Zachary Klaassen: Hi, my name is Zach Klaassen, urologic oncologist at the Georgia Cancer Center, and I'm delighted to be joined on UroToday by Dr. Geoffrey Sonn, who is a urologic oncologist at Stanford University. Today, we're going to be discussing Unfold AI and some of the implications in prostate cancer. Geoffrey, thanks for joining us today.

Geoffrey Sonn: Thank you for having me. It's a pleasure to be here.

Zachary Klaassen: We're excited to chat with you. You've got some really cool data and some slides just talking about Unfold AI. Before we get into that, just by way of background for our listeners, why is it important for AI-based decision support tools today for localized prostate cancer?

Geoffrey Sonn: I think there are many places across the spectrum of prostate cancer where AI could potentially play a role. As you mentioned, for today, we'll talk specifically about patients after diagnosis of localized, intermediate, or higher-risk prostate cancer and how an AI tool like Unfold may be helpful. I think there are several different potential applications for this, some of which we'll cover here in a moment. But for me, using this clinically, and this is a tool that I'll use fairly routinely in my clinical practice, it can be helpful for counseling patients to help them have a better understanding of what they're dealing with regarding their prostate cancer. I think it can be tough for a patient who gets a systematic and targeted biopsy. It may come back and show they have seven cores, and to understand, does that mean that I have seven cancers? Does that mean I have one cancer that was sampled seven different times, or somewhere in between?

So, a tool like this can be helpful to be able to display visually to a patient so they really understand what is the nature of my cancer, how big is it, and how that might impact our selection for treatment. I think furthermore, with the emergence of focal therapy, it becomes increasingly important to know more about the cancer size, location, and margins, and this is where Unfold can be particularly helpful. Focal therapy is really promising, but the recurrence rates in the short term can be upwards of 20 to 40%, which, in my view, can be unacceptably high. So we'd like to drive that down further, and I think we can do that using software like this to get a better understanding of size and location of the tumor and who is really an appropriate versus inappropriate candidate for focal therapy.

Zachary Klaassen: That's a great background. I know you've got some great slides you're going to pull up for us just to walk us through how Unfold AI works and some of the data that led to the clearance from the FDA in November 2022. So, why don't you pull up those slides and walk our listeners through those?

Geoffrey Sonn: Absolutely. So, we're going to talk about prostate cancer assessment using Unfold AI. And fundamentally, Unfold AI is a multimodal AI program that's designed to provide a more accurate representation of prostate cancer size and location. And when we say it's a multimodal AI tool, that means it takes as input several different types of data. So, we can see the T2. This platform will incorporate the T2 MRI, PSA value, biopsy pathology, as well as the biopsy core tract location, and it incorporates all of that information into the neural network, this AI architecture, and then will output a cancer prediction map, essentially a heat map where we can see the areas that it thinks cancer is not present in blue. And the areas that cancer is present will be in red here. We can see the radiologist annotation as well as the cancer core tract location. It'll also generate this superimposed on the T2 MRI. This purple line here represents the area that, if focal therapy is performed, it's suggesting ought to be the margins for that focal therapy treatment.

So, when you think about how this might be used clinically, one is for patient counseling. It can be helpful to display this to patients so they actually can see what their cancer looks like, its location, its size, and what implications that may have in our treatment selection. Specifically for patients who are interested in focal therapy, it can help them understand why or why not they may be a candidate for focal therapy, and it can help us as urologists to define what treatment margins we want to conduct for maximal success. It's also helpful for patients who may not be interested in focal therapy to help with surgical planning. So, there are a couple of validation studies that have been published using Unfold to date that show that it improves the ability to detect extracapsular extension, as well as another study showing that it does a better job of detecting seminal vesicle invasion than just relying on the MRI images and/or radiology report alone.

So, what does this look like clinically? This is a sample of the output that we can look at as urologists, or we can show to patients. In the middle panel here, we can see this heat map that shows where the radiologist thought the cancer was present. And once you incorporate the biopsy core tract location, so these little red dots here indicate grade group 2 or higher cancer on targeted biopsy. The yellow are low-grade cancer. The blue are benign. So, it's generating this cancer estimation map as well as the planned focal therapy margin. On the right, on the T2-weighted image, we can see those biopsy core tract locations. And then all the way on the left side, this is actually an interactive tool, so the urologist is able to drag this around to generate... If we want to be more aggressive, treat a larger area to get higher confidence of treating all the cancer, that is something that's certainly possible.

Now, of course, this is a three-dimensional problem in a two-dimensional display, so when we're going through this, you're able to scroll. And this is just a short little video clip scrolling through a patient of mine. You can see as we come through here, the biopsy core tract locations. This is a particularly large cancer that encompasses virtually the entire left side of the prostate, so this would not be a great focal therapy candidate, but we can see here, taking in the T2 image as well as the biopsy core tract location, we get a better sense of the cancer size and location.

So, again, clinical applications of this: it can be helpful for patient counseling. It's just a nice way to display for patients what their cancer looks like. It's easier to talk them through this than it may be just through a written pathology report. It's helpful also for us as urologists to truly have a good understanding about cancer size and location to help us know, is this somebody we ought to be discussing focal therapy with. And/or if we're doing a radical prostatectomy, what sort of margins, and when we ought to nerve spare more aggressively or less aggressively.

Now, focal therapy, I think, is really where this shines, and I'm going to talk a little bit about a couple of the studies that have been published. The first one I'm most familiar with, because we did this at our institution: we took 50 consecutive patients who had a radical prostatectomy done at Stanford for intermediate-risk prostate cancer, and who had had a targeted biopsy that was done before. We then processed those specimens in a whole-mount fashion and had our pathologists annotate the cancer after scanning it in digitally, and this served as our gold standard. We then ran the AI model on the biopsy location and the T2 images to generate treatment plans and compared that to the more standard hemi-ablation strategy or the radiologist annotation, plus a one-centimeter margin, and showed that using this AI platform, the negative margin rate was 90%, which is better than the 66% we'd get with hemi-gland ablation, or the 82% you'd get with simply adding a one-centimeter margin on top of the radiologist lesion.

Now, since then, there's been another study that was fairly similarly performed. It was also 50 patients in a population of patients who had a radical prostatectomy. This was a multi-reader study, where 10 physicians went in and annotated where they would have planned to do focal therapy initially with the standard-of-care information, that is the pathology report from biopsy, the MRI, and the radiology report, and annotation of the lesion. And that was then compared to the same readers after a washout period, who were provided with the AI output and then asked to annotate where they would treat. And this was a really remarkable difference. Using the AI tool, the negative margin rate was 73%, which is much better than the abysmal 2% that we got just using a standard-of-care approach. So, a couple of validation studies that show how this can be helpful in a focal therapy population. And then finally, here are a couple of the papers that have been published, the validation studies, several of which I've mentioned in the preceding slides.

Zachary Klaassen: Geoffrey, beautiful pictures, beautiful video, I think, just really highlighting visually how this looks for the patient. I was going to ask you about applicability. You addressed that beautifully in the slides as well, so I'll maybe switch up the question a little bit. What's the patient feedback been when you're able to sit down in the clinic, show them what this means, help them make those decisions with Unfold AI?

Geoffrey Sonn: I think patients usually do appreciate the ability to look at some of the images. It's really common that someone will come for a second opinion and say, "Wow, gosh, I never was able to see these images before." And Unfold makes it even easier than taking them through the MRI. It graphically displays the full extent of the tumor. A lot of the patients that I see may come to see me because of an interest in focal therapy, and it can be pretty disappointing when I just tell them no. And to be able to actually take them through and thoughtfully explain, "You know what? This is why this is not going to be a good treatment option for you, and instead this is something that we should offer." It's an easier pill to swallow for patients, and I think it's really helpful to them to understand not only what their options are, but why.

Zachary Klaassen: Yeah, I think that's awesome. When I think about focal therapy, it's in the Wild West now, and really honing in on who are appropriate candidates, this really does help us pick who would be an appropriate candidate, doesn't it?

Geoffrey Sonn: It does. It does.

Zachary Klaassen: Awesome. Favorite question of the evening: what's next for Unfold AI? Where are you guys taking it? What's coming down the pipeline, if you can share a couple of thoughts about where you're going future-wise?

Geoffrey Sonn: Sure. Well, I mean, first thing that I'll mention is that even though I use this clinically and have been involved in some of the validation studies, I'm not a member of the company and not on an advisory board or anything, so I don't know all of their details. I think they probably do want to expand even beyond prostate cancer into other malignancies. They are trying to expand within prostate cancer in other applications we talked about, just general counseling. It can be fed... this output can be fed into the context of doing radical prostatectomy for someone who's interested in that. I think the biggest focus is really on getting this out there so that, as urologists, we are aware that for those of us doing targeted biopsy, this is a tool that is out there and available that can help with counseling patients and help us with treatment selection.

Zachary Klaassen: Awesome. And maybe just on that lead-in point, how can we get this into our clinics? What's the next step for those listeners out there that may be interested in learning more?

Geoffrey Sonn: I think with most companies, there are teams that are in place to help us bring any sort of clinical tool to our practice, so I would just encourage urologists who are interested in learning more and seeing if this may be helpful to them in their practice to reach out to the folks at Avenda Health. I'm sure they'd be happy to help talk them through the process of doing this. There is a CPT code that is now recently available, so this is something that can be billed and reimbursed. I'm not an expert in that. The folks at Avenda are very helpful with this.

Zachary Klaassen: That's great. Last question, just anything we haven't hit on that you want to hit on before we wrap up, maybe a couple of take-home messages?

Geoffrey Sonn: I think we've covered this fairly comprehensively so far. I would say that in my mind, there are some nice validation studies to date that show that this really does improve our ability to assess cancer size and location and to help with treatment planning, decision-making about what type of treatment. This is something that is applicable really for those of us using targeted biopsy. So you do need targeted biopsy with biopsy core tract locations for this to be useful, or to run these maps in the first place. But for those of us doing targeted biopsy, whether or not we're doing focal therapy, this is a tool that is available and helpful to patients.

Zachary Klaassen: Awesome. Geoffrey, great discussion. Thank you for educating our UroToday listeners. Really appreciate your time and expertise on UroToday.

Geoffrey Sonn: Thank you for having me.