Melanie Cole, MS (Host): Welcome to Better Edge, a Northwestern Medicine podcast for physicians. I'm Melanie Cole. And today, we're highlighting applying AI to improve local staging for prostate cancer. Joining me is Dr. Hiten Patel. He's an Assistant Professor of Urology at Northwestern Medicine. Dr. Patel, thank you so much for joining us for this fascinating conversation today. And before we turn to the application of AI in prostate oncology, I'd like you to start with the role of prostate MRI to detect and stage prostate cancer, and then we'll evolve into the goal of using AI to improve this.

Dr. Hiten Patel: Thanks for having me, Melanie. Yeah, I think this is an exciting area because prostate MRI was basically not used at all 10 years ago to detect or stage prostate cancer. We used to use our fingers to be able to tell, "Hey, is there potential for that prostate cancer to be extending out of the prostate Or what we call extraprostatic extension?" And now, prostate MRI in the last 10 years has really taken over, where we can get a good picture of what the prostate looks like, and a radiologist can look and see if any lesions that look suspicious for prostate cancer or something we've already confirmed as prostate cancer, does it extend out of the prostate? And they kind of give us their interpretation of how likely it is that's happening.

Melanie Cole, MS: So then, let's talk about your AI research on the role of prostate MRI and speak about how that's all tying together. Really an exciting time.

Dr. Hiten Patel: And I think the way to frame this is to say, well, why do we want to locally stage prostate cancer? And it's important because extraprostatic extension is associated with recurrence of cancer after treatment and the future risk of metastasis. And so, it really does tie to outcomes for the cancer. And even though we had methods to predict these things in the past, MRIs helped us. And at Northwestern, we use it over 90% of the time for anyone who we think might have prostate cancer, so we have a lot of good data to work with. And one of our early work has looked at whether we can improve upon old models where they use clinical factors, like patient's age, the grade of the cancer, and things like that to predict whether there might be extraprostatic extension after surgery on the pathology, because that's really the gold standard. We took the prostate out. Is there really extraprostatic extension or not? And along with grade, these are important factors. And we have an early study that's coming out soon where we made a risk calculator that could help clinicians by using these factors predict on the dominant side where more of the cancer is, and on the non dominant side, what's the likelihood there could be extension of the cancer.

And that's important because, one, not just because it's associated with outcomes, but two, it's actually associated with what decisions I might make during surgery. So for urologists doing radical prostatectomy, if there's cancer extending out of the prostate on the dominant or the right or left side, that can impact how close I get to the prostate when I do the surgery. And that's important because if we can spare nerves on the prostate, that helps men keep their erectile function, and quality of life that can be improved. But if the cancer is going to be extending there, sometimes we favor going a little wider to make sure all the cancer is removed, if that's the priority, the oncologic control is the priority. And so, actually having a tool that tells us better is there something there or not is important. And so, that's where I think we stand with our prior work is that we've created that risk calculator, but it's good, but we think we can use AI to make it better.

Melanie Cole, MS: As we're talking about the clinical significance of using AI to improve that accuracy and efficiency of prostate MRI and staging and what you've just discussed, speak about the benefits or implications of successfully integrating AI because it's so interesting. How is that going to improve what you just mentioned when you're talking about local staging?

Dr. Hiten Patel: So, I think for the prior work I mentioned, we recently received a Prostate Cancer Foundation Young Investigator Award that is going to help support this project. And what we're trying to do is, one, actually look at what the radiologist variation is in identifying or describing EP or extraprostatic extension. That is something that radiologists vary on how they interpret. They may say there's abutment of that cancer on the shell of the prostate on the scan. Some may say there's actually an irregularity or extension outside the prostate. And so, because there's some variation across radiologists, and there's no real set of rules that tells us how to grade this, what the likelihood of grading the extension is.

There's a rule called the PI-RADS system that helps us grade the likelihood of cancer being present. But when we know cancer is present now, how do we grade the likelihood that it's pushing out of the prostate? And that's where AI, I think, can improve upon it. Because what we can do is with the lab by Dr. Ulas Bagci, who's one of my mentors on this project, he can basically take prostate MRI images and train a model on the right or left side where we can tell the model, "Hey, on surgery pathology, there was definitely extraprostatic extension there. And on this side, there wasn't." And we can use a model to train it and say, well, we don't use what the radiologist said, but the final pathology to help the model learn and look at these images and help tell us on the right or left side, is there extraprostatic extension present? And then, we can look back and compare what the AI model says to what the radiologists suggested on the scans initially, and see if that's better or worse than what they had said.

Our hope is that we can make it better than what they had said, and that this could later lead to a tool that they could then use, that the AI model gives them a pre-interpretation. And then, it's still up to the radiologists to look and say, "Oh yeah, I agree or don't agree with what the model says." But if they can get that information to start with, it gives them a better starting point than they were looking at it more blind at the beginning. And so, that is a what we call assisting tool. But at the same time, if the AI model is just overall better at predicting it, we can use that directly from the images too.

Melanie Cole, MS: And can it also then, along those lines, help improve reproductibility for those radiologists and detecting subtle changes when you're doing active surveillance monitoring. Are these value additions that are being investigated?

Dr. Hiten Patel: I think AI for detection and monitoring is already happening. Dr. Beck, she has a grant looking at detection using AI. And so, that's why we were hoping to now push it towards staging. For detection, there's also some variation, but at least there is a standardized schema for how radiologists can grade likelihood of cancer. We can potentially reduce that variation if they get a baseline, or if there are some radiologists who are newer or not trained on prostate MRI as much as others, maybe be able to get them to a good level of interpretation out of the gate, as opposed to requiring building up experience over time.

And so, it gives them a good floor. And then, their ceiling obviously can get increased from there. So, those are tools that are already in process. And the staging tool, I think, will also be important, because those interpretations will help the surgeons, or even radiation oncologists, make decisions where I can decide, well, the likelihood of extraprostatic extension is low enough that I could go closer and do more nerve-sparing for a patient, or I can advise them ahead of time. Because right now. I usually tell a patient, "Well, I'll see how it looks during surgery. And if it looks like I expect, I'll do more nerve-sparing or less to help optimize cancer control." but now, if I can tell them that more ahead of time to say, "Well, I'm now I'm more confident because of this model or because of this interpretation," I can tell them with a little bit less uncertainty of what's going to happen.

And same thing with radiation oncologists, they can potentially apply more or a boost of radiation to a specific area where the cancer may be extending out of the prostate and do it more confidently now that there's a model that's telling us what's going to happen, because when patients get radiation treatment, their prostate is not removed. And so, we don't know for sure was there extraprostatic extension there. But now if we have a tool that can tell us more confidently, they can change the treatment plan to help with that.

Melanie Cole, MS: It's really changing the landscape of prostate cancer, for which there are already a number of tools in your toolbox. But this could really, as you just said, help to see if it's gone out of the prostate when you couldn't really tell before. And so, that is so interesting. And Dr. Patel, what about other AI work being conducted at Northwestern Medicine to improve general urologic cancer care? What are some of the goals of future research?

Dr. Hiten Patel: The goal is to have AI help us, and even though it's a black box, use it where we know or we think clinician judgment can improve care. And so, we could try applying it to everything, but the issue becomes that if there's some things that are going to be useful and some things that are not helpful, here we try to target use of it onto a clinical question. So, can we identify extraprostatic extension or improve staging? Can we improve detection? And we have to see what are we already doing and what are the easy ways to make it better? Dr. Ross uses this and Dr. Cooper, who is a PhD working on digital pathology and applying AI to actual pathology slides of biopsy tissue and prostate cancer specimens that have been removed and seeing if that can help pathologists by already identifying where the cancer is in the prostate, is there extraprostatic extension and things like that on the slides on the pathology side? So, we're trying to push the radiology side to get better at it. The pathology side is already ahead of us where we have tools that can detect cancer and then help a pathologist hone in on different areas to look at. And I think those applications are important.

And my ultimate goal is also to look at in kidney cancer or kidney tumors that are removed have different levels of complexity. And so we can look at the images, but maybe AI can tell us a little bit more about how complex or what the risk of removing that tumor is compared to another patient. So, it really depends on the clinical question because with the prostate, removing the whole prostate with kidney tumors, we're trying to focus on a tumor and preserving the rest of the normal kidney tissue. And so really, we have to adapt it to what is the clinical question where can AI help us?

Melanie Cole, MS: Do you have any final thoughts and key takeaways for other providers, Dr. Patel, on this exciting work? Your research, future goals, kind of summarize everything with the important message about how AI is being used in urologic care and really expanding that toolbox.

Dr. Hiten Patel: For this project, we're very thankful to the Prostate Cancer Foundation for providing the award to pursue this angle. But I think in summary, we know that there's variation in care. And so, the problem here being variation in identifying extraprostatic extension. And so, there are measures we can take from a health services standpoint of evaluating how that's being done and helping clinical interpretation to become more standardized. And I think AI as a tool in that area, if it fits into your clinical question, it's important to try to use it. And I think we're fortunate at Northwestern to have Dr. Cooper and Dr. Bagci, who are two AI experts on Pathology and Radiology that help us advance these fields. And I think applying it to prostate cancer because of how common it is and because how much of that we care for at Northwestern is a great place to start to apply these tools.

Melanie Cole, MS: Thank you so much, Dr. Patel, for joining us today. What an exciting time. And thank you again for sharing your expertise. And to refer your patient or for more information, please visit our website at breakthroughsforphysicians.nm.org/urology to get connected with one of our providers. That concludes this episode of Better Edge, a Northwestern Medicine podcast for physicians. I'm Melanie Cole. Thanks so much for joining us today.