Chapters Transcript MRI in Prostate Cancer Detection - What We Know and What We Don't Course: Advances in Screening, Detection, and Treatment of Prostate Cancer Uh, well, thanks everybody for coming to our course. I, I see a lot of familiar faces, uh, who seem to come every year, but in case you haven't come before, my name's Samir Taneja. I don't look anything like the picture that Herb put my name under with a probe up some guy's ass. So, uh, this is what I look like, and, uh, this, uh, a lot of my work over the last, uh, now going on 15 years has been in the area of imaging. Um, I've given this talk many times, so I tried to switch it up a little bit, maybe jazz it up a little bit, uh, to, to keep your attention, but really what I want to convey to you is what I think urologists should know about MRI, uh, what they should know about what it can do for your practice, but also then what the limitations are and maybe how we're gonna try to address some of those limitations in the future. Uh, some of my slides will probably overlap with Angela. She emailed me yesterday and said, how do we not overlap, but a lot of our work has been very collaborative between neurology and radiology. So let me see how to do this, so I a few disclosures, none of which are uh related to this talk. So what do we know about prostate MRI? Well, in summary, when you use it before a biopsy, it can detect and localize the disease. It can potentially avoid the need for biopsy in some men and can be used to guide a biopsy strategy. When you use it to guide the biopsy, it can allow more accurate risk stratification. It can allow selection of candidates for focal therapy, and it can reduce the biopsy extent or or sampling. Uh, and when it's used after the biopsy, it can identify occult missed cancers in some cases, but it also can help in staging, uh, the cancer as well. So when we started, uh, Andy Rosenkranz and I started our imaging program here probably around 2011, right when he was finishing his fellowship, and our initial goal was really just to refine the imaging. So you know, the, the, the images you just saw in the last talk didn't just emerge, they were really a long standing refinement. Uh, of, of a process and we, we felt that the imaging now, uh, uh, it's, it's at NYU is quite high quality, but along the way we learned a lot, uh, and the way we did that was when we first, uh, I'm I gotta get this, when we first, uh, developed the technique, we then also developed a way of trying to do MRI targeted biopsy so we could query what we're seeing on the imaging, and that was the fusion biopsy that helped us to further refine the imaging. And from that have come several offshoots as I mentioned, improved biopsy outcomes, risk stratification methods for active surveillance that allow biopsy de-escalation, better data for which we can counsel patients on treatment and of course uh our initial efforts led to our ability to empower a new treatment paradigm in focal therapy. So since 2012 at NYU we made sort of a commitment that every biopsy we do will be under MRI targeted biopsy approach. So all our patients, nearly 98%, have gotten pre-biopsy MRI's and uh and undergone typically a transrectal but more recently in many cases a transperineal fusion biopsy. What we learned early on has has really persisted in our follow up, and that is that MRI is very good at predicting the likelihood of significant disease and so you can see in the red when the pyre score is high, very high likelihood we have what we call significant disease when it's low, very likely the patient either has no cancer or low grade disease, and of course there's this middle ground that sits as sort of the Achilles heel. Uh, what do you do with the patients with equivocal imaging or the patients with suspicious imaging whose biopsies reveal lower risk disease? Well, how does a pre-biopsy MRI really help you as a urologist in practice? So here's a case of a 62 year old comes in with a PSA of 6.9. MRI shows an anterior region of suspicion. So you could argue that here you've used the MRI to increase the the chance of identifying clinically significant prostate cancer. You're more confident in the results or the disease risk you've created a better disease map and if cancer is present, you may be able to answer a number of clinical questions that you might not have if you didn't have that data coming into the biopsy, even surgical questions about nerve sparing or or or ease of surgery. And in the case of the patient who has a normal MRI, you can use that data then to further risk stratify the patient decide if they need it, decide if you can find another explanation for their elevated PSA, the size, the anatomy, and if the biopsy is done, of course have greater confidence in wrist stratification again. So there are many ways in practice now that we use MRI as a standard piece of our clinical decision making. Now when we look at our initial data, this is a 4x4 table. MRI has good and it has bad, right? The green is the good, it's the patient who if they had undergone a systematic biopsy, they would have come back as benign or low risk, but in fact by the targeting, we find the higher uh the higher risk disease. And of course the, the bad, as many critics of MRI point out is the red here, the people who, uh, uh, it seems the MRI or targeted biopsy misses the cancer. Now some of that is technique related. We've published that MRI targeted biopsy is a skill. It's not just something that you can sit down and do and in fact in our own experience when we put our biopsies early on into stratiles of 200 patients roughly. Each 200 patients we saw an increasing rate of significant cancer detection in the Pyrette 4 category. Now that may be in part improved reading of films, but I don't think so. I think actually it's mostly getting better at doing biopsies, and we see that learning curve among early surgeons just starting in our program as well. All right, so what do you need to know as a urologist? I think there are 4 fundamental studies that tell you about how the MRI performs and and that are good for urologists to understand. Of course, the PRIS study taught us that MRI is not perfect, right, that it doesn't see all cancers. So when I just showed you, learning curve is one of the reasons we miss cancers, well, we probably miss it because MRI is not perfect. Now this study has to be taken within the limits it has. It was done with 1.5 Tesla MRI, but in the PRIS study, what happened is that all the patients underwent a 1.5 tests of prostate MRI, a systematic biopsy, and then a transperineal mapping biopsy. And the outcomes of the MRI and the trust biopsy were compared to the mapping biopsy as a standard reference. So the way the study went was that there were 576 patients. They chose to define clinical significance in a bit of an odd way. Gleason 4 + 3 or higher or more than 6 millimeters of cancer, but they segregated the prostate MRI. Into what they perceived to be a risk of clinically significant cancer in about 80% and no risk in about 20% and what they found was basically that MRI had about a 51% true positive rate if you correlated it with the biopsy, but remember these were not MRI targeted biopsies, they were just template biopsies. And it had about an overall 11% false negative rate in that when they didn't see it, they found clinically significant cancer by that definition. The trust biopsy's performance was, as you would imagine, higher false negative rate, uh, and uh a very high true positive rate uh when the trust biopsy showed significant cancer, of course the template biopsy should as well. Most important with this study was and what people focused on was the negative predictive value and depending on how you defined clinically significant cancer, most of us use this arbitrary greater than GD2 designation, greater than equal to the negative predictive value declined. So if you used a looser definition of clinical significance, up to 25% of cancers were missed by the MRI prediction. Now, some of that relates again to learning curve with interpretation. I've seen some recent data where people have gone back and rehashed the raw data, and maybe the prediction is better, but the bottom line is MRI won't find every cancer. How significant that is, uh, we have to understand. Precision is the next study. This was a validation of the approach. It was actually a study we designed as a collaborative group at a workshop here at NYU. And it was carried out on a global level randomization of systematic biopsy as a standard of care versus no systematic biopsy, no biopsy for pyrid 1 and 2, and just an MRI targeted biopsy and intended to show non-inferiority between the two arms. So the way that study played out, men were randomized to MRI or trust biopsy. If they had an MRI, about a third of them didn't need a biopsy based on the prior had 12 designation, and in fact that's very consistent in the literature dating back over the last decade. Uh, 2/3 needed a biopsy, and of those, um, The 40% roughly had significant disease. The trust biopsy about 30% had significant disease. So the conclusions were 30% avoid a biopsy, more clinically significant cancer and less indolent cancer in the men who underwent the targeted biopsy arm. The aftermath of precision of precision was somewhat disparate. Pre-biopsy MRI has largely been accepted as a risk stratification tool in Europe. Uh, and, uh, has made it into a lot of the guidelines was criticized in the United States initially because of a failure to, to, to query the people who weren't biopsied and uh how many cancers were missed in that arm, uh, and it's really been largely rejected up until very recently as a risk stratification tool but used pervasively to guide biopsies. I think that's a bit of a problem because to justify the cost of MRI. You probably have to use it to avoid biopsies. If we're just using it as a biopsy guidance tool, the cost escalation is probably not warranted. Uh, the next study is MRI first, so this study was conducted differently. 250 men who underwent both a trust biopsy, 12 core and ultrasound guided, and then, uh, uh, a targeted and what you see in the blue box is that this study is more consistent with what most most of us see that systematic and targeted biopsy finds similar amounts of significant cancer, but when you combine the two, you increase the overall rate. Uh, importantly, the added value of the targeted biopsy is about 7%. The added value of the systematics about 5%, and I think that's pretty consistent in the literature as well. Important to note though is if you do the systematic biopsy in the red box here, uh, you will find a lot more low grade cancer in your patients, uh, and so in looking at carefully at the MRI first data, uh, the, the men who underwent systematic biopsy. Were accounted for 80% of the indolent cancer detection in the study overall, so you quadruple the rate of overdetection if you choose to oversample or increase sampling of the patient. The final study that I think it's important to look at is a screening study that came out of Sweden. This was a study in which they tested two screening strategies. The reference strategy was a traditional, um, uh, all men with PSA greater than 3 get an MRI followed by a systematic biopsy and a targeted sample. The experimental strategy was that those men would not have a biopsy if their MRI was low suspicion. And they would have MRI targeted only if it was suspicious, so basically the precision strategy. They concluded that avoiding systematic biopsy in favor of MRI directed targeted biopsy reduced overdiagnosis by half, but I think their conclusion actually was slightly incorrect because if you look at their data, you don't have to look at all this data, uh, but if you look at their data, what you see is that in the men who are pyrid 34 5 in both arms. Whether they had a systematic biopsy or not, the detection rate was about 7.4% for low grade disease, so it really wasn't different whether you did the systematic or not. The risk reduction came from avoiding biopsies in men who were pyride 1 and 2. And if you avoided biopsies in those men, you really only missed 2.2% of what we called significant cancers. So the real opportunity for reduction of overdiagnosis in my opinion, has less to do with biopsy technique and more to do with avoiding biopsies in low risk patients. Now what's the consequence of that? Well, one of our residents, Zach Feuer, did an analysis where we took the precision, uh, strategy. And data set and we superimposed it on our institutional data set and what we saw is that if we avoided biopsies in our pyred 12 patients, the majority of cancers we missed were low grade, but there were some clinically significant cancers we missed, keeping in mind that we don't biopsy every pyrid 1 and 2. If we applied risk stratification scores like Cara or Epstein to our cohort, we found that the vast majority of cancers that were missed in by avoiding biopsies and pyrid 12 would be low risk, uh, on CAPA scale. So how do I manage patients with low risk MRI? Well, I do further risk stratification. We'll talk about it a little bit in the case presentations later, but you really have three options when you meet with a patient who's got an elevated PSA and a low risk MRI. One option some need a biopsy anyway if they're high risk, uh, and those would be people with a clear PSA elevation, a high PSA density, marked PSA velocity. People with a strong family history or genetic risk, some can have further risk stratification, so if you're not really sure, these I think this is a good population to uh use the biomarkers or nomograms uh to try and help with that decision. I'm still a big believer in MRI first for everybody as a primary risk stratification tool. I know I I might be a growing minority in the room. Some can have further deferral. The vast majority of patients who come to me with low suspicion MRIs, I don't do any further risk stratification. I follow their PSA over 2 years, and if it remains stable, I tell them to go back to their primary physician. That's been shown to be safe in the data, and only about 5 to 10% of those patients will have a rising PSA that warrants a deferred biopsy. All right, so this is where MRI has left us as urologists. You have a choice to make when you do a biopsy. You can increase sampling and biopsy more people, and you will find a little bit more clinically significant cancer but greatly more indolent disease. On the other hand, you can rely on the MRI, decrease your sampling, you will find a good amount of clinically significant cancer but undoubtedly miss a few. But you will greatly reduce indolent cancer detection, uh, and, and perhaps the number of biopsies you need to do. The other application that Angela alluded to for MRI is how do we use it in surveillance. So I'm gonna briefly show you this. So since 2011 I've used a very empiric surveillance protocol, and that is if men come to me referred with low or or intermediate risk favorable risk prostate cancer or if they're simply referred with elevated PSA we do an MRI and we do a targeted biopsy. If the targeted biopsy is low risk, we then put them on surveillance, we repeat the MRI and the biopsy at 1 year, and if that's confirmatory biopsy is low risk, then they go on to a 5 year biopsy interval schedule with uh with interim follow up with PSA and MRI and, and of course 4 cause biopsies is needed. We've just submitted some of this data for publication so we've hashed it out, but this is what thus far it looks like. Uh, of the men we've enrolled, uh, 537 of my patients have undergone a one year biopsy. They have a median follow up of about 39 months, and the Gleason upgrade rate on that one year biopsy is about 3. So it's very interesting that even when you do MRI targeted biopsy, if you bring them back a year later, there's a reasonable chance they're gonna be reclassified. We think some of that's early progression. We think some of it's targeting error, uh, and, and that's something we're gonna be dissecting, but it is about a third. Pyre is predictive of that to some extent. Within the for cause biopsy compartment of the men who are being followed in year 2 to 5 at a median follow up of 75 months, 50 of them have required a 4 cause biopsy, and the reclassification rate and treatment rate in those men is pretty high. So this has taught us that when you monitor men. MRI is pretty good at picking up significant progression events and when you need to do a biopsy. The reason I put the 6 year biopsy in is without it you can't define the negative predictive value of the MRI and follow up so we ask all these men to come back 5 years later and get a biopsy so that we can see what is the rate of reclassification when your MRI didn't predict reclassification and in fact in those men it's about 50% will will reclassify predominantly as I'll show you GG1 to GG2. But only about 10% have required treatment and some of those have just been people who chose to have treatment. Uh, Within the 6 year schedule biopsy, um, 12 of 99 went on to treatment after that 6 year biopsy. 3 of those had stable disease. 7 of them had focal therapy, so part of the treatment was, hey, now we have this focal therapy to offer you, and only 5 have really required radical treatment. So our own and, and I'll show you the the GG1 GG2 classification, but you can see that of the men who were upgraded on the 6 year biopsy to GG2. The 20 men, uh, some of them were baseline GG2 as well, only one higher than GG2 reclassification. So we think our data shows that MRI has good positive predictive value, good negative predictive value in in in detecting reclassification events on surveillance. Uh, the challenge is what Angela alluded to before. How do you define change on MRI? I'm not convinced the precise score is the right score because when we use it, our performance is not as good, but it is, we do need ways of trying to detect this. This is a great application potentially for AI but you can see here in the upper panel in the green, there's a gentleman who came in, targeted biopsy at baseline. He's one of these AS patients. 2 out of the courses from the target showed uh GG GG1 disease. A year later, the target appears to be slightly larger, more conspicuous on ADC, and in fact he was reclassified to GD2 and actually chose to have a prostatectomy, um, although he would have been a good focal candidate. So what, what is it that we don't know about MRI? Well, one of the most important topics that's interesting people now, and I won't, I'm not gonna say a lot about it because I think there's going to be a lot of discussion. Is what does it mean to have tumor on MRI or not to have? What is the difference between tumor visible and tumor invisible disease? And we know that presence of tumor on imaging predicts worsen outcomes after surgery, more likelihood of progression on surveillance, the quantitative metrics as Angela showed you in this slide are predictive of higher grade disease, uh, with, with strongly overlapping confidence intervals. Andy and I, Andy Rosenkranz and I, one of the very first questions we asked was that very question, what is the difference between these? We saw a number of morphologic differences between visible and invisible tumor, uh, more densely packed cancer cells, less stroma, higher Gleason scores, all these things were independently predictive. Uh, and now going forward we're still at that place. I just reviewed a thesis from London on MRI invisible disease and it's insignificance, but it's still really suggestive data, morphologically different. Now there's some genetic different uh data to show it's genetically different, but I'm not sure we can, and I'll show you some cases later, not sure we can always conclude that if you don't see it on the MRI, it's not significant. I think MRI is one variable. That helps us with the risk stratification, but probably not ready to use it in isolation as as some are proposing. OK, final, uh, couple minutes. What are the limitations of prostate MRI? Well, I've already told you it misses clinically significant cancers. We know it's costly and probably costs can't be sustained, particularly if we don't use it correctly. I'm now starting to see people coming to me from the community after they've had their 4th MRI. You know their primary orders an MRI every 2 years and if it's fines you don't need to see the urologist. These are things we have to do away with through education, uh, and otherwise it's gonna spiral out of control. Access of course is a problem in parts of the world. So how are we trying to address this? Well, the clinically significant cancer Angela and I started a project. Uh, a couple of years ago where we decided to try to understand why we see discordances between MRI and uh and uh pathology. So we, uh, every Friday tried to sit down for a couple of hours and go through cases and we defined a process so we defined discordant definitions in the left box we then created a process for evaluating those discordant cases with the pathologist. We then classified them according to different categories and then we created a validation to say how do we prove if that's correct. So examples would be where we thought there was targeting error. So in this case, the patient has GG2 and core number 7 systematic, but a benign target 13, well, we suspect that's targeting error, right? In this case, the patient has GG2 and a pyres 2 lesion, but when we do a consensus re-review, it gets reclassified to GG3. So maybe that, uh, I'm sorry, Pyres 3. So maybe that's a class misclassification due to the reading. Uh, and sometimes it's histology. Here's a patient with what's called a pyre 4. It's benign. It shows acute and chronic inflammation may be the cause for the appearance on the MRI on the histology. So if you find something on the histology that explains the MRI finding, you understand the discordance. So we looked at over a period of a year, uh, only my biopsies, 472 biopsies consecutively, there were 61 that we defined as discordant so that's not insignificant. Of those 12 of the studies were MRIs that weren't performed here, so we attributed to that and uh uh uh uh a subset of those were post treatment imaging studies after focal ablation. Uh, or radiation, so maybe those need to be excluded as well. But even if you exclude those 40 out of 472, roughly an 8% discordant rate, what were the types of discordants? Well, predominantly here in the bottom I've showed you the most common discordant we see is a pyrid 45 that returns as benign, or a pyrid 45 that returns as low grade cancer. That's about half of the discordants and then of course we have this subset up here in the gray of clinically significant cancer and a low pyrid reason. And what, how do we classify those after our validation? Well, incorrect pyred's designation in a subset, the purple here targeting error is still a major problem with our biopsies and the high pyrid. Uh, with benign low grade cancer, still we see a subset of these that are just high pyrid but don't have high grade disease. So how do we improve some of those radiologic discrepancies? Angela alluded to this, but I think there is a growing literature that assisting radiologists with AI or deep learning protocols can improve reader accuracy and this study that was published. Uh, the ability to accurately report pyre 4 or higher lesions improved overall in a reader concordance improved, and the reading time was reduced. So I think it's inevitable that AI will have to be used as a tool to smooth out some of the interpreting error that we see. What about costs? Well, Angela alluded to the biparametric MRI. The prime study, which was done by the Precision Study Group, has just been completed. It was presented at the EAU, but it, and, and we participated in the study at NYU, but this was a randomization of men, uh, to well, not a randomization, a comparison in the same patients of biparimetric MRI with multiparametric. I won't say a lot about it because, uh, I'm close to time. But the reality is that uh the the the biparimetric seems to have similar diagnostic accuracy even in a um even in a uh uh um a multi-center uh type of of uh setting. Uh, what about accessibility? So again, Angela alluded to this, but I wanna quickly show you that I think the idea that you can use AI to improve accessibility for MRI is really important, and it's something that Angela and Hirsh Chandarana in our group and, and Patricia Johnson have really been working on with some really exciting work. So this is the idea that you uh rather than processing imaging for a long time, you acquire a limited data set. We've just published a similar approach with using AI to interpret histology that you can take every other slice or limited data sets and then reprocess it uh using AI and you can see the image quality is equivalent between the 33 minute scan and the 38 2nd. This is an example of a reduced or rapidly accelerated DWI uh at the bottom and you see similar quality and detection. So just to give you an idea of what that means, if you do a conventional multiparametric MRI it's 30 to 45 minutes, a biparimetrics about 15 to 20 minutes, this type of rapid screening exam is 5 minutes or less, and that could certainly improve accessibility and reduce cost. The other thing that Hirsch's group is working on is low field imaging, the idea of using a low field scanner, acquiring lower resolution images, but letting AI query than the uh raw data to improve the resolution, and this is an example of that, uh, from their work which shows that this is a T2 weighted image obtained at a low field, 0.55 Tesla, and then reprocessed with AI to improve signal intensity. So inclusion, where do we sit with prostate imaging in 2024? Growing utilization globally, it's, it's certainly here to stay, uh, growing use in surveillance, treatment planning, selection for focal therapy. I think I didn't touch on it. I'm gonna touch on it in our next session about how PET maybe could be combined with MRI to improve diagnostic accuracy. We know the quality still varies widely, and I have a concern that we've gone through a bell-shaped curve. Early on in the learning curve, everybody pointed out, well, you know, the, the inner reader observability is, is really bad, um. People got better and better and better and we reached the top of the bell shaped curve where a lot of people were reading MRI as well then it infiltrated the community and everybody's doing MRI and even at our own institution we have more people reading and I think the quality of the readings is slowly declining. I think the thing that's gonna buttress that perhaps is the AI approach I hope but but it is a concern and cost remains a major concern. Uh, and so I, I, I don't wanna go over time, but where can we go further? We'll talk more about MRI PET in visible cancers and, and cancer in the future. Uh, what can we predict? Tumor behavior with imaging alone this is something that our group is now starting to be interested in. Can you find radiologic signatures, molecular signatures that you see in the imaging that might predict behavior beyond just a simple pyra ed score or a Gleason score? I think they're very exciting things coming with MRI in the future, uh, and you know, hopefully if I'm still giving this talk 10 years later, uh, we'll have some new content for you. Thanks very much. Published June 26, 2024 Created by Related Presenters Samir Taneja, MD View full profile
CME Knowledge Hub
