Hello everyone. Thank you for joining us for today's webinar, Advances in Artificial Intelligence and Breast Health, sponsored by Hologic, in partnership with the RS&A. This webinar is one of many ways RS&A is reaching the radiology community to provide education and resources. Today's webinar will be recorded and available for free in the RS&A Online Learning Center, as well as the RS&A's YouTube channel. I'm Tom Geisbers, and I'm the Senior Manager of Medical Relations at Hologic. Hologic empowers people to live healthier lives everywhere, every day, through its extensive portfolio of innovative medical devices for diagnostics, surgery, and medical imaging. I'll just jump to this disclaimer slide for a second here. This presentation includes a pre-recorded webinar, followed by a live Q&A with our speaker, Dr. Terri-Ann Gazinski. Please use the question panel on the left to submit any questions you have. The chat panel is also available for sending instant messages, but the questions you submit will go into that questions section, and that's what we'll be choosing these questions from. Now I'd like to introduce our speaker. Dr. Terri-Ann Gazinski, MD, is the Chief of the Community Breast Division at University of Pittsburgh Medical Center. As a seasoned radiologist, she has previously presented on all breast imaging modalities, including stereotactic biopsies, ultrasound-guided biopsies, MRI biopsies, breast ultrasound, MRI, and digital breast tomosynthesis. She has presented throughout the United States and other countries, such as Italy, Japan, and Korea. Her clinical interests include automated breast ultrasound for dense breast tissue, an extent of disease in newly diagnosed patients, and all types of breast biopsies. And now we'll begin the webinar. Hello and welcome to Advances in Artificial Intelligence in Breast Health. I'm Dr. Terri Gazinski, and I am currently the Chief of the Clinical Breast Division of the University of Pittsburgh Medical Center. I'm going to give you a little bit about my background before we start. I went to medical school at Pennsylvania State University, and then I did a radiology residency at the University of Virginia, and did a breast fellowship one year at the University of Virginia. I then went to several private practices and eventually ended up at the University of Pittsburgh Medical Center, where I'm now the Chief of the Clinical Breast Division. So today, what we're going to talk about is artificial intelligence, really trying to understand it a little better in our field, some of the applications of how we can use artificial intelligence in breast imaging. And in that realm, we're going to talk about how we can utilize synthetic images, and then also some of the advancements with utilization of artificial intelligence to allow us to decrease the overall number of tomosynthesis slices that we have. And then we'll finish up with genius AI detection, which is a new and exciting part of what Hologic offers. And then we'll discuss the future of artificial intelligence. So I think one of the first things that I want to do is just pause and have all of us remember why we do what we do every day. Really, we do it for our patients. As we know, you know, to do breast imaging, you really have to have the passion to really care for the patients. And it's a very special field, and I feel very lucky and blessed to have worked in this field now for over 20 years to help those patients through the journey of breast cancer. And, you know, every day you are reminded what an impact you can make in another person's life, just with a handhold, just with a hug. And that's really why I feel breast, you know, imaging is obviously so near and dear to my heart. And so we just take a moment to, you know, really acknowledge, you know, all of our patients who I feel like sometimes they've helped me more than I've helped them along the way. So the first thing we're going to do is talk about, really, what is artificial intelligence? And the idea, you know, to some is certainly overwhelming, certainly scary. Some people feel like, you know, what is artificial intelligence? And I think one of the big things we need to understand is that, you know, we need to really understand what it can do for us and how it can help us so that maybe it can allow us to make better decisions to perform our work better. So artificial intelligence really is helping machines to think like we think, like humans think. And there's a whole variety of what you talk about with artificial intelligence. And most people, when they talk about artificial intelligence, are talking about machine learning, where we train a machine to get better at a task, trying to allow it to make, you know, help us make good decisions. And what's exciting is now we're into the realm with the Genius AI product from Hologic of deep learning, which really allows more layers and more precise figuring out exactly where abnormalities are. And it's a whole lot more information. And I think that's where it's really exciting to get into more and more of the deep learning to, again, assist us in our diagnoses. So, again, deep learning really allows the machine, the algorithms, can really detect patterns that we can't see as humans because there's so much information coming at us. And so the deep learning takes the pixel by pixel and really deciphers as opposed to machine learning, which oftentimes just looks at overall characteristics. So it really is the next level after machine learning. And so, you know, why is it important? I think, again, at the end of the day, what is our goal as breast radiologists? We want to find every cancer that we can, of course, but we also want to do it with a reasonable recall rate and reasonable false positives. Because, you know, that's one of the things that we often get criticized is, you know, we call all these patients back, we cause a lot of anxiety, we do all these extra workups, and we don't really, you know, have a lot to show for it. And so if deep learning is able to help us really find better cancers with improved sensitivity and allow us fewer false positives, that's going to, again, improve our specificity. That's really the win for deep learning. So the idea is that it looks at a group of pixels. So it actually looks at, you know, all these little pixels, and it can help determine is the lesion more likely to be cancerous or not. The other exciting parts are then it can look at how is it responding to different treatments? Is the cancer, is it actually responding to this type of chemotherapy? Or maybe we need to switch the patient to a different type of, you know, medicine, different regimen that will allow better response to treatment. And then the last part, which I think is really where we're heading with a lot of this AI, is looking at really the genomic properties and really understanding, you know, why certain tumors in certain patients act a certain way because of their specific genetic makeup. And I think that's really where a lot of this is going to explode in the future as we get more and more sophisticated at being able to determine, you know, again, why if we give two people the same treatment, one responds and one doesn't. So I look at it as not really one against the other, but I look at the machine and man working together for the best for the patient. And the reason is, you know, as radiologists, we get a lot of information and data thrown at us consistently. And so, you know, we really have to be able to process and figure out what is, you know, benign, what is cancerous pretty quickly. And so that is where I think artificial intelligence can help us make better decisions for our patients. So the other thing is, you know, in our work, you know, I was very lucky and blessed to, you know, be one of the early adopters of tomosynthesis, and that was a great boost for us. However, it definitely slowed us down. And, you know, radiologists who were used to doing a certain volume of work, we quickly found that we were not able to keep that up once tomosynthesis was incorporated because it was just more data, more information, more slices, more images, more images coming at us. And so we want to see, can artificial intelligence help with our efficiency? And again, overall, our performance, the performance of the radiologist is obviously very key in how our patients do. So if artificial intelligence can help, that's going to be wonderful for us. And then looking also really at globally, you know, not just necessarily in our little neck of the woods, but looking at, you know, the other countries, looking at the United States, looking at places that maybe, you know, how can artificial intelligence help us for the subset of populations take better care of the patients overall? And really, you know, helping our patients take the best care of themselves and helping them understand maybe what is in their best interest. So with artificial intelligence, I think it allows us an opportunity to look at different ways to utilize it. And it may not be the same for every center. Certain centers obviously have their own little nuances in how they want to look through their workflow, take care of their patients. Are they doing anything for the patient? Are they doing any sort of risk assessment? How are they evaluating the breast density? Are they talking to patients about what they can do to, you know, take better care of themselves? And I think artificial intelligence has a role that it could play in this realm. So some of the challenges of artificial intelligence is, you know, we need to figure out how to make it flow into our current workflow. Because again, as radiologists, we want things to be easy. We want things to be efficient. We get into our patterns. And so we have to make sure it's not going to create an extra step that is going to slow us down or detract from our usual patterns of working. Also talking about making sure interfaces with your PAC system. Because again, as good as the information is, again, if you have challenges with your PAC system and integrating into your PAC system, that's kind of a deal breaker a lot of times of being able to utilize it. And so again, understanding, you know, how do you market it? If you have these features, how to best, you know, let the patient population know, let your patients know, so that they can, you know, get, you know, the improved care. And again, make all radiologists better. That's the goal, you know, if all the breast radiologists, you know, all got better throughout the world, you know, that would be wonderful. We all know, you know, in the breast realm there, there's definitely a big need for additional breast radiologists to be able to take care of patients. And, you know, we know a lot of people who shy away from breast imaging for a variety of reasons. So again, if this makes it easier, maybe we can encourage some people who maybe had an interest, but were nervous about, you know, breast imaging to allow them to reenter. So the idea also, again, is to certainly, you know, decrease the misses that we have. If artificial intelligence can, you know, help us in that realm, that's obviously a great benefit. So looking overall at breast density software, there's been so much, you know, talk about how do we determine breast density? You know, what do you use in your practice? Is there a consistent way to evaluate breast density? And, you know, I always chuckle at myself when I read a mammogram one year and say, I give it a heterogeneously dense. And then the next year I decide in my head, I'm going to give it a scattered. And then I look last year and see who read it as heterogeneously dense and it's me. So I'm sure, you know, some of you may have had a similar experience. And so if we had more of a consistent way of really evaluating breast density, allowing us to see it quickly, I think that again would be a very good positive thing for our patients. And so there is Quantra, which is, you know, a product of Phylogix, which is very good at evaluating and determining breast density. It's very reproducible, which is, again, a very important thing. Again, better than radiologists. I think I, you know, read one study and it said we only agree with ourselves, I think one third of the time in terms of breast density. So not great results for sure. And then even between radiologists, it can be a challenge as to who's calling what density. The wonderful thing also about Quantra is that it's compatible with both the 2D images and also the tomosynthesis images. And so here's just an example talking about Quantra products. And the Quantra 2.2 is the newer version of the breast density. And the old version was based on the volumetric evaluation of breast density. And the new 2.2 takes into consideration pattern and texture. And so the reason we were going to talk about why that change is really because of the change of the Bi-Rads. And the Bi-Rads, you know, Edition 5 came out and they really wanted to focus more on pattern and texture rather than overall volume. And so, you know, Quantra really aligns with how Bi-Rads has moved away from the volumetric assessment into the pattern and texture because they felt that was more important when you're evaluating and trying to figure out, again, you know, whether or not we're able to identify lesions on a mammogram. So what we also really want to stress with the Quantra software is that it's immediately available on the workstation. So you don't need a separate server. You don't have to worry about, you know, any of those issues that you may run into, but immediately you're able to see the breast density. And so, again, understanding and really appreciating how breast density, you know, can impact your reading. So the Quantra is, you know, objective, reproducible, allows you quickly to see, you know, results, which are very nice on the workstation. And so in terms of pattern and texture, people were trying to figure out, you know, is it overall density, just the volume or really the pattern and texture? And I think most of us who read really understand the difference that pattern and texture can have and how that can impact whether or not we're able to identify a lesion. So here's an example, and there is a little cancer here seen in the superior aspect of the breast. And you can see both breasts have, you know, a 25 to 55 percent, you know, breast density if you were to do it in this, you know, second quartile by the old edition. And you can see, though, more of the breast tissue and density is centered right over where this cancer is. So, again, if we're going by volume, they would be the same. However, if we're identifying and evaluating by pattern and texture, we can see that it's going to be more challenging to find a cancer in this focused area of breast tissue. So, again, trying to figure out ways to help us streamline our workflow and really give patients, you know, the information about their breast density so that they can make better decisions for their care. The nice thing also about Quantra is it can be integrated into the mammography reporting systems that we all have in our practices. So, now we're going to transition a little bit and talk about tomosynthesis with a synthesized 2D imaging. And this, for me, was really exciting when we started talking about synthetic views. And so we all know that breast tomosynthesis obviously performs very well. It was a great advancement. And, you know, I've been around for a while, and I read film screen, and I went through film screen, and I went through the digital transition, and then I went to the through the tomosynthesis transition. And tomosynthesis was really wonderful, as we know, because it decreased our recalls and improved our cancer detection. And, you know, there have been studies out there, of course, that have been published, you know, in JAMA, in many of our journals, for sure, that show the efficacy of tomosynthesis. We know it is superior with 2D than just 2D alone. And when you do both the 2D and the tomosynthesis, again, you know, it's going to be more accurate for our patients. We can see areas definitely better with tomosynthesis, again, just because we're taking away the overlying tissue. So those areas of distortion, really, we're able to see much better. And, again, helping all radiologists across the board really perform better. So, again, you know, with tomosynthesis, you can have up to a 40% increase in cancer detection rate, invasive cancer detection rate, and then overall cancer detection rate up to 27%. And, again, recall rates typically go down. And, you know, my own personal experience is that initially my recall rate went up a little bit with tomosynthesis, because we are seeing so much more, and we really had to kind of reset what we were recalling. So initially, mine went up, and then it did go down a bit. And the nice thing about tomosynthesis is that across all breast densities and age groups, it does better. And so some of us who went through the digital, you know, transition, initially we were focusing more on the dense patients, but we know tomosynthesis does better in all breast densities, which is, again, I think really something that we need to recognize. And so where we started was back, you know, once the Hologic was approved back in 2010-2011 by the FDA, it really started with a single compression where we did a tomosynthesis scan first, the grid was retracted, and they did a tomosynthesis sweep. And after that, there was a 2D exposure where there was an exposure with the grid. So that was what we called the combination mode. And so the next step after the combination mode was the synthesized view, which, again, I think really was an advantage for our patients. So what we did is we performed the standard scan, and it was quicker than doing the combination mode, which was a little bit longer because we had to do the tomosynthesis sweep and then do a 2D exposure. So with the synthetic view, we get our tomosynthesis sweep, we take our 15 projection images, and we turn those into tomosynthesis slices. And then we create from those tomosynthesis slices, we create a synthesized 2D image. And so together we use the tomosynthesis slices and the synthetic 2D, and that allows us to get a lower dose examination for our patients. And this is really available in any view. So you can do it in XCC views, full laterals, whatever projection you want. And so now what we were able to see is with the synthetic view, there were some advancements from earlier images, which were called the first generation. Then we went to the second generation of the synthetic view and had increased resolution, which we're gonna discuss now. So being able to quickly identify the synthetic view is really easy. There is something, you know, on the image that tells you that it's an intelligent 2D image. And the idea is that you always read the synthetic view with your tomosynthesis slices. So it's not, you look at one or you look at the other, you look at them together and it's a full study. And that allows us, that marker to tell us that it's an intelligent 2D, tells us that there we know it's not the full field digital image, but it's a synthetic view. And the idea is we use the synthetic view instead of the full D, full field digital mammogram. And the reason is we can really allow us to compare current and priors. And it's gonna give us the important details from our tomosynthesis slices. And again, we use them together, the synthetic view, tomosynthesis view, we use together. So when we compare dose, because of course, you know, patients initially, when we started doing tomosynthesis, people were concerned about how much dose am I getting? Is it gonna be too much? And really what is nice is with the synthetic view is that we're able to decrease dose significantly. And so both doses are well within limits of what we're allowed to perform on patients, but the synthetic view does really decrease our scan time and decrease our dose. So looking at how does it perform? We know that using the 3D mammogram and a synthetic 2D does better than a 2D alone. We are increasing our cancer detection and again, reducing our recall rate. So again, for all those reasons, we feel like this is really a good direction to go for our patients and utilizing the synthetic 2D view. So just as a comparison, when you look at the images, you can see the 2D, then the tomosynthesis slice in the middle, and then the synthetic 2D. And so here is the lesion that I think we can certainly appreciate much better on the tomosynthesis slice, which we expect, but look how well we can see it on our synthesized 2D view. It really does stand out. And so there have been, as I talked about different versions of the Hologic synthetic view. The Intelligent 2D is the newest version. And that really, the key to that is that it is a 70 micron resolution, which is an improvement over the first generation, which was the C-View software, where the resolution was 100 microns. So many of us, myself included, started with C-View, and then we transitioned to the Intelligent 2D. They both serve the same purpose. And again, decrease in exposure, decrease the time that the patient is in the room and the scan time, and then it allows us really to get good information from our tomosynthesis slices. So again, just comparing, I think the big thing, and I'm gonna show you some examples, is the resolution. The 70 micron is better than what we started with the 100 micron. And again, the difference in scan time is the same, the angle's the same, reconstructed time, all of that is really the same. So it's really the resolution. And so we're gonna show you the advantages in skin line and the demeddling, and just really the overall, what we call the more natural look. We also have improved the visibility of the calcifications with the synthetic 2D, the Intelligent 2D over the C-View. The other really nice thing is the smart mapping feature, which we'll discuss a little bit to allow you to quickly identify where lesions are on the mammogram. So again, the Clarity HD, the high resolution Intelligent 2D is the 70 micron. And so it's gonna give us a really nice picture and image for our patients. So again, just the comparison. We're now getting back to our original resolution with conventional full-field digital, which was 70. With the C-View, we went down to, I mean, we increased to 100, and then now we're back down to our 70 micron, which is gonna give you a crisper image. So the big things I think that those of us who went through the transition really notice are the skin line and the contrast. And I'm gonna talk a little bit about my experience after I show you some examples, but those were the things that I really picked up on initially. And then also just the speculations and the distortions. And the reason that is more visible is again, because of the artificial intelligence algorithms, which were improved with the second generation. So this is a picture that really highlights the overall image quality. So if you look at the Phantom, comparing the first generation C-View with Intelligent 2D, you can see how well you can see the resolution. Even the numbers here are more clear and crisp. So that I think speaks volumes to the improvement in the resolution. And again, just a nice example showing you how you can really identify some of these cuprous ligaments and the fatty tissue and all that fibrous tissue is more sharp in the Intelligent 2D software. Here's an example showing calcifications. And again, really highlighting the difference in the number and the morphology of the calcifications is just clearer and more crisp in the Intelligent 2D software. So the idea is that we're gonna reduce background noise, which is gonna give us a better picture, but we also don't wanna lose the clinically relevant information. And then also blurring. Again, the imaging process techniques allow us to see better overall image quality. And so here is an example, again, showing how sharp things are. And initially there was question, were there calcifications? You can see calcification here and here. And there was a question as to whether there were additional calcifications, but on the Intelligent 2D, we can clearly see there are no calcifications. So image blur. Again, just looking at the definition of the breast parenchymal tissue, you can see more sharp and clear. And then contrast in terms of overall contrast in the skin line. And that is definitely something that was a huge improvement. And some of us struggled with the C view just seeing the skin line and you're gonna see an overall improvement. And this is one of the examples that we would often see this gray area where the skin line was. And it was hard to sometimes tell, is the lesion in the skin or is it just under the skin? With the Intelligent 2D, we can clearly see how sharp the skin line is. So that really took us back to what we were used to seeing. And here's just a highlight example. Really, we're losing where the skin line actually is. And with the Intelligent 2D, you can clearly see how sharp and crisp the skin line is. And just the overall contrast. When you look at the overall contrast in the C view versus the Intelligent 2D, you see how much darker the fat is. And that's really more in the skin line and the skin line is more in the skin. And that's really more in alignment with what we were used to with our full-field digital images back in the day. So artifacts. Artifacts are something we obviously struggle with, trying to see around calcifications, clips, surgical, anything that's put in there artificially. So the synthesized Intelligent 2D really decreased that. And here you can see what we used to see oftentimes with the C view was the streak artifact, we called it. And you can see how much it decreased in the Intelligent 2D software. And the same thing, just a nice example of another calcification, reducing the artifact around it. So then moving on to the margins of masses and speculations. Again, the Intelligent 2D had elevated software and algorithms which allowed us to really improve the visibility of lesions. So here you can appreciate when comparing the C view, this nodule, you can see more of the border around this nodule. And speculations, again, very sharp and clear in the Intelligent 2D. Here it's a little bit more challenging to see in the first-generation software. And calcification cluster, again, very difficult to see in the C view, and we see them a lot more clear in the Intelligent 2D. So overall, basically, better resolution, the noise is reduced, blur line, the skin line looks so much better, and just overall contrast. And again, those bright objects in the artifact. So again, better for our patients, easier for us to find the lesions. So for my experience, what I did in our practice is we were trying to figure out how do we transition from the first generation to the second generation? And anytime there's a change, obviously, we as radiologists get very set in our ways for sure, and sometimes it's hard to sort of push us out of our comfort zone. And so what we decided to do was to go really all at once to transition our unit so that we had sort of a start and stop point so that we got used to it pretty quickly. And again, sometimes it takes time, and sometimes you may not be in the situation where you can sort of turn all of your machines on at the same time, depending on your practice. But I think the sooner you're able to do it, then the quicker you're able to start adjusting your eyes. Now, I will say our transition from the combo mode to the CV was a little bit slower. We were a little bit uncomfortable. We weren't sure. We were like, are we gonna be missing things? Do we not feel like we need a true 2D, full-field digital image? And so for a while, we did do the combination mode for our screening patients. And so then we decided to look pretty critically and say, are we missing anything? Are we calling something that maybe we're not seeing on both images? And so we did a study. We had each of us look at the images without the full-field digital, just using the synthetic view, and then together and recognizing, were there any lesions that we weren't seeing? And so we were able to quickly identify that really we were not changing our calls. We were calling the same things. The 2D was not really adding any additional information, or we were not losing any additional information by not doing it. So we then transitioned into just doing the synthetic view for our screening mammograms and for our diagnostics now too. So now I'm gonna switch gears and talk about the 3D Quorum, which is, again, super exciting. I was very, very excited when I first saw this technology, and I wanted to get my hands on it. And I remember saying, how soon can we get it? And I'm gonna talk about some of the reasons why and the advantages of the 3D Quorum. So again, using artificial intelligence, we are gonna be able to create the six millimeter, what we call smart slices. What's really nice is the smart slices are gonna have overlap, which allow very smooth scrolling and not losing any information. Because again, at the end of the day, we don't want to make something better or make it more efficient for us, but yet we're losing information. So the idea is that the AI, again, is gonna keep all that data and preserve all the information we have for our abnormalities, but yet give us a decreased number of slices. That's really the advantage, in addition to reducing our file size. And the idea is the smart slices are really meant to take the place of our one millimeter tomosynthesis slices. So overall, I wanted to look at, can this really help read times? Again, the big complaint about tomosynthesis is it takes longer for us to read. So if we could decrease that read time, would that not be a huge advantage? And then also read your fatigue. I myself know, if I have a tough case at the end of the day, it's often a little bit harder to get through as opposed to the beginning of my day where I'm more fresh. So we're gonna talk a little bit about the 3D technology, the 3D Quorum technology. And what we do is we take the tomosynthesis slices, we use artificial intelligence to really highlight areas that are of interest. And that's gonna be highlighted on our smart slices. And it's gonna really, again, as I said, preserve the abnormality. So we take six tomosynthesis slices and combine them into one smart slice. And that's how we get our smart slice. So we're gonna talk about, is this slabbing? People said, well, is this a slab? And this is not, because traditional slabs are not FDA approved. It doesn't reduce the number of images to review. So it's not gonna help with your read time and it's not using AI. Smart slices do all of that and more. So two thirds reduction in the numbers of slices that you have to look at. Overall file size is gonna be reduced up to 50%. And that's gonna save on average, if you're reading eight hours, about one hour out of your day. And again, how you spend that time or what you're gonna utilize. I think all of us would love to have a little bit more time in our day to do whatever, whether it's reading more cases, taking care of other patients that we need to take care of. So what we do is we take six slices and we combine them into one slice. We then use an overlap of three to allow us to generate the smart slices and really not lose any information. Because we don't want to be taking all that information, combining it and not having some overlap, which allows us to be very confident that again, we're not losing information and we're preserving all of the features that we need with our smart slices. And there's a white paper and reader study that was done looking at really, are they comparable? Are they, we don't wanna do this and realize that maybe radiologists aren't performing as well. And then also recall rates. So on both areas, we ended up doing better and equivalent. So that's really important is again, the average ROC curve went up for people who use the 3D quorum versus the standard slices. And again, the recall, we did better overall. So again, we don't need the one millimeters, we can just transition into the 3D quorum. So now some examples, just highlighting how wonderful and beautiful these images are. And you're able to really again, appreciate the abnormality better in the smart slices. And here's a blown up view. Again, looking at the definition and the speculations of this mass, you can see them to better effect on the smart slice as opposed to the one millimeter tomosynthesis slices. And again, the idea is that the artificial intelligence allows us to highlight those areas that are abnormal and gives larger weights to those locations in the algorithm so that you will then not lose the features of the abnormality and the speculations and the calcifications. So here's a nice example, also showing the comparison from the one millimeter tomosynthesis slice to the six millimeter smart slice. And the same thing with the calcifications, we are finding that we are able to see and visualize calcifications to greater effect on the smart slice as opposed to the one millimeter slice. So my experience with 3DQuorum is we transitioned immediately from the, I should say immediately, which was, I think, two weeks. For two weeks, we sent both the one millimeters and the six millimeters to the workstation because again, we wanted to do a comparison and make sure that we were comfortable, weren't losing any information. Honestly, about after two days, I was ready to just say, forget it, just send me the smart slices. But I wanted the others in my group to be as comfortable as I was. I would encourage you to just make the transition, which I told others is, feel, please just think about just going right to the 3DQuorum. Because what happened was I had the one millimeters and I had my smart slices. So I felt like I had to look at both of the datasets, which in fact was not saving me any time. So I'm like, jeez, oh man, I'm actually doing more work than I was expecting. So we then quickly transitioned. The other thing that I did do with my 3DQuorum is I slowed my CineSpeed down because there were fewer slices. I wanted to make sure, and there's more information, I felt like on every single slice. So I slowed my scrolling speed down so that I was able to capture all of those abnormalities. But again, quickly was able to transition to the smart slices. So now I'm gonna show you a couple of cases. This was a 54-year-old female and she has a very dense breast here. And there's an abnormality here in her left breast and lateral aspect. And just comparing the smart slice here on the left with the one millimeter, you can see how much sharper the area is and the spiculations. And there's a blown up view, again, just highlighting how well you can see these lesions. And here's her medial lateral blight. And again, a blown up view of the area of abnormality. And again, the skin line definition is even better in the smart slice. And that turned out to be invasive ductal carcinoma. Our next patient is 58 years old. And here is her mammogram. And on her mammogram, we saw an area in the medial breast. And again, comparing SmartSlice and looking at the overall texture adjacent tissue, I think you can see it's a greater effect on the SmartSlice than you can on the one millimeter slice. And here is another projection of the same lesion. And here's the lateral projection, again, showing and highlighting, really being able to appreciate the mass and the margins. And that was an invasive lobular carcinoma. Next patient, 72 year old female, came in for her mammogram. And we were able to see this little nodule here. And I'm gonna show you comparing side to side here, again, looking at the SmartSlice versus the one millimeter. And again, on our medial lateral blanks, blowing it up, looking at being able to see the margins to greater effect as where they're a little fuzzier on the one millimeters. And that was an invasive ductal carcinoma. Our next patient is a 48 year old female. And her abnormality is in the right breast. And you can see this abnormality there. And again, highlighting the difference and the improved resolution with the SmartSlices. Invasive lobular cancer. Now we're gonna turn to SmartMapping technology. That uses the ability to quickly identify where a lesion is on your tomosynthesis slice relative to your synthetic view. And so again, allowing you to not have to scroll through all those images and really get to where you need to be. And then the Genius AI technology, again, is what I'm very excited to talk to you all about because it does use deep learning. So we're gonna find more cancers, gonna allow us to be more efficient. And the difference in this type of artificial intelligent product is that it is a concurrent reading workflow. So the images come up as you're looking at the case, as opposed to traditional CAD, which a lot of us are used to, where you hit a button and you can see it at the end of your case. There is a study that was done, of course, and it did show that for all radiologists, regardless of their experience, all radiologists did better. So again, this is important information that it can help all radiologists perform better. And it was a multi-reader, multi-case study. And again, two sessions with washout period, done very well. And it did show, again, better performance for all radiologists. The nice thing about this Genius AI technology is it can be used different based on your practice. There's a lot of information that's available. And so how you choose to use it is completely up to you. It gives you things such as a lesion score, an overall case score, a reading priority index, indicator, I'm sorry, a case complexity index, and then a read time indicator. So again, being able to implement this into your practice in some fashion is really gonna be key to how you utilize it. So you're able to quickly get a lesion score, which tells you, basically, compared to all the algorithms that were placed, that were utilized in this artificial intelligence product, how confident that a suspicious area represents a cancer relative to the other lesions and throughout the algorithm. The reading priority indicator also, again, if there are multiple findings, if there's something very concerning, it can give you a higher reading priority. So again, you could potentially utilize that in a different fashion so that maybe you do the higher priorities earlier in the day or while the patient's there rather than at the end of the day. And the read time indicator is also something that, again, you can potentially use in a different fashion so that if you know this case may be very complex and take a little more time, you may wanna put aside different time as opposed to early in the morning or later in the day, depending on if you batch read. And case complexity, again, if there are multiple findings. So again, you can sort those cases if you feel like maybe I don't wanna deal with multiple findings right now or at a certain time in the day. And then the case score, again, how confident the product is that there is a cancerous lesion in that case. So we're gonna finish with a couple of cases. The first one is a 57-year-old female. And here is her mammogram. And on her mammogram, she has a lesion that is marked. And you can see quickly that there is a case score given to this patient. So then it is up to you, again, to really look at it. And what I really wanna stress is through the algorithm, the idea is that we are gonna improve our sensitivity. So we're gonna find more cancers, which make us more accurate. But overall, the algorithm decreased the number of false marks by a factor of four. So again, we are used to so many marks and so many mammograms. This is really gonna help us minimize the number of marks and really pay attention to relevant abnormalities. So again, just showing you how quickly you can see a number and the area. You can have the overlay so that you can identify whether or not the area does need to be evaluated. And it will quickly also take you to that slice on the tomosynthesis stack where the abnormality is. And that is the little mark here in the middle that allows you to know where in the stack it is. And that turned out to be atypical ductal hyperplasia. The next case is a 48-year-old female. And here is her mammogram. And there are several lesions. So again, it's gonna give you a lesion score for each lesion that you will then figure out and evaluate and determine whether there is something there to work out. And again, just seeing how you can see it so well on the SMART slices as opposed to the one millimeter tomosynthesis slices. And again, pointing to the lesions, whether you like the overlay or whether you like just an error to show you the abnormality. And then again, you get to determine the workup. And the first lesion was an invasive ductal. The second lesion was a radial scar. Next patient, 58-year-old female. Here is her mammogram. And there are several areas that were marked on her mammogram. And here we can quickly see again, the lesion score for these areas. And then again, we can determine, some of these areas may have been present before so we don't feel like we need to work them up. And then other areas maybe we do wanna investigate further. So in this case, we can quickly identify where the lesion is on our synthetic view and then on our tomosynthesis slices to be able to evaluate it. And that turned out to be bilateral carcinomas. Our next patient is 57-year-old. She came in with a palpable lump, but we saw something also, I should say the Genius AI found something on the contralateral breast. And there you can see that little speculated area. And that turned out to be bilateral invasive ductal carcinomas. Our next patient is 67-years-old. And her mammogram here shows an abnormality. And again, looking quickly to be able to identify a case score that you can toggle on and off to evaluate and determine whether or not you decide to biopsy it. And that turned out to be ductal carcinoma in situ. So Genius AI, I'm very excited about how you can utilize it in your practice. I think there's a lot of opportunity to help with workflow efficiency, to help with all radiologist performance, to make all of us better, to make sure we're not missing cancers, but also decrease the number of biopsies that perhaps we're recommending. So the end of the day, I think AI is gonna play a large role for us. And I think we need to really look at it, try to embrace the advancements, try to figure out how in our practice we can make it work to the best of our ability. Thank you so much for your time. I'm now gonna take some questions and answers. All right. So we've heard from Dr. Gazinsky on the advances in using artificial intelligence for breast imaging and improving diagnostic accuracy with Hologic's newest technologies, 3D Quorum and Genius AI Detection. Thank you so much, Dr. Gazinsky. That was wonderful. We'll open up the discussion for live Q&A with Dr. Gazinsky. Again, the questions, we've had a lot of great chats and the questions can go into the questions box, but we encourage you to submit those questions if you haven't already. And Dr. Gazinsky, we'll turn it over to you. Thank you, Tom. And I just wanna also say thank you all for taking time out of your day. Now we're all busy and I appreciate RSNA and Hologic partnering to allow these events to happen and to share knowledge. So we have a bunch of great questions. I'll try to get to as many as I can throughout our Q&A period. And please feel free to definitely add questions. And the first question is, are you totally comfortable not using the one millimeter slices and relying on the smart slices with the 3D quorum for your screening and your diagnostics? And so the answer is yes. The Hologic Reader Study demonstrated there was equivalent performance between the one millimeter and the six millimeter slices. And the six millimeter slices are FDA approved to be used alone for clinical use. So definitely very comfortable with it. And one of the follow-up questions was, did I do any sort of formal evaluation? And we did. What we did was we had our radiologist actually look at the images, make a diagnosis or make their assessment on the six millimeters and write that down. And then we went and gave them the one millimeters and had them write them down. Now, again, not super scientific, but just wanted to be sure we all were comfortable before we transitioned into just using the six millimeters and getting rid of the one millimeters. And so we definitely did that, noticed we really weren't changing our readings and that's why we were comfortable. And I will tell you the other big advantage we noticed in our system personally is that because of the change in the file size, it made a big difference in our speed. And that was also something that was really, to our practice, very advantageous. Now, you all may not have some of the same issues in terms of speed and file size, but we definitely noticed a difference. And I've talked to other people, some people have said they don't have issues with slowness or too much volume, but in our system, we definitely noticed a difference. So one of the other questions I had is, is Quantra compatible with synthetic 2D? And the answer is yes. The Quantra is used with the synthetic 2D and the tomosynthesis lysis. So you can use it for that. Next question that we'll see is, a question about CT of the breast, is it coming as a good choice anytime soon with reasonable dose? And I personally know there are ongoing studies, some of which I've talked directly to people who are involved in it and they do feel it's coming along. I, you know, not sure what the timeframe is. I think, you know, anytime we try to, you know, make these improvements, there's obviously a lot of things that go into it and we have to make sure, of course, you know, we're making the best decisions for our patients. So I think it's not completely figured out. I think there's a lot of, you know, new interests now in contrast mammography rather than the CT of the breast. And so I'm not sure exactly what direction, you know, everyone will pivot to, to really decide, you know, where we're going. And then there was a question about the T-MIS study. The question is, why is all this continued effort going into the T-MIS if DBT is proven to be better in all categories of performance? Should they halt the T-MIS? And so the T-MIS study was set up back in July of 2017 and they wanted to accrue 10,000 patients and do the thorough evaluation. So we're almost to the end of that five-year timeframe. That was their sort of goal. And I think, you know, really seeing all the information that most of us have, I think everyone sort of agrees. And so I think, you know, we're sort of at the end anyway. So I think it's going to naturally just sort of play itself out. Now, that being said, I will tell you some of the latest figures that I have seen is that we have about, of all the radiology facilities in the United States, about half of them are tomosynthesis. So we still have about 50% of the facilities that are not tomosynthesis. And I think, you know, obviously there are barriers. You know, the first one, of course, being cost to the facilities to engage in, you know, moving from, you know, to the tomosynthesis. And I will tell you personally in my area, I'm kind of surprised, but there are two facilities that are still doing film screen. So I feel very fortunate that in my facility, we were, you know, one of the early adopters, as I talked about with, you know, digital breast and tomosynthesis and, you know, I feel lucky every day that I have that technology, but many places do not have, you know, some of the, even the digital technology available. And I think, you know, as a breast care provider, we need to make sure we are out there recognizing, pushing, acknowledging, educating our patients. You know, that's one of the things that we need to educate our patients. We also need to recognize, you know, as a public health problem, you know, how are we addressing this? How are we going out and educating those individuals and also our referrers, you know, the family doctors, the OBGYNs that are sending us patients. We need to really make sure we are educating them in what is happening, what is the technology and how we can provide the best care for our patients. And a few more questions. So one of the questions is, why is there, why do different radiologists read breast density differently? And I think that's a great question. And I will tell you that it's very challenging oftentimes to really get a real flavor or idea as to how dense is this breast. Many of us do very well with the very fatty breast and then the very, very dense, extremely dense breast, but those ones in between can be challenging. And I myself have found myself where I will look at the mammogram and feel very comfortable that this is a scattered breast. And I look, last year they read it as a heterogeneously dense breast. And then I look at who read it and it's actually me. So we know radiologists, there was a study done that showed we agree with ourselves about 60% of the time. So we need to understand that, you know, we still don't necessarily even agree with ourself. And so that's why some very consistent way of evaluating breast density, I think, is a huge improvement for us as, again, healthcare providers providing the best care for our patients so that they can make informed choices. You know, in the end, we really want to talk about helping the patient make choices that they can live with and, you know, the choices that make the most sense for them. It doesn't have to be one size fits all. As we know, you know, now more and more of us are trying to help patients understand what supplemental screening can look like for them, what it means in terms of, you know, their care, the pros and the cons. And so, again, educating our patients, educating our referring physicians is really part of our role rather than just sitting down, you know, reading mammograms all day. All right, let's see what other questions we have. Let's see. Okay, so one question is, during the transition from the old to the new generation of the imaging system, is there any performance degradation of the AI-CAD algorithm? And if so, how do you tackle those problems? And really, there is no degradation of the CAD algorithm. In fact, I think it actually, you know, improves. And that's the wonderful thing, you know, Hologic has invested, obviously, in making sure they really put the time and effort into getting these deep learning algorithms as opposed to some of the, you know, earlier algorithms that weren't as sophisticated. And so the more neural pathways that you have, the better it's gonna be for our patients. And that's one of the things that I really, you know, wanted to highlight is the reader study for the Genius AI, you know, it took 17 radiologists and they had almost 400 cases, and they did two sessions, you know, with the washout period. And they noticed that in all radiologists, regardless of their experience, we had an increase in the ROC curve, meaning they perform better. And we also had increased sensitivity by about 9%, which again is significant. 40 seconds, Dr. Gazinsky. Okay. And so the last thing is that it also, I think that one of the big bonuses is that it decreased the number of false marks, false positive marks, the number of marks by a factor of four, which again is one of the criticisms that people have oftentimes that, you know, CAD marks so many things that are really nothing. And we'll have to stop there. Okay. Thank you so much. I'd really like to thank Dr. Gazinsky for sharing her expertise. To our audience, thank you also for attending today's webinar. And we hope you have a wonderful rest of your day. Take care. Thank you.

Artificial Intelligence

Breast Imaging

Dr. Terri-Ann Gazinski

3D Quorum

Genius AI Detection

Synthesized 2D Imaging

Quantra

Breast Density Evaluation

Diagnostic Accuracy