: So. so he already has for training pigeons for all kinds, researching

Richard: Very sophisticated investigations. And in fact, he recently published a paper showing that  pigeons, humans look at images the same way. In other words, what the components that are important  to humans are for

important to pigeons. Which makes sense because pigeons are very visual creatures. And if you think  about the world they live in, they live entirely visually. They don’t do things by touch. They don’t do by  hearing, they don’t

do by smell. [00:20:00] They’re out in that. So they have two tasks. One is to, you know, sort of find their  foodstuffs and the other is to avoid being eaten by camouflage snakes. Okay. Okay. Camouflage snakes.

Aleks: What do you mean by that?

Richard: I mean, literally a snake can can have colorations. I have an image somewhere that look just like,  for example, falling leaves on the bottom of a forest floor,

Aleks: But were talking about literally about snakes? I thought bout falcons or birds of prey. Richard: Yes! Well, I would know. But I mean, a pigeon is is wandering around on the ground.  Aleks: Okay.

Richard: And snakes like to eat birds, and they have to make sure that they don’t wander into a snake.

Aleks: I have never thought snakes are a threat for pigeons, maybe because I come from Poland and we  only have like two species of snakes.

Richard: Maybe so. But yeah, I mean, I actually I’m not making this up. I looked up sort of, you know,  potential predators for pigeons in the in the world. And the point is that snakes have [00:21:00] their job is to catch prey. And they are extremely sophisticated in their camouflage. And if you look at a camouflaged pigeon,  among, I’m sorry, a camouflaged snake among, you know, litter that leaves on a forest floor, it’s pretty much the same problem as telling the difference between, you know, sort of normal and person cancer.

Aleks: Malignant…

Richard: But they’re, you know, they look very, very similar. Can be certainly to an unsophisticated person.  You can tell them apart. And so pigeons have the skills to tell, you know, tiny, tiny pattern differences of heart.

And that’s what pathologists do.

Aleks: Oh my goodness. This is hilarious. So you’re the pigeons at UC Davis?

Richard: Yeah, I didn’t want to do. One of the huge frightening things for me was to come to a place and they say, Here’s your, here’s your office, here’s your lab, here’s your startup package. Good luck, Dr.  Levinson.

What do you [00:22:00] do? Lot. Since I had no clinical responsibilities, everything had to be research.

Aleks: Research.

Richard: And I really didn’t know what to do. The pigeons weren’t a serious thing. They were just a hugely fun experiment that went viral. But that’s not something you can make a career out of. But I knew I was  into…

Aleks: And then it went viral. Yeah, like can if there is any research paper in the pathology world that went viral, this is the one. And I was I was just presenting at the conference at ACVP, the American College of

Veterinary Pathologists annual meeting, and I had slides where I had like snippets out of literature. And then I had this like compressed literature versus real life and my real life example was and the number of followers that clinical pathologists using static pathology and veterinary medicine have and like one account had 29,000 followers, the other one 17,000 they are on, I don’t know, 15,000. And I think that this is more than [00:23:00] all those other publications that they showed before and they showed like ten publications ever had. Like I don’t think people read the stuff, but the pigeon thing, everybody in pathology world, they know about this.

Richard: Just lucky, I guess. There was a paper last year called the title of the paper was Pathologists Are  Not Pigeons.

Aleks: D Yeah, there is a whole subculture of this. They know exactly I need to make a video on TikTok about it.

Richard: And you know, I was told just to get a little bit of sort of career story. I mean, people said, You’re wasting your time. This is not going to help you. This is not serious, you know? So don’t listen to  people. Do what you want to What’s fun?

Aleks: I totally I totally signed this. This now that sentence of wisdom.

Richard: However, I do have to say that said from a perspective of late in life, full professor with tenure  right there at the start position.

Aleks: Who got the position out of the blue.

Richard: Yes. So I even I even wrote a poem says I’m professor, I’m vice chair, I’ve got tenure and I don’t care. And, you know, [00:24:00] that’s I have to admit, that’s a very true situation to be in.

So but fortunately, I did manage to connect up with an old friend of mine who developed some interesting technology using ultraviolet illumination for imaging. And what he was doing was in vivo imaging. And the trick there was use the app, not just what the pathologists think of as UV, which is like 353 80, you know,  basically just below blue. This is so sunlight-deep, UVA to 60 or to 80 nanometers, which is so UVA ish that it  doesn’t even go through glass.

So you can’t use a standard microscope lens. You have to use quartz or.  But what it does is it did an amazing thing because it’s you may know, like traverses tissue, the depth in  which it goes is very, very dependent on the wavelengths. So, you know, it goes all the way through. You can see a red laser pointer through your finger. You put in one side and then you look at the other side and it’s what, green or blue laser? [00:25:00] You don’t see anything because it all gets absorbed right away.

And ultraviolet is even more so it gets absorbed within ten or 20 microns or you don’t. And here’s the trick that means you don’t need to make a thin section slide to get thin section histology because this U.V. light goes in ten microns, which is sort of like a thick slide. And then what? And then what’s really cool about when it goes in is that it excites autofluorescence.

And interesting enough, unlike sort of the fluorescence that we’re used to, when you look at a, you know, a fluorescein or a PSI three or a Alexa, this or that, where the excitation, let’s say is 680 and the emission is at 710, they’re within ten or 20 nanometers of each other.

The x the excitation at to 80 actually generates fluorescence way out in the visible, so you can extend it  to 80 and you can still get red, green or blue light coming back out, which means that you can collect the  light with a standard microscope lens and go and collect it with a color camera. So it is amazingly simple technology  to do what we now what the field calls slide free patholoy, you don’t have to make a slide and you get  histology. And lots of people are doing this many, many different approaches. Some of them are extremely  compelling.

However, often they use lasers, they use computers, they use optical shaping devices to to make  structured illumination. They use sound waves, all kinds of crazy things. And this just uses an LED, a UV  LED, the regular microscope lens and the color camera. So it just and in fact, it’s been implemented by a junior colleague of mine on a cell phone.

Aleks: So I’m like staring into desk. I’m like, yes, I want to talk about it. Because when I was researching  you, I’m like, okay, let’s talk about this, this main thing in your eye. But there are so many different things that you did. And I’m like, We need to talk about this as well, because it is something that everybody  thinks is like a [00:27:00] very for future to go glassless pathology. When is it going to be and my goodness.

And now you’re saying it’s on the cell phone. Tell me about that. And then I wanna you to tell me about  it!

Richard: Well, so we spun out a company from from our lab and by our I just want to give full credit to my  colleague Prasad Peruduni, who started as a postdoc in my lab and is now just coming up to the associate  professor, which is a nice evolution for for him. And so he he basically sort of built and all of these systems  that I’m I can talk about.

Aleks: And so that’s the company name?

Richard: Well, it’s now it also went through a few changes, but I think it’s now called Smart Health  Diagnostics.

Aleks: DX?

Richard: Smart Health DX. Have you heard of it?

Aleks: Yes, I haven’t read it.

Richard: So they they took over this technology and they’re moving ahead with it. But in the meantime,  my colleague Prasad came up with a different approach, which doesn’t use UV at all.

It uses just blue light, but it generates images immediately that look just like any thin section and that it’s a terrible name. But I called it FIBI – Fluorescence Imitating Brightfield Imaging.

So it’s a very not very good, not very sophisticated name FIBI, but I actually did it as a kind of run because  I had previously had a brief exposure to another kind of multiplexed imaging using lanthanide  labeled antibodies.

Aleks: I don’t know what that is.

Richard: Well…

Aleks: I know what antibodies are, but I don’t know what’s different.

Richard: There are two two groups that are doing that. Both companies and you can look at between, you know, 40 or even 100 different antibodies simultaneously because the the the signals can be separated using mass spec and so I worked on that actually that one of those technologies, a name which was Multiplexed Ion  Beam Imaging, which is MIBI. So then I did FIBI.

Aleks: I know MIBI

Richard: You know MIBI.

Aleks: Yes I know MIBI [00:29:00] I didn’t know what it stood for.

Richard: So I named MIBI, MIBI. And actually the other being I like about maybe I get to say how it works.  It’s you. You scan the the tissue with oxygen, dual plasma drone. That’s just the device that makes oxygen ions that are strong enough to blow up these lanthanides and send them off into the mass. That this gives me the chance to say oxygen, do a plasma shock. And I don’t do that anymore. But anyway, MIBI became  FIBI
We founded another company called Histolix, and we’re trying to get that up and running, actually a paper was published earlier this year, a validation study. We looked at 100 different cases and compare the FIBI images with Standard H and E and got a 97% concordance with pathologists had never been trained on  FIBI. So the image…

Aleks: evaluation…

Richard: To be just read directly by untrained or unprimed pathologists.

Aleks: FIBI that blue light

Richard: F.I. B. I. [00:30:00]

Aleks: Okay. And what was the previous one? The MU…electroviolet…

Richard: M.U.S.E. Microscopy UV Surface Excitation.

Aleks: Exactly.

Richard: And I call it is in honor of the co-inventor who was Stavros Dimos, who is from Greece.

Aleks: So Muse and Greece, because the muse is in Greece. My knowledge is okay, I see this like words are important. They are just important. There is a theme there.

Richard: Exactly.

Aleks: Okay. So Muse is now acquired by Smart Health DX and then FIBI, you’re starting with FIBI, and you  just finished the validation study,

Richard: Right?

Aleks: Okay. Is MUSE being used for any, like, diagnostic at the moment?

Richard: I don’t know if that’s being used, but it is certainly being I think, geared up to go through the FDA  and get out there. [00:31:00]

Aleks: And I know that the company is going to be advertising on into veterinary professionals and veterinarians as well.

Richard: Yes, they have a couple of things going on. Of course we do with with FIBI, because UC Davis has is home to the world’s number one vet school. I’m not mistaken.

Aleks: Yes.

Richard: And so we’ve got some collaborations going with with veterinarians. That’s people.

Aleks: Interesting.

Richard: And there’s some technical advantages to FIBI. For one thing, the images look like H&E directly strangely.

Aleks: Now, how do you make them like H&E?

Richard: They are like that way… So how does how does

Aleks: like you just but how they are that way?

Richard: Okay.

Aleks: Do you stain it with something?

Richard: Sorry, the the the thick specimens which can be fresh or fixed or frozen, but just not they don’t have to go into paraffin. They don’t have to be thin-cut, so you can just take a piece of tissue. Right. Right out of the…

Aleks: It’s a surface-like surface…

Richard: Surface. And then we stained for [00:32:00] 20 seconds with hematoxylin at 20 seconds eosin,  and…

Aleks: That sounds familiar, its H&E. It’s it’s a need that image different lick.

Richard: That’s right. And then the magic is it’s bizarre. So and we it’s like with four or five nanometer blue light from an LED okay so you think…

Aleks: So a step back how do you make MUSE images to look like H&E?

Richard: You have to, they’re fluorescent so and the color

Aleks: You have to translate it into…

Richard: Or you it’s it’s easy you can you can invert them

Aleks: You can basically like translate the colors into different things.

Richard: Exactly.

Aleks: Map the colors into different colors…

Richard: Right, they can look very, very H&E. And a lot of the other techniques, you know, generate two grayscale images and they just take one image and make it blue in the other image and make it pink and squish

them together. And it looks like H&E. The advantage to MUSE into FIBI over some of the other techniques or an advantage is that it uses a color camera, which means that you get lots of colors [00:33:00]}], not just  a  blue channel and a red channel. And as you know, pathology has more. There’s more to life than red and blue.

Aleks: There is green.

Richard: This reveals things about the tissue that you don’t see, and other things. Aleks: Amazing! So, okay. FIBI is H&E without glass,

Richard: Right.

Aleks: With and normal microscope camera with a microscope lens.

Richard: Yup.

Aleks: And what about the…

Richard: And the blue light excitation And then why does it look like Brightfield

Aleks: Yeah, why?

Richard: Because the excitation light, the four or five-nanometer excitation light goes into the tissue  past the surface down into the bulk tissue because don’t forget these are thick slices, quite thick,

I mean 100 microns and on, you know, a 10th of a millimeter. So not thick in any real sense, but thick certainly in terms versus a thin slide. But we typically look at one or two millimeter, three-millimeter sections that you can easily generate just with a razor blades.

What happens is the light goes into the tissue and creates tissue auto [00:34:00] fluorescence, which just becomes this sort of cloud of whitish light in the tissue, half of which comes back to through the surface, through the lens, into the camera. And when it goes through the surface, it gets absorbed by the hematoxylin-eosin stains, which is exactly what happens on your regular microscope. Right.

But from the other side, from…

Richard: Well no, I mean…

Aleks: Like this one goes from the inside and illuminates. And in the…

Richard: Same idea. I mean, so basically your your microscope light is from below the surface. And FIBI, the autofluorescence is generated below the surface and then it comes back up.

Aleks: Okay. Yeah, It’s like a microscope light underneath the surface that you’re trying to image.  Richard: Exactly.

Aleks: Oh my goodness. So, Richard, we’re like, is this now? Are we going to get rid of glass within the next year? Like, where do you see it coming and what are the disadvantages of this? Like, why is it not yet everywhere?

Richard: Well,

Aleks: Cause you patent it, and you don’t want to…

Richard: I know it’s first of all, MUSE came to life in a happier investment time as you know with some investments these days are much harder to come by people people don’t just throw tens of billions of dollars at

you know, the way they used to. So it’s been a little bit

Aleks: Too bad

Richard: Hard in thr fundraising world. And secondly, I think it’s it’s a matter of getting the work done,  getting the word out and and then demonstrating it as not just as pretty pictures, but actually functional,  useful and the utility of this kind of work. And we will get to AI in a second is that you can get you can  speed things up so intraoperative you can get rid of process.

Aleks: This is like basically you’re taking pathology to where radiology is, where you don’t need the analog  part of it.

Richard: Exactly. And so it goes right to digital, direct to digital from the from the tissue, you know, no slide, no film. Right. [00:36:00] Same thing.

Aleks: You’re basically a radiologist pathology with this.

Richard: Right. And in fact, we can interact with radiology because a lot of interventional radiology sort of culminates in a needle biopsy and then they don’t know what’s in the biopsy. Right! So we can you just

Aleks: Phoebe it and you see whats in the biopsy.

Richard: Exactly right here. And you know, if if you hit tumor whether you have adequate tumor for your molecular and maybe even do it definitive diagnosis just all at once.

Aleks: My goodness. People who are watching this on YouTube are going to see me like with very wide eyes.

Richard: Yes.

Aleks: This is amazing. So let’s how many how much more time do you think till it hits the market?  Richard: If only if only you know…

Aleks: Like in the best case scenario and in the worst case scenario, one scenario…

Richard: Is that something will certainly hit the market. I mean, slide-free microscopy, there are five or six or seven people with different technologies all vying to do this. It will happen. Who’s going to win?

It’s really an execution story. You know, basically, you know, [00:37:00] VHS, beat Apple, Betamax, right?  Or it was better execution, poor technology, but they won. So we’ll see. But I would say within three years  it will be

frozen section robots, which is pretty

Aleks: okay.

Richard: And before that of the researcher’s hands. And depending on what the FDA does, it could be a  LDT. Right. If you just validated house. We don’t wait for the whole FDA approval story you validated that  house and then you can use it internally and then veterinary you know you bring in a racehorse for you know 50  miles away that’s not easy. And then you do a biopsy and then you send the horse away and they have to  come back a week later because…

Aleks: Yeah, For another biopsy.

Richard: Yeah. So now you can do it all with the horses there.

Aleks: Amazing. Oh my goodness. And you also touch on that on something that is like not so obvious because a person can come back for whatever procedure immediately. They usually go to. Well, they,  they travel as  well, but the horses are difficult to transport.

Richard: Yes, [00:38:00]

Aleks: Amazing!

Richard: And what’s really, really cool is the the NIH seems to like what we’re doing and we’ve received…  Aleks: That’s a good sign.

Richard: Bunch of major grants recently, one just went through all of the paperwork. This has been funded to develop FIBI for corneal biopsy, breast cancer and develop and test in the country of Ghana in Africa.

Aleks: Oh my goodness!

Richard: Five years of work…

Aleks: Why Ghana out of all the countries in the world?

Richard: A good question. Well, I had reached out to a colleague of mine at ASCP, Dan Milner, I don’t know if you’ve come across him, and he’s been very active in in global health stuff. And I said, you know, “I’ve

got this really technology, where should we go?” And he said, “I have this great person in Ghana, Beatrice  WIAFE ADDAI…

Aleks: Connections is everything.

Richard: So just lucky. She’s the CEO and she’s a breast cancer surgeon and founder and CEO of a hospital in Ghana and a [00:39:00] very prominent person with lots of, you know, all of the right influence and this work. And then

I, I got into trouble recently because I was asked to talk.

Aleks: That’s a theme as well.

Richard: I know I get into trouble all the time. So I got into trouble because I was had to give a talk on the in the funding agency. You know, a lot of people who just received the said that describe what they were doing. And I saw I was talking about this work in Ghana and I said, you know, I discovered I was sort of put in touch with people in Ghana by my colleague Dan Milner.

Aleks: I went and I want to also know the part with the smartphone.

Richard: Well, that’s that hasn’t been pursued. It was it was so my colleague, he knew at Cleveland’s Case  Western Reserve basically just figured out how to how to hook up a little LED, this was a UV LED and on  an apple phone and a nice microscope lens, which is basically the same lens that’s in the phone. You just  [00:40:00] flip it, invert it, and it becomes a micros… Microscopy system. And he was able to power and then use the the course, the camera and the phone and he was a power everything with the iPhone battery so  there were no cables at all and you could do really wonderful histology just by touching the phone to the surface of of the tissue and it was published.

Aleks: You need to find this publication.

Richard: I think I’m on it somewhere. So if you look up me on google.

Aleks: Yeah. I’m going to Google scholar you and…

Richard: Yeah, but if the images are gorgeous and it’s from a cell phone with and the extra costs turn a  cell phone into a microscope was like under $100.

Aleks: Wow. It’s just like rebuilding the smartphone or having, like an attachment.

Richard: It’s a little actually very thin. I mean, he did a great job. It’s just a machine, a plastic machine part that just fits over where the camera and the lens are. Right now. It’s about, I don’t know, two or three millimeters thick and and it’s freestanding. In other words, it plugs into the phone somehow, but it doesn’t [00:41:00] need any other wires.

Aleks: And is it being used anyway?

Richard: I don’t think so. It’s being used.

Aleks: Why?

Richard: I mean, it’s I have no idea why its not being used.

Aleks: Okay. But I will. I will invest in that topic.

Richard: And yeah, absolutely should be. And the same thing we can do with FIBI, we haven’t done it. It would be a gorgeous grant. It would be wonderful for school kids. Right. Imagine having this and being able to do microscopy at in the fifth grade. Imagine how that would open up people’s eyes.  Aleks: Yeah, but also like everywhere.

Richard: Everywhere. Exactly. You’re out in the forest. I mean, you can go anywhere.  Aleks: Amazing! Maybe. Can I buy it?

Richard: We can take this offline and we can see if it can be revived. And somebody needs… Aleks: Yeah, I’m going to. I’m going to find the paper.

Richard: Yeah.

Aleks: Check this out. Okay. So glassless pathology, pathology that actually is now on par with radiology.  Regarding the technology, we don’t have that. We basically have the potential to very soon eliminate the complaint, it’s analog. And then you have to [00:42:00] add digital pathology and AI on top of it and it’s overhead and that’s everything because we still do analog. Now there is a potential not to do analog.

Richard: Its direct to digital.

Aleks: Direct to digital. So exactly like radiology. So this is where your encounter with A.I. happen. Or tell me your story with A.I. and your relationship with A.I. in pathology.

Richard: My career is very, very strange. I’ve had one graduate student in my entire life  Aleks: As a full professor.

Richard: As a full, as well, I mean, don’t forget, I was sort of ten years, 13 years in industry, right? For 15  years in industry. And then I was just an autopsy pathologist way back when, you know.

So I didn’t really have any standing to have a graduate student. But I joined the biomedical engineering graduate group when I was at Davis, and a student reached out to me to join my lab and do his Ph.D. with me, which is remarkable. And he actually did it with four years through through the COVID thing. I never saw him. He just all he did all his work from home, and [00:43:00] he was doing AI…

Aleks: P.H.D. from home. I love it!

Richard: Yeah, it was right. And he got his Ph.D. just now. He’s nationally or internationally famous. Why?  He got his Ph.D. at the age of 19.

Aleks: I think I saw news something news and news, a press release or something about it. How we how,  how? No, me, how? Like so 19. He started at 15. You’re not even done with high school…

Richard: You know…

Aleks: Tell me the story

Richard: He’s way outclassed his sister who is doing a mas… not a Ph.D. but right now with just a master’s degree and she’s already 17.

Aleks: Oh no I think failure, failure, it’s mind blowing right?

Richard: At 17 You know, she’s she’s she’s an absolutely professional level singer. And in fact, during the graduation ceremonies for all of the graduate students at which my students got his received his Ph.D.,  she was the one who sang the national anthem for this.

Aleks: Oh my goodness, [00:44:00] this is like some movie or…

Richard: Exactly, were still working together, we just had a paper accepted in Optica.

Aleks: Okay, there is a good many good collaborations between the and the optics and like the light vision scientists and pathology is the super cool intersection. Actually, the first company I worked for and where I encountered digital pathology, one of the founders had the Nobel Prize in. I would have to Google would exactly in, but it was something with microscopes and optics. Let me Google him.

Richard: So he and I were working with another imaging technology, a sustained free imaging.  Aleks: Stain free, so not only glasses free now this its stain free.

Richard: This is glass-free and stain-free and high resolution, but it’s grayscale. So my I didn’t say what my  my graduate student did. He was working on A.I. tools to convert one image into another image type. And so he was taking those MUSE into FIBI images [00:45:00] and making them look identical to H&E. And then  we did the same thing for this other technology, which is called it’s another phrase I wanted to QOBM,  Qualitative Oblique

Back Illumination Microscopy, which sounds frightening, but is…

Aleks: It does. It’s my sounds like this other word that you like to use, oxygen something.

Richard: But the oxygen duoplasmatron on thing, that’s a 500,000 million dollar box. This is another one that fits into my my theme of appropriate technology. It also requires just a couple of LEDs, a microscope lens and a camera. That’s all. And it does.

Aleks: Okay.

Richard: High resolution, super fast imaging of unstained tissue. And this is not only just a surface, but also a little bit depth through is all really fast, really cheap. And then what my graduate did and I with this other group, I didn’t develop the technology at all, but we developed the AI tools to convert these grayscale images into things that look just like H&E. And so that was what was published.

Aleks: And so what are you going [00:46:00] to do with all those different microscopy technologies? Is there going to be one that’s going to win it? or…

Richard: No. While they’re all good for different things. So this QOBM thing, I don’t like FIBI for invivo  because it also stains.

Aleks: Ok, this is true.

Richard: It can be handled and one of the first potential uses is, for example, looking at brain tumor,  intraoperative brain margins. And so that’s what part of the paper was. We showed that you could tell the difference between the normal brain using to QOBM and convert it to H&E.

Aleks: So this was your first encounter with the AI?

Richard: No…

Aleks: Like, hands on…

Richard: I go way back.

Aleks: Tell me about your history on your relationship with AI in pathology I want to know about.

Richard: Well, back to this company that I work for, Cambridge Research and Instrumentation, and we  developed this multispectral imaging and then one of the people who I’m still pals with develop neural  net tools [00:47:00]  for segmenting pathology images. So it wasn’t deep convolutional neural nets, but it was like a three layer  neural net. And this is back in the early to mid 2000s, right? And it it developed into a softer call Inform  which remark.

Aleks: Yes, that’s Akoya.

Richard: It’s Akoya. It’s still being, still being sold.

Aleks: And its being use for IF.

Richard: Yeah. So this is way back when and actually I have a couple of patents related to that, none of  which got me any money. All of it went to the company, unfortunately. But so I think I’m one of the  pioneers in  A.I. tools for pathology, amazingly enough. You know, I just happened to be at the right place at the right  time.

Aleks: Fantastic. Okay, so that was what which year?

Richard: I dont know, mid 2000.

Aleks: Okay.

Richard: And then that’s still there. And actually, I really like it. You know, it’s very low overhead, very fast.  It’s trained by example, much faster than convolutional neural nets, at least to sort of work [00:48:00] through needs less training…

Aleks: Use it, it only has like three layers not that less (than CNN.)

Richard: It’s not necessarily as robust, but if you have a simple problem or like given a state a single image  and you just want to find all the nuclei in that image, you don’t need hundreds and hundreds of examples because, you know, it’s a very defined problems that…

Aleks: I managed to find what Gerd Binnig got his Nobel Prize. And it was 1986 and it was the scanning,  tunneling, tunneling microscope, scanning, tunneling microscope. And he shared it with Ernst Ruska because Ernst won the other half of the prize. So Gerd actually won half of the prize. I didn’t know you  could win half of a prize.

Richard: Well, I think you only have you can have a group, but I think you can only split it among three.

Aleks: Amazing. Okay, so now let’s go to the paper A.I. in pathology. What can possibly go wrong? And I’m  going to start sharing this here again, because there is there are some visuals in the papers that we  [00:49:00] want to address. But it was recent. This was published in 2023, which is this year.

Richard: Which is recent.

Aleks: Very recent. But yeah, tell me, why did you decide to write?

Richard: Well, actually, I and K.Nakagawa and I have been talking about writing a paper like this for years.  Five years now. And I kept saying, blame it all on him because I really get a chance to blame somebody  else.

He said, I’m. I’m busy.

Aleks: He is also the first author of the paper and your the last to reason…

Richard: Exactly, I’ll get to it this week. And of course this week went by and then this month with AI. And  then it just never happened. But we had we’d have like a page or two written, you know, five years.

I honestly don’t know how this happens, but I think he and I talked again about sort of resurrecting it in  the modern era where it would suddenly resonate with a lot of people.

Aleks: And now it’s like the topic of every discussion, like with large language models. [00:50:00] I don’t  know if you’re an address to the large language models here. Yes. And of course, because you generate  that a  problem with a large like.

Richard: Yes. In fact, the first thing we have to look first, two paragraphs of the paper were written by  Chap GPT and the final poem was written by ChatGpt.

Aleks: Yeah. Do you want to read the poem, Richard? Do you want read the poem?

Richard: I want to claim, I think, a priority and I don’t know how many pathology journal articles have  poems in them.

Aleks: I have not read any other one with.

Richard: I know. So I think I’m pioneering a whole other genre. I think we should have, you know, limericks  and things.

Aleks: I don’t know how many enthusiast you will find for that. Maybe all those with similar background  through yours with the English major. English literature background.

Richard: Well, shall I read the shall I read the poem that’s on the. So go ahead and read the poem. Okay.  So I asked ChatGPT [00:51:00] basically I just gave it the title of the paper and said Please write a poem in  the style of Edgar Allan Poe. For whatever reason.

Aleks: Yeah, what reason? Why did you not want us to.

Richard: We couldn’t think of a you know, I don’t like Emily Dickinson as much as other people do. Aleks: Okay

Richard: But so I just said that and it really came up with this right away. It wasn’t as if it generated, you know, 400 and I had to pick the best. This was the one that came up with. And I just go with this.

And that’s the the amazing power.

Aleks: And which chatgpt, did you use the 3.5 or.

Richard: I don’t remember all of the details. And there were some issues about setting some settings to I  forget I mean, I have I have it written down. But basically you’ve…

Aleks: Probably in the materials and methods of this paper, it’s because now you’re required to say, like what you did with ChatGPT [00:52:00] and how much and which model and everything.

Richard: Yeah, I know all that stuff. I forget there are some, some some knobs you could tweak and I forgot everything about how they were tweaked, but it was very straightforward. You know, I’m not an expert in large language models. I just was a user.
But anyway, so this was a whole paper on. Well, all the dangers of AI that might be discovered in as it gets developed for pathology and they are substantial and people  you know this is by now well discussed and to some degree well understood. So anyway I asked AI, the chat GPT what it about this whole thing and to write a poem about it and so this is what it says on the screen it’s  it’s up there but I’ll read it in case you’re just listening to it.

Aleks: Yes. Yes.

Richard: “The machines, they may seem infallible,

But they’re errors. They can be terrible.

A missed cancer, a false positive too

The consequences, [00:53:00] dire and askew.

But yet we still march on with this quest

For faster results with little rest.

But let us not forget the human touch,

For in the end it is what matters much.

So let us tread with caution and care

As we delve deeper into this AI affair.

For the stakes are high and the risks real

In the field of pathology, let us not seal

Our fate with machines, but with our own eyes

And judgment and expertise that never dies.

Aleks: Oh my goodness. Its deep, wrote by ChatGPT, I know, totally captures the spirit because,  you know, these are powerful tools. And both on the imaging side and on the large language model side.

And now there is a combination like on the, I don’t know, popular side. And there is this ChatGPT vision where you can interact with images, you can generate the image, which we’re going to be talking about,  one generated image in this paper as well by generative networks. But you can also have an image  [00:54:00] and have the language model extract the content of the image and describe it. And there have been attempts already to do it with pathology images. And the results weren’t like crazy all. I would say,

Richard: And it’s sort of an oxymoron to have GPT three tell us about the risks of it, but do it in such a  beautiful manner, right?

Aleks: It’s it’s amazing. I mean, this has not been available. I would not have imagined like that. You will have a tool that you can tell them what you want to have when the presentation on the PowerPoint presentation is going to make of this slide. And this is basically and maybe not entirely, but I did it for the last conference, I was presenting it. It’s like mind-blowing.

Richard: It is.

Aleks: But getting back to the dangerous, let’s talk about the image in the in the face so that we don’t actually.

Richard: Let’s go to the first image.

Aleks: Yeah, the first yes, let’s go to the first and the second actually the second image. And everybody who’s watching this [00:55:00] on YouTube can look at that on YouTube. We’re showing it right now.

And I’m going to also link to all those papers and the both of the papers in the show notes. But yeah, figure  1, I’m going to read the description “Image generated by DALL-E mini based on prompt: “Pathologist

confronted by AI. Richard, can you tell me about this artwork.

Richard: It just did it literally that’s what popped up. And I haven’t seen a better visual representation of exactly the topic that we’re there were discussing. There’s a pathologist clearly identifiable and then some A.I. one sort of humanoid, the other not, but with a glowing red thing. And it’s supposed brain. And  it’s just a very gentle way of capturing this whole storyline where and there, you know, it is a  confrontation,

but it’s not, you know…

Aleks: Not aggressive confrontation.

Richard: You know, it’s just kind of what’s going to happen next, guys sort of thing.

Aleks: It’s the style of like the brains [00:56:00] representing A.I. or like brains slash neural networks.  Richard: Right. Exactly!

Aleks: Presentation of the AI.

Richard: But I do like just stylistic. And again, this is amazing, that sort of orange, you know, glowing disk which was thrown in there by the artist DALL-E Mini. And it, you know, it just dramatically improves the content of the of the image from a style point of view.

Aleks: Interesting. So let’s go to the second image. The second image actually references the pigeon paper.  Richard: It it did indeed.

Aleks: Right,

Richard: DALL-E Mini.

Aleks: Let me read the description. “Image generated by DALL-E Mini based on the prompt: “Pathologist training pigeons (Style of Rubens)”. Looks promising except for one thing.

Richard, tell us what this one thing is.

Richard: Yes, I didn’t ask for it, so. Yeah. Pathologist pigeons. Perfect. I’m happy, AI is going to rule the world. But then if you look down at the bottom right, there is a strange [00:57:00] creature that it’s impossible really to describe. It looks a little like a combination of cat and rabbit that well, who knows  how many it has.

Aleks: It has like, eye-like structures on tentacles.

Richard: I know, who knows what it is.

Aleks: Like a road, face or…

Richard: I didn’t ask or but I mean, if you look at the prompt, I didn’t ask for it. Throw in something horrible.  I just said pathologist training pigeons and…

Aleks: Maybe it was supposed to be a pigeon that didn’t work out. But I know pigeons are so realistic.  Obviously, it’s kind of well known that the AI has problems with with details like the fingers always look a  little bit dysmorphic that the pathologist has I assumed these are glasses, but they’re a little bit like thick..  Richard: Well…

Aleks: like I mean a jar.

Richard: It’s not fair this is, you know, a year and a half ago.

Aleks: That’s true.

Richard: By a sort of minor player in the field. Nowadays, that generative art is just astonishingly good and it would not have that little funny thing at the bottom [00:58:00] right? That just wouldn’t be there.

Aleks: And it also has these colors like a little color scale is this…

Richard: I don’t know about the colors, but he’s he’s wearing a nice dirty white coat.

Aleks: It is it is because of the style of Rubens, I guess. Yeah.

Richard: Or the style of pathology lab.

Aleks: Anyways, that I’m gonna do. I’m going to try to include those images and but yeah, those images represent what can go wrong. And I love what you have in this paper as well. It’s a table basically of what all can go wrong. So if you know…

Richard: One of my one of my co-authors is one of the leading scientists in the world, Faisal Mahmood,  and he came up with a nice list of this.

Aleks: It’s a great guiding table, I would say, because you have a you have a column with challenges. You have the column of impact of the AI and mitigation strategies, which is basically what’s needed, [00:59:00]  because I  don’t think there is a way to avoid it. And it’s some point at the moment where it’s mature enough and it’s going to give us enough advantage in providing care. It’s going to be unethical to not use it.

Richard: Right.

Aleks: But it doesn’t mean that with the maturation of this technology dangers are going to disappear and I think, like there is no way to have two camps over one of those who are using it and the other ones who are not using it. There are against it because it can make so many mistakes.

It’s going to be it’s going to it is going in the direction of, okay, what are the mitigation strategies, What risk assessment points can we have along the process and how can we like when can we say we did the best job we could have done?

Richard: Well, I think, again, I hate to say this, but read paper, [01:00:00] there are some serious issues down the road which has to do with the finances of of these air tools. They’re very expensive to develop.

And the question is, you know, how do you do You lock them down, and not improve them. If you improve them and spend another $10 million on a bigger data set, who pays for that? The company. And how do you charge for  it and etc., etc.. So there there are, you know, just getting beyond kind of the theoretical questions of…  Aleks: Sorry, getting into the like reimbursement side of of health care.

Richard: Well, there’s that. You know, and the challenge here is that the problem is that pathologists are only already doing, quote, the gold standard level of care. So how do you add costs to to generate just at best, gold standard of care?

Aleks: Well, if you take away the glass, then you kinda took away…

Richard: Now we’re talking [01:01:00]

Aleks: And it kind of like offsets the cost of developing everything on top of analogs. But you know what?  This is a question basically, how do we implement digital pathology regardless of air? Because know and people are the implemented. There is no there are codes. There’s no reimbursement yet. And I it’s better I mean,  that’s better for me to do my job as a vendor.

Richard: Yeah I you know, it’s it’s a problem. I mean things that are objectively better still have trouble making it in the market. And then you have some, you know, and then you have downstream issues like where do you store all these digital images? How do you pay for the storage? How do you what when do you get rid of them? You know, it’s sort of it it kind of rhymes with the problem of of electric vehicles, you know,  perfect

Tesla works…

Aleks: very parallel.

Richard: Yeah. Looking at some of the infrastructure issues and downstream repair issues, all kinds of things that really stand in the way. And you can’t be naive about them. [01:02:00]

Aleks: That is a very good point. And they want to emphasize you can’t be naive and like I think pathologists as a profession, they kind of like see beyond the marketing message, marketing message and messaging of any tool. But I would want to emphasize that we cannot be naive because obviously everybody is going to like off who is going to be offering this.

And it’s the same with electric vehicles like narrative is they’re better for the environment, like in which aspects are their better, what are the other aspects? And I think for any technology, it’s not that, you know, it’s for any technology. I what I’m seeing is, okay, if there is enough noise generated that takes the narrative in a certain way, we tend to go with it.

But basically it’s let’s put it that way. It’s human nature to like want to trust something. And if there is enough messaging around something, then at [01:03:00] some point you assume, okay, and this is what it is, I start believing it right In health care, we cannot allow for this to happen regardless, like how much messaging about anything there is.

And I think we are in a kind of better position to not be naive about it, but I think we still need to pay extra attention, especially with tools that are like so much better than anything else. Like up to now,  everything was like incremental. You could compare it to what was immediately before and now with generative AI and with the glassless imaging, it’s like leaps forward and that that’s my like point.

Okay, If it’s so much better, is it then ethical not to use it, but is it ethical to use it if there are dangers? So let’s talk about that a little bit. And I don’t think there is a good answer…

Richard: About boring AI. Yeah, let’s look at boring AI.. [01:04:00] It’s things like can you count the nuclei?  Can you…

Aleks: Image analysis? Yeah, like the classical applications that we had with and you know, scoring of…

Richard: And you can quality control on your slides. Can you prioritize the, the, your workflow, Can you order stains automatically, Can you do all of these workflow issues? And I don’t think anybody as long as the sort of the pricing was right on it helped with productivity. I don’t think anyone could have a problem with that.

Aleks: See, I wouldn’t have a problem with that. Most of the people wouldn’t. Where is why is it not implemented everywhere? That’s the like boring and low-risk A.I..

Richard: Well, it does still require an infrastructure. In other words,

Aleks: Thats true.

Richard: You need to be. Yeah, yeah. A digital shop.

Aleks: Okay.

Richard: Right.

Aleks: And I would argue again, I don’t think it’s even in every digital shop. Let’s let’s be optimistic and say  10% is digital, which I’m just taking out of my head. I don’t think it’s 10%, but let’s say [01:05:00] 10% is  and

I don’t think even 10% of 10% will have those things implemented at the moment.

Richard: Well, and then there’s just to go down that road, there’s a sort of nice little tributary to all this,  which is basically on your microscope digital. So you just have a camera that’s sitting up there and you’re removing the slide around and it’s doing everything you want it to do, right? It’s it’s looking at the images. It’s it’s counting mitosis. It’s stitching together a large field of view for you automatically.

You don’t think about it. You don’t change anything about what you do and don’t do. By 200,000 on the scanner, you don’t even know.

Aleks: You know what my talk was at ACVP conference? One of the talks, Static Telepathology and thing of the past or a new trend on the rise.

Richard: Yeah so you know there lots of flavors to this and I’m not sure what’s what’s going to win.

Aleks: Yeah, I don’t think we can answer this question. Maybe we should meet next year and have a follow-up conversation.[01:06:00] And that’s the thing you know, you have there’s fear during the enthusiast who cannot be naive and they’re not naive, so they kind of like want to promote, but all the concerns are legitimate. So yeah, I guess time will tell and when there’s going to be enough proof that’s going to tell the science,the applications. Because we will just see which way the world goes with this.

Richard: Let me let me hit to air dangerous one is bad training sets. And I think a sort of classic example of that is breast cancer. I forget which class, but there seems to be a different biology of breast cancer for West Africa than for the rest of the world. And the AI tools that are trained in Pittsburgh don’t work in, let’s say, Ghana. And, you know, how do you monitor that? How do you document that? How do you fix that?

Aleks: This is [01:07];00] just how they do it. You just you provide the different data sets, right?

Richard: You pay for it.

Aleks: Yeah, how they pay for it. That’s always like because there are so many fantastic initiatives and all the questions that they ask, like, where is it? Where is the smartphone microscopy, where is the low-hanging fruit? It’s like probably nobody backed it up with enough money to implement it, right? So that there is the answer to my question. But this is also super interesting.

This is so recent because I just came from this conference in my colleague of mine. She works for a large pharmaceutical company and she was asking the same question in the context of diversity. Right. And diversity, when you talk about that, it’s more like let, let’s say, in the socio-economic context, but context and like to be fair and just, but it totally translates into their scientific context and the functionality of the tools like and example.

So I was preparing something for another webinar and I gave the prompts to DALL-E, a pathologist at the microscope out of  [01:08:00] four images, obviously not obviously, but obviously sadly, obviously four of them were white male,  right? So yeah, it’s all the biases that we have are being translated into the AI development sets. The thing is, okay, some of the biases we know about and many we don’t know, I think biologically or like, yeah, biologically and through our experience, we are biased creatures and to like under this bias we have to be conscious that we are biased creatures and like, find the biases. Can you find all the biases? No. Right.

Richard: They’re smarter at generating biases than we are and finding them famous classic case of of some imaging challenge. And they looked at a bunch of patients from different hospitals and it turns out that the patients rooms in one hospital had a blue stripe behind the patient and but it was also a different patient population. And so the AI [01:09:00] immediately clued in to the blue stripe that was there.

The differentiating.

Aleks: Very much. There was no story like that, like with a red dot on that malignant or something. Yeah,  well, another one that I heard from Anant Madabhushi who was a guest on my podcast, he was differentiating the Wolves from Husky and that was a fantastic performance. And it turned out there was snow behind the  Huskies always that us with no answer.

Yeah. So, so what like if you were to express your opinion on AI,  it’s your opinion. I mean that’s part of your research. On the other hand, you are the author of the paper  What Can Go Wrong? And we kind of get dead ends in our in the whole discussion. And then parts of the discussion we like, which there are no answers to the questions. Do you have like an opinion?

Well, you if you wanted to to give an opinion, a short opinion on this, and I’m [01:10:00] going to give you my mind is we need to be informed of how these things work. And then that’s the best we can do at the moment. But I don’t have anything deeper than that, unfortunately.

Richard: My analogy for air and already based on something I just read, was on automatic bartending robots and what this does to servers serving drinks in a bar and you think, they’re not enough servers,  making them bartenders will just do this. And and that means that they can sort of be more efficient get the drinks faster to the customers, get higher tips. The answer is no, it’s the opposite. Why?

Sometimes that the customer wants a drink with you know, this bourbon or that bourbon or with or without the lemon or whatever it is, and the bartender robot can’t do it. So this poor waitress or waiter has to go running down to find a human bartender to put the drink together, who he may be, he or she may [01:11:00] be way overwhelmed.  Meantime, the customer gets tired of waiting and leaves so there goes a whole the whole small sip is  gone.

And she and  the and waiter has been running around like crazy trying to make it happen. And, you know, so you got all these unintended consequences of something that works 90% of the time, but that fails miserably  10% of the time. What is what happens to them? And I think the short answer is I haven’t the faintest idea. We will we will watch this space, I think is the best I can do at the moment.

Aleks: Yeah, because what’s the space? I mean, you have to watch pretty closely at the high cadence as well, because there’s going to be like every week there’s going to be some revolutionary new perspective or new paper, new technology. To me it’s like, okay, if something doesn’t work at the moment, don’t make it work. Wait for the next thing to come out. But yeah, a fantastic discussion.

Richard: Thank you. It really was.

Aleks: Thank you so much for joining me. [01:12:00] I’m wondering what what the listeners are going to say about that after this episode. And I love that it meandered like that and that basically it takes it takes out of the at the simplistic binary nature of things like, I don’t know, this was something when they came to the U.S. it’s just like, okay, it’s pretty binary because there are only like two parties and you either are four something or against like kind of noticed that it was more than in Europe. I don’t think anything is binary.

Richard: It will be fun to watch.

Aleks: It would be fun to watch. Thank you so much for joining me.

Richard: All right. Thank you!

Aleks: And I hope you have a wonderful day.

Richard: And you, too, as well. It was real good. Thank you.

Richard: Bye bye.

Aleks: Thank you so much for staying through the end.

Aleks: My favorite thing about this episode was understanding how close we are to glass-less pathology.  And I would love to understand this topic more. So I’m actually [01:13:00] thinking of a more in-depth,  serious,  or a virtual event or a mini-conference about it. So if you are interested in taking part in this, drop me a  comment below.

Writing GlassLess Pathology and I will know that you are interested and I’m going to  start organizing this event. I’m looking to hearing from you in the comments and I talk to you in the next episode.