Read the Full Video Transcript

Alicia Morgans: Hi. I'm so excited to be here today, talking with Dr. Adam Dicker at GU ASCO 2025. He's joining me from Thomas Jefferson University. Thank you so much for being here.

Adam P. Dicker: My pleasure.

Alicia Morgans: Wonderful. So there's been a lot of excitement about MMAI, this artificial intelligence that helps us really understand pathology specimens on a deeper level, maybe being prognostic or predictive of certain outcomes for patients. And you're trying to help us understand, in an abstract that you presented, really how we should put the specimens together. Are there differences in different specimens-- TMA versus regular specimen-- in the way that we should understand and interpret our MMAI? And you also did a really interesting analysis trying to understand race in this as well. So I wonder if you could share a little bit about what you presented.

Adam P. Dicker: So the story goes back only about 30 years, where we had a pathologist, Roger Peterson, whose standard was, in pathology, to do whole-mount sections. So a regular microscope slide is like 3 inches by 1 inch. When you whole-mount it, you're kind of cutting like a loaf of bread. And this way, you can look at cancer, noncancer. You really can get a whole picture. We got interested in this because we were interested, through a generous grant from the Prostate Cancer Foundation, about the immune microenvironment.

And in particular, we had 2,000 patients' worth of these whole-mount specimens. And as you know, there's a great literature about, for reasons that I don't think we really understand, why is it that African-American patients don't do as well, whether they're diagnosed later or it's the biology, social determinants of health, we don't really understand. So we thought, wow, what a great opportunity to use AI to study whole-mount specimens. Theoretically, you have so much more information than a regular biopsy or regular little piece of a radical prostatectomy.

And we also made a tissue microarray. We identified about 100 patients, about 80 African-American, 20 Caucasian, locally advanced disease. We worked with a pathologist to map out, meaning they drew by hand the areas of the Gleason score, et cetera. And then we created, before we scanned them, a tissue microarray. Because one of the issues in this whole AI business is, so what is it seeing? Right? What's the biology behind the AI? What's reflected?

And the tissue microarray allows you to then study, using spatial biology and other transcriptomic approaches, what might be reflected in AI. So that was going into this. So we said, hey, there's this FDA-approved AI thing called Artera MMAI, or Artera Prostate AI. And why not look at the whole mount, and why not look at the tissue microarrays, and see if there's a difference and see if it can show differences vis-a-vis self-identified race?

So it didn't work out so well, meaning in a perfect world, the score that the Artera MMAI gives would be identical on this whole-mount specimen, and it would be identical on the tissue microarray. Unfortunately, there was a lot of discrepancy. So it was a fascinating project. It raised more questions now than we have the answers to. We didn't see any differences in terms of race between how the scores were for the MMAI versus the microarray, et cetera.

But it brings up questions about sampling error. The MMAI from Artera was trained on regular microscope slides, not all this information. Maybe the classifier kind of peters out or something. What is ground truth? It's uncovered a whole number of issues that we're trying to sort out. They're a little technical in the AI space, but I think that's a good thing, and we'll learn from it.

Alicia Morgans: Well, it's interesting that you say, unfortunately, it didn't work out, because I think that this question, whether you get the answer that they're equivalent or not, is actually just an important question to answer. And I wouldn't say it's unfortunate. You found a difference between these two approaches that now you can investigate to understand in terms of the clinical application of this MMAI and what we should do as clinicians, and as we send those samples over and as we're trying to understand how to interpret the results.

So I think it's really interesting that you found this discrepancy because now we can try to understand, does it affect more than that difference between the tissue microarray versus the whole-mount specimen? Is it something we should consider as we're sending our samples to Artera? And how do we use this in clinical practice? Does it affect anything you do in your practice, or still just baby steps to ask more questions?

Adam P. Dicker: Yeah. So your viewers should know that whole mount is not the standard of care in pathology labs. It consumes enormous resources, and it's not clear that it yields more data.

Alicia Morgans: Yes.

Adam P. Dicker: Remember, we got interested in this from an immunology standpoint because, as you know, in the world of prostate cancer, the checkpoint inhibitors that have had success in renal cell cancer and bladder cancer and a number of other malignancies haven't had the same level of success. So we are trying to unravel that. Then the Artera opportunity was just a nice opportunity to look at whole mount versus tissue microarrays.

So we've learned a lot on the AI part. So for example, there are recent papers about, how do you create a digital whole mount from pieces of prostatectomy samples. So there's recent papers on this. So there's a very active area of interest within the AI community in this space. We're still quite interested in the tissue microarray because if you can work that out, and if it turns out that the MMAI technology works in that space, it opens up a lot of opportunities. So it's early days still, but I appreciate your cautious optimism.

Alicia Morgans: Well, I think any time you answer a question, we can be optimistic that we've at least checked the box on one. Now we may have opened the box on five more questions or 10, but at least we've answered one. And this answer is actually pretty interesting, from my view. Can you comment-- I think it's important that you did not see differences by race in terms of how concordant, maybe, the TMA was with the whole-mount specimen. So any thoughts there? At least it doesn't sound like there are differences by race on that.

Adam P. Dicker: No, but again, the MMAI score was created off standard microscope slides. So that will be done. I mean, we're going to find the standard microscope slides that result in, especially pre-prostatectomy, from these. There's a lot of work we're going to be doing to sort this out. There's a question about, if you take-- the whole mount's about double the size of a regular microscope slide.

So what happens if you just kind of use the algorithm on portions of it? So there's a way to kind of tile along the whole mount to see, does it-- it's a sampling error question, right, which is very important, in terms of when tissue biopsies come in. I mean it's an enormous issue in the field about sampling error in general. So we'll find out.

Alicia Morgans: Very good. So what would your take-home message be from this work to the viewers?

Adam P. Dicker: Take-home message is so far, at least in this limited set, whole-mount prostatectomy versus tissue microarray, there's not that much concordance. We also didn't see any differences vis-a-vis self-identified race, but look forward to more coming out in this story.

Alicia Morgans: That sounds great. I always look forward to hearing more from you, and I really appreciate your time and your expertise.

Adam P. Dicker: Well, thank you.