Unfold AI demonstrates utility for prostate cancer treatment planning in real-world case

In this interview, Joshua Trachtenberg, PhD, discusses his prostate cancer journey and how Unfold AI helped with his treatment decision. In addition, Wayne G. Brisbane, MD, explains how he used Unfold AI to help visualize Trachtenberg’s cancer and plan his treatment. Trachtenberg is a professor of neurobiology at the University of California, Los Angeles (UCLA), and Brisbane is an assistant professor of urology at UCLA.

Dr. Trachtenberg, please discuss your cancer journey and how you came to be under the care of Dr. Brisbane.

Joshua Trachtenberg, PhD

Trachtenberg: In brief, my father had prostate cancer, so in my 40s, I started getting semi-regular PSA bloodwork done. [My PSA level] gradually moved up over the years, over about a decade, and then the urologist I was working with retired and I was assigned to a new urologist. He immediately suggested I have an MRI, followed by a targeted biopsy. That was done in 2020. It was considered a watch-and-wait type tumor. And so I waited another year and a half and came back for a follow-up MRI and again, fusion biopsy, and this time, in out of something like 14 cores, 2 of them, I think, came back a higher grade Gleason score of 7 - 3 plus 4 in my case. And so that urologist suggested that I have my prostate removed post-haste. My father had gone through that, and my understanding was that it was a pretty bad procedure in terms of outcome for him. I was not willing to follow his footsteps, basically. I also understood that prostate tumors grow fairly slowly, and so that gave me time to research alternatives. And I should stress to anybody that receives the bad news that they've got prostate cancer, they have time to research things, they shouldn't immediately try and address the cancer. They should take their time and look into different options. I told my urologist at the time that I was not willing to consider surgical removal of the prostate and I asked him to put me in touch with people at UCLA who were looking at alternative approaches that might be based on different treatment paradigms. That put me in touch with Dr. Brisbane, and Dr. Brisbane worked with an interventional radiologist, Dr. Steven Raman. They were dealing with an approach that uses ultrasound to heat up tissue focally guided by MRI and therefore to focally destroy the tissue, just heat it up and kill it. I should note there are lots and lots of different focal approaches. Some are based on radiology, some are based on freezing tissue, some are based on using lasers to kill tissue with light. Others are based on thermal destruction using ultrasound. There's a bewildering array of approaches, and so finding the right one is very difficult. I had read something about targeted thermal ablation, and it seemed very appealing to me. When I spoke to Dr. Brisbane, first of all, he was the first guy to show me like exactly where my tumor was, to show me the MRI datasets and to tell me that it was a "unicorn" lesion, which was really nice to hear, and that, given its location, far away from things like the ejaculatory duct, far away from things like the neurovascular bundle on the periphery of the prostate, that it was in a great location for targeted treatment. The question then is, how big of region of the prostate are you going to ablate? And that's where he has close contacts with the people that developed Unfold AI. And he, really generously, ran my MRI datasets, which were also fused with the core biopsy datasets, to create 1 probability image. It shows you, with a red to blue color scale, which regions of the prostate have the highest probability for cancer. That gives you an idea of where the margins are and how much tissue would be destroyed and how much tissue will be left behind. And when you destroy tissue, what important structures could you avoid, and which perhaps are the structures if it's in a different location that you have to accept that you might destroy with thermal damage? That really grounded my expectations for how to move forward. And then we moved forward with thermal ablation approach.

Can you provide an overview of Dr. Trachtenberg’s case as well as how Unfold AI was used to help form his treatment plan?

Wayne G. Brisbane, MD

Brisbane: It will be very clear to readers that Dr. Trachtenberg is a pretty sophisticated patient. He's published in Nature and Cell and works in neurobiology. That's an organ where you obviously have to be very picky about any kind of treatments. The prostate is a little different, but more recently, there's been a lot of interest in dose de-escalation or just kind of precision in how we're approaching these treatments, which has been fueled a lot by MRI. MRI is not perfect; you can see the middle of the tumor, but there can be extensions outside of that. That's where Unfold AI has been helpful. For Dr. Trachtenberg, we did an MRI, we also did a PSMA-PET to look at tumor volume metrics. And then we used the fusion of the MRI and his biopsy cores to create this heat map that he alluded to, to show exactly where the tumor is located. For his particular tumor, I just kind of came up with the term "unicorn." It's rare, but it's really nice when you get them. It's an anterior tumor that is not huge in volume and appears in the mid gland. It's far away from all the critical structures, like he said, the ejactulatory ducts, the erectogenic nerves, even a little bit farther away from the urethra. So there are not a lot of penalties for doing a thermal ablation up there. Those tumors also have traditionally better natural histories, so they're probably more amenable to a thermal ablation. And he's also just a very sophisticated gentleman. He's very comfortable with risk percentages and understanding the follow-up nature and natural history of prostate cancer, so that was very helpful in selecting his treatment. But to get back to your question, you can use Unfold AI anytime you have an MRI and tracked biopsy cores. This is not required, but we got a PSMA, which showed that the center of the PSMA was the same location as the MRI, which is in the same location as his Unfold AI, which was in the same location as his biopsy positive cores. There was a lot of overlap so we are very certain about the tumor location, the tumor phenotype that we're working with, and in that case, it's pretty nice to go with some kind of ablation.

If you have an elevated PSA, in our practice, most guys are going to get an MRI following that. We did a tracked biopsy with one of my colleagues, and so he got targeted biopsies to the MRI, and systematic biopsies using the UroNav platform. And then basically, those biopsy core locations get exported and uploaded into a web portal. That gives you the MRI contour, the lesion, and then the pathology, and you just select all those things along with some demographics and it generates out a 3-dimensional heat map that you can move around. With Dr. Trachtenberg, I got to show him that and say, "This is your predicted location. This is where the nerve bundles would be. This is where your urethra would be. This is what we would do. And these are kind of the expected sequelae of that."

How much did having this 3-dimensional heat map help you with your treatment decision and comfort level with it?

Trachtenberg: I think there are 2 things that are super important to stress here. One is that, with all these different kinds of imaging approaches, and biopsies, and also genotyping with tumors to get information about how aggressive the cancer might be, fusing all these data together is really difficult. And AI really excels at this. If you go into ChatGPT4 for example, there are any number of sub-GPTs that you can now use for large-scale data analytics, and not just in medicine, but across all fields. These days, we're getting into this realm of big data, where there's so much data, and it's across so many dimensions that it's hard to really come to any kind of real understanding of it without some kind of tool like AI. And certainly for a regular Joe who's not spending all their day looking at patients, this kind of information that you get, showing where it is, what the probability of cancer would be, where the margins might be, is hugely, hugely helpful in deciding what approach to take. In my case, this was very focal, and I was comfortable going with a focal targeted ablation. But the second point that's important to make is that in all fields, one's happiness with an outcome is usually the difference between the actual and the expected. And so when you have this information from Unfold AI, you can see the location, you can ground your expectations for how you're going to recover in the months and years after the ablation. What are your reasonable predictions for obtaining normal urinary continence or obtaining erections or any number of other aspects related to your urinary or sexual function? If you have well-grounded expectations, you're likely to be happy with the choice you made. That's super important.

How has the use of Unfold AI improved your ability to have a treatment conversation with patients and hone in on the right treatment for the right patient?

Brisbane: What I would add from the clinician side that's useful is it's kind of a surrogate risk predictor of focal therapy success. We're going to hopefully be publishing very soon on that from a quantitative perspective. But you can get a pretty good gestalt of who's a good focal therapy patient and who's terrible, based on the kind of heat map. If it involves 90% of the prostate, you're not really saving much for the patient, and you're probably compromising cancer. And so visually demonstrating that, as Dr. Trachtenberg said, it just helps patients understand why you're even recommending radical treatment. That's very helpful for them to just conceptualize that, and own that treatment later if it's necessary. And then I think that the heat map is a very intuitive tool. It comes in a 3-dimensional structure so you can move it around. That kind of red to blue heatmap is just easy to conceptualize; it takes imaging, genomics, pathology scores, PSAs, all these things that are influencing the decision, and distills it down into, this is our model of what your prostate is, and then now what are we going to do about it?

Is there anything you would like to add?

Trachtenberg: It would be really helpful, I think—and I'm not sure if you can visualize this from the MRIs—but if the AI can also render, for example, in the 3-dimensional image, if it could render the urethra and the ejaculatory duct, and if it could render the position of the neurovascular bundle, I think that would also help patients get a much more intuitive sense of where their tumor is relative to all these structures, because that's currently not rendered. You get this heat map, which is very intuitive and useful, but adding a little bit more information would go a long way. Again, these are big datasets that are being fused together, and even after they've been fused together, it's still a very daunting task to digest all that information in the context of all the published literature out there and all that is known about best practices, and I think AI could, on the fly, take that data, and possibly give patients some recommendations for best-case approaches they might consider given the information they've been delivered. I'm super fortunate to be at a massive research center; we have the best and the brightest here. But there are a lot of people in more rural areas who would benefit hugely from this kind of deep AI information about their tumor and its prognosis.

Brisbane: I think obviously, Dr. Trachtenberg is a very good use case of it, but in general, I really like it for patient education. I think it takes a complex thing and distills it down very nicely without simplifying it. You can always oversimplify things, but I think you're able to educate patients by giving them real data in a way that makes sense to them and gives them a very accurate picture of what's going on. We use it for margin planning as well, and then I do think, as the sophistication improves, you'll also see risk stratification, better understanding of which focal therapies are more appropriate, given certain tumor phenotypes, and stuff like that.