 Hello, this is Tom Stavros, and today we're going to talk about ultrasound features of breast masses as prognostic biomarkers. My disclosure is that the study from which the data comes that I'm going to present today in all their extensive data was enrolled and funded by Cento-Medical Incorp. However, there will be no mention of Cento-Medical or its products during this talk. Now, when we talk about ultrasound features of breast masses, we have to question whether we ask enough of ultrasound, and I think we aren't. We could ask more of it. We're going to look at ultrasound features from a scoring system that was designed for a different purpose. And we decided to use these features to assess them as prognostic biomarkers after the fact, after it was designed, after the original data was evaluated. We'll talk more about the original purpose in the talk that follows this. And I'm going to show you in this talk that there are numerous ultrasound features that are significant prognostic biomarkers, but they are so numerous and they interact in such complex ways that it's difficult to put them all together without the help of artificial intelligence. Now, why we want to use ultrasound features as biomarkers is to get closer to personalized medicine. We want to get away from the one-size-fits-all medicine to precision or personalized medicine. Now, the types of biomarkers, imaging biomarkers that can be used are diagnostic, monitoring, predictive, and prognostic. This comes from a very good article by Olina Weaver and Jessica Leong, that is my kind of go-to article that I rely on for biomarkers. And today we're going to talk primarily about prognostic biomarkers and primarily how they correlate with tumor grade, receptor status, and of course, receptor status is used as a surrogate for molecular subtypes. Now, we do know that some other type of breast imaging features can be used as prognostic biomarkers. From the biography, we know that triple negatives tend to be round and circumscribed or indistinct, and that the aeropositive and luminal molecular subtype cancers had to be speculated. And we know that segmentally distributed fine linear and branching microcalcifications are most likely to occur. And her two cancers are a little less likely in luminal B. We know from MRI that ring enhancements suggest triple negative or high grade, and that clustered rings of non-mass type enhancement on MR tends to be caused by grade 3 DCIS or what Laszlo would call ductal adenocarcinoma and Cy2DAB. And we know that on mammography and MRI, multi-focality is most common in her two type and luminal B types, less common in luminal A, or triple negative, which are mostly unifocal. And that large bulky lymph nodes tend to favor her two or luminal B cancers. The triple negatives often skip the lymph nodes and go hematogenously. Now, in the distant past, all the way back in the 1990s and early 1990s, I noted the correlation between sound transmission and histologic grade. I also noticed the taller than wide tended to be low grade. And that the size of enlarged ducts correlated with grade, big, big large involved ducts in the surrounding tissues, or what the associated findings in byrads tend to go with grade 3 and small ducts with grade 1 and medium size with grade 2. And that the relative thickness of the ecogenic rim and the length of the spicules tend to go with low grade. So these are all things I noticed anecdotally, but really didn't have a lot of data to support except for sound transmission. Now, here's this, here's data that we accumulated way back in the 1990s. We had no idea what molecular subtypes were, whether people had positive ER or PR or R2s or KI-67s, because we didn't do any of that stuff back then. But I could clearly show that acoustic shadowing reduced the chances of being grade 3 to about 0.4 of what the overall population had, and enhanced sound transmission increased the risk that it was grade 3 by almost 2.4. So these were relative risks, not odds ratios, but clearly there was a correlation in sound transmission and histologic grade. In fact, we came up with differential diagnoses for the tuber type based on sound transmission or what we now call posterior features. Acoustic shadowing tended to correlate with grade 1 or grade 2 invasive ductal cancer or grade 1 or grade 2 invasive lobular cancer or tubular cancers. Enhanced sound transmission tended to go with grade 3 invasive ductal cancer, especially triple-negative with some larger mucinous carcinomas, with some grade 3 carcinoma in situs that didn't have a lot of calcations with medullary carcinoma and invasive papillary carcinoma. Now, more recently, some articles have come out showing some sort of correlation of ultrasound features with prognostic biomarkers. Unfortunately, these are both from outside the United States. This one is from Toronto from a friend of mine, Fred Au, and they looked at 60-sum cancers and found that speculated margin was associated with low grade and ER and PR status and shadowing was associated with positive progesterone receptor. On the other hand, they showed that microlobulated margins and angular margins tended to go with ER negative status and enhancement with grade 3. Another recent article out of China showed that luminal A cancers tend to have speculated margins with posterior acoustic shadowing and that luminal Bs tended to have indistinct margins and triple negatives. We're more likely to have circumscribed and microlobulated margins lack a thick echogenic rim and to be markedly hypochoic and have posterior sound enhancement. So, you know, the findings on sound transmission or posterior features are holding up. I now have a larger database of 653 invasive malignant breast cancers to which I have scored all the ultrasound features and which represent the basis for the data that I'm going to show you. Now, if we use ultrasound features as prognostic biomarkers, we need something to compare to. So, receptor data turns out to be a good thing to use in addition to histologic grading. Estrogen receptor, progesterone receptor, HER2 receptor status, and KI67 are the things that are used as surrogates to decide molecular subtypes. So, first we're going to look at the individual receptor correlations, then we'll put them together to come up with surrogate molecular subtypes, and we'll look at a ultrasound correlation with the molecular subtypes based on this. It's also important the latest ICCN guidelines for TNM staging use ER and PR status to either upgrade or downgrade the staging of tumors. So, it's become more important in staging to have an estimate of what the receptor data are. Now, if we look at sound transmission, which is what we noticed way back in the 1990s, this is an extremely powerful biomarker for estrogen receptor status, progesterone receptor status, HER2 status, and KI67. And if we look at these p-values, some of them are ridiculous. We can see that the correlation with ER and PR status is 10 to the minus 15th and 10 to the minus 16th. We can even distinguish ER positive, PR positive from ER positive, PR negative, which has a worse prognosis than when they're both positive. There is a correlation with HER2 status with a p-value less than 0.5, 0.5, but it's not as strong as for some of the other features. We also have an extremely powerful correlation of sound transmission or posterior features with KI67. And we can even look at the primary breast lesion and look at some of its old sound features, especially sound transmission, and get some information about the likelihood of positivity of axillary lymph nodes. Again, the p-value is less than 0.05, but not better than 0.01. So I've color-coded these. If the p-value is less than 0.001, I've color-coded a green. If it's better than 0.05, but not as good as 0.001, then I've color-coded an aqua. But you can see that all of these are statistically significant. Now if we specifically look at sound transmission versus ER status, we've got a p-value less than 10 to the minus 16th, or p-value equal to 9.8713 times 10 to the minus 16th. We can see that if we have enhancement, the relative risk of it being grade 3, or ER negative, I'm sorry, is 3.8 times more than it is to be ER positive. On the other hand, if we have acoustic shadowing, it's 3.6 times more likely to be ER positive. And so if we look at this case, clearly there's acoustic shadowing on the left, there's enhanced through transmission on the right. The likelihood is that the shadowing lesion is ER positive, and the lesion with enhanced sound transmission is ER negative. If we look at the correlation of PR status, it's just a teeny bit less strong, but it's still p-value equal to 3.6 times 10 to the minus 15th. Again, enhancement increases the likelihood of being PR negative by 3.8 times, and shadowing increases the chance of it being PR positive by 2.3 times. And again, if we look at these two lesions, the one on the left has enhanced through transmission, the one on the right has weak shadowing. It's more likely that the shadowing lesion is PR positive, and that the lesion with enhanced sound transmission is PR negative. If we look at a combination of ER and PR, we can see that enhancement makes it 2.7 times more likely than its ER positive PR negative, whereas shadowing makes it 1.4 times more likely that both ER and PR are positive. Now if we look at HER2 status and sound transmission, we do have a significant p-value, but it's nowhere near as powerful a correlation as there is with ER and PR. But if there is a partial enhancement, we found that it's 2 times more likely to be HER2 positive, and if they're shadowing it's 1.8 times more likely to be PR negative. So what's interesting about these HER2 lesions is we found that they're quite wide, and that the sound transmission behind them tends to vary, and so we found out that if you have a mixture of shadowing and enhancement, it about doubled the chances that it was HER2 positive, especially if it was a lesion that was much wider than tall. So in our scoring system, a complete enhancement is a 0, normal sound transmission is a 1, a partial normal and partial enhanced is a 2, partial enhanced and partial shadow is a 3, partial shadow is a 4 and complete shadow is a 5. The HER2s would be these 2s and 3s, a mixture of enhancement with either normal and or partial shadowing. And so these are two HER2 lesions. In fact, these are some of the reference keys, the images that we use. Here's partial enhancement, partial normal, and here's partial enhancement with an area of shadowing right next to it. Both of these were proven HER2 positive cancers. Now, Ki67, again, these are color coded, so that if it's aqua, it's a p-value of less than 0.05, and if it's green, it's a p-value of less than 0.001. So we can see that just about everything correlates with Ki67. Now, if I say negative, that means that the lower the score, the more likely it is to lead to a high Ki67, and the higher the score, it leads to a lower or correlates with a lower Ki67. We can see that sound transmission is the most powerful, it has the best p-value, but there are a lot of other tremendously significant findings on Ki67. So again, in our sound transmission scoring system, enhanced through transmission is a very low score, it's a zero, whereas partial shadowing or mixture of shadowing enhanced through transmission, as we're seeing here, would be a four. So basically, the left one is likely to have a very high Ki67, this one in fact had a Ki67 of 60, and the right where they're shadowing is more likely to be a lower Ki67, and in fact this one had a Ki67 of 5. Now, what's interesting is we can look at the index lesion in the breast without even looking at the lymph nodes, and ascertain something about the likelihood of positivity of lymph nodes. And so what I did is when we looked at the lymph nodes, we tried to make it correlate with the Z11 study. So we had either negative lymph nodes, one or two positive lymph nodes, or three or more positive lymph nodes. And this kind of perplexed me initially because the things that would normally correlate with high grade, such as enhanced through transmission, did not necessarily correlate with positivity of lymph nodes. But if you think about it, it kind of makes sense, because we know that triple negatives tend to skip the lymph nodes and go straight hematogenously. And so since enhanced through transmission goes with grade three, triple negative, it's possible that those are so aggressive they go straight hematogenously without going through the lymph nodes. Anyway, shadowing increases the risk of positivity of the lymph nodes, whereas sound transmission decreases the chances of three or more positive lymph nodes. So you can see that there's about a two and a half fold increased risk of three or more positive lymph nodes when there's acoustic shadowing versus enhanced sound transmission. Now, we talked about histologic grade, and that was the first thing I noticed way back in the 1990s. So if we look at sound transmission, or in the current version, edition five or by reds, we'd call it posterior features, we can see that all of these have a very strong correlation with sound transmission versus grade. So distinction grade three from other grades, 10 to the minus 17th, distinction grade two from other grades 0.00011, and distinguishing grade one from other grades 0.000196. So we can see that if we look at enhancement, the relative risk of it being grade one or grade three, I'm sorry, is 3.5 times more likely than grade one. So enhancement goes with the increased risk of grade three with a relative risk of 3.5 times. On the other hand, acoustic shadowing makes it 2.9 times more likely to be grade one than grade three, and it makes grade 3.35 times as likely as grade one. Notice that there's a much bigger difference between grade three and grade two than there is between grade two and grade one. Grade one and grade two are more similar, and I'm going to show why at the very end of this talk, but this does make sense when we look at the genetic assays that are done in grade two cancers. If we look at the feature scores for grade one and grade two with 95% confidence intervals, we can see that indeed grade three is way, way, way lower scores, and remember lower scores go with better sound transmission, higher scores go with more shadowing, so we can see that grade one and grade two shadow more, and there's a higher likelihood of enhanced through transmission and very low scores with grade three. So we're looking at these two lesions. The left shadow is intensely. The right has enhanced through transmission. The right is likely to be the grade three. The left is likely to be the grade one. Shadowing goes with grade one, enhanced through transmission with grade three. Now what about molecular subtypes? Well, this is the ultimate in personalized medicine because they respond differently to treatments, and treatments can be customized for the molecular subtypes. Now the best way to define molecular subtypes is by genetic assays, but many times people use surrogates, and so we've defined surrogate definitions of the molecular subtypes using ER, PR, HER2, and KI67. We've already shown that sound transmission correlates with all four of these things extremely strongly, so it should not be surprising that if these are the four components for molecular subtypes, sound transmission is going to work to some extent, at least, for molecular subtypes. So if we talk about the subtypes, there's four. There's luminal A, which is defined as ER and or PR positive and HER2 negative with a KI67 of less than 13, or ER and or PR positive with a HER2 positive. Now, I'm sorry, that's a little bit. The ER2 will always be negative. In luminal B, you've got ER and or PR positive and HER2 negative, but a KI67 greater than 13, or ER and PR positive with a positive HER2. Now, I will say something about the cutoff value of KI67. In Europe, they're using St. Cal in 2013, which uses a cutoff of 20, and ASCO and CAP, which is the Pathologic Society, are using 10, so there are different cutoffs that people use to distinguish A from B. Now, some people are also using grade, but the best way is to use the KI67. The HER2 positive group is ER and PR negative, but HER2 positive, and the HER2 positive is defined as either an IHC immunohistochemistry 3+, or an IHC 2+, with a re-amplified HER2 fish, either one of those counts. And then a triple negative, obviously, is ER and PR negative, or HER2 negative. Now, we kind of talk as if triple negative is one thing. It's now been shown from genetic assays and prognoses that there are really multiple different subtypes of triple negatives. If we look at the survival curves, luminal A has a much better survival curve than any of the others. You can see that the others have much worse survival curves, although I will say that HER2 is the manifestation of personalized medicine. Without the HER2 blocking drugs, the prognosis of HER2 is terrible, but with the right combination of HER2 antibodies, the prognosis of HER2s has gotten much, much better. In general, the luminal A's make about more than a third, almost 40%. Luminal B's are similar. Triple negatives on HER2s are usually 10% to 15%. Now, luminal A's are the most common subtype. They arise from the epithelial cells of the assymus of the Tdlu. They're unifocal about three-quarters of the time, multifocal about a quarter of the time. They're very positive ER and PR and HER2 negative. They have a low Ki67 of less than 14 or less than or equal to 13, and they respond to tamoxifen and aromatase inhibitors. And they have the best long-term survival, but they do tend to recur late, so you'd have to follow them for a long time. And when there are metastases, they tend toward bone metastases. A typical path report for a luminal A would be grade 1, irregular shape speculated. ER 100% positive, PR 100% positive, HER2 negative, 0 or 1 plus, and then Ki67 to 5. Luminal B's make up a slightly smaller percentage than luminal A's. They also arise from the epithelial cells of the assymus of the Tdlu, and therefore they're AADs, assynure adenocarcinones of the breast. They're more typically, they're more commonly multifocal in about half of the cases. The ER is positive, but more weakly so. The PR is positive, but more weakly so. The HER2 is negative and 70% positive and 30%. The Ki67 is going to be over. It's going to be 14 or higher. There is a higher chance of P53 mutation rate, and P53 is a tumor suppression gene. So when the tumor suppression gene is faulty, the normal tumor repressive mechanisms don't work. They respond to tamoxifen, aromatase inhibitors, but these people usually also get chemotherapy. Luminal A usually gets hormone therapy only, whereas luminal B's get a combination of hormone therapy and chemo. Their survival rate is worse. It's about 40% of 5 years, and like luminal A's, they tend to metastasize the bone. A typical path report for a luminal B would be a grade 2 or 3, irregular shape, indistinct margins, ER 30%, PR 10%, HER2 1+, or 2+, with non-amplified fish, and a Ki67 say of 25%. The HER2 molecular subtype is only about 10% to 15% of all cancers. These are thought to arise from the luminal epithelial cells of the duct. So they're DABs, and they may undergo epithelial or mesenchymal transformation. These are multifocal in the highest percentage of cases, 65%. And it's very common to see almost a few lesions with lots of small foci that involve two, sometimes even three quadrants. So the largest focus in my experience might be 12 millimeters, but you might have 10 or 15 foci. These have the highest percentage of positive lymph nodes, and I've noted, personally, I've not seen this literature, but when I see a lesion where the lymph nodes are so bulky that they have a larger volume than the breast lesion, I think HER2 first. Some of these may have a little bit of HER2, a VR and PR expression. They tend to have pretty high Ki67s. I find it 50 or 60 often. These have a higher P3 gene mutation rate, up to 75%. They more often have extensive DCIS, or what Laszlo calls DAB components, usually grade three, and they are often associated with fine linear branching calcifications. So you may see an evasive mass together with extensive calcified DAB components. These respond to the trastuzumide, pertuzumab, targeted antiher2 drugs that are antibodies or kinase inhibitors. They have a 31% survival of five years without treatment and a high recurrence rate, but they do respond very well to antiher2 drugs. Interestingly, they tend to have visceral myths and to metastasize towards brain. And brain is very important. A lot of these insurance companies are not approving PET CT scans that include the brain. It's a big mistake if there's a HER2 positive case, not to include the brain on your PET CTs, because about a third of these patients get brain myths. And the reason for that is that trastuzumide is a large antibody, too large to cross the blood vein barrier, so it will prevent disease below the blood vein barrier, but not above the blood vein barrier. So a typical report here would be grade two or three, irregular shape, indistinct margins, nuclear grade three with comedone necrosis on the DAB components. ER put negative PR negative HER2 three plus or two plus with an amplified fish and KI67 at 60%. Triple negatives is again 10 to 15%. They are thought to rise from the myoepithelial cells of the assymus, so they are AABs, but they are thought to rise from myoepithelial cells. They're usually unifocal. They're more common with BRCA1 mutations. They're more common in African-Americans. They have the highest P3 gene mutation of all lower lymph node metastasis rate. They tend to go hematogenously more early. They're the highest chance of presenting as an interval cancer between screening mammograms. They tend to have less associated DCIS or DAB components, but they have a higher occurrence rate, usually within one to four years. So if you can get to five years with a triple negative, you might be cured. They also tend to have metastasis to visceral liver lung brain. Typical report here again would be grade two or three, usually three. Irregular in shape or indistinct margins. ER negative, KR negative, HER2 negative, and KI67 might be 90. Now let's look at how sound transmission works on these molecular subtypes. As you would expect, we can distinguish triple negative from other types really well with a p-value of 10 to the minus 13. Distinguishance luminal A from other molecular subtypes, again, 10 to the minus 14. Not quite as good with HER2, but it's a significant p-value of less than 0.05. So if we actually look at the graphs of what percentage of cases have what kind of sound transmission, we can see that triple negative by far has the highest enhanced sound transmission rate, and it makes it 5.5 times more likely that it's triple negative than luminal A if there's enhanced sound transmission. On the other hand, if there's acoustic shadow, we can see that luminal A has the highest rate there, and if there's acoustic shadowing, it's 6.8 times more likely to be luminal A than triple negative. So again, if we look at these two lesions, the one on the left has enhanced sound transmission, the one on the right shadowing, it's taller than wide with a thick halo. The one on the right is going to be the luminal A, the one on the left is going to be the triple negative. Now, the sound transmission with molecular subtype and histologic grade, you know, it's all interrelated. So the, you know, molecular subtype and histologic grade are not independent of each other. So if we look at our 524 invasive breast cancers where we have all the receptor data, we can see that luminal A's are about half and half grade one and grade two. Luminal B's in the other hand have many fewer grade ones, and mostly grade twos and quite a few grade threes. Her twos have virtually no grade ones. They're all grade two or grade three, mostly grade three, and then triple negatives even more. But, you know, we can say that it's not surprising that grade and molecular subtype correlate similarly with enhanced through transmission because, again, we can see that sound transmission, enhanced sound transmission goes with grade three and with triple negative, but grade three also goes with triple negative. So you can see the interrelationship, and by the same token, if we look at shadowing, shadowing means it's more likely to be grade one or luminal A, but we can see why because most luminal A's are grade one. So we can see the interrelationship. You know, what I'm saying is the histologic grade and molecular subtype are not independent of each other because they're related. Now, if we look at the actual scores for molecular subtypes with a 90% confidence intervals with luminal A and the left luminal B and then her two and the triple negative, we can see that triple negative and her twos have much lower sound transmission scores than luminal A or luminal B, but we can see that there's even, there's not even any overlap between luminal A and luminal B. So we can't really tell her two from triple negative, but we can tell her two and triple negative from luminal A and from luminal B, and we can also tell luminal A from luminal B. So again, if we look at this sound transmission, molecular subtype and histologic grade, the lesion on the left is taller than wide with the thick halo and speculations and acoustic shadowing. It's going to be a luminal A grade one, and the one on the right is circumscribed or indistinct with enhanced through transmission, not much of a thick echo than a halo. We're going to think triple negative in grade three. Now, what about taller than wide? Well, when we talked about an atom, we talked about how the TDLUs are located in the periphery and that AABs arise from the TDLUs and that, you know, enter and posterior TDLUs tend to be oriented not parallel to the skin, whereas terminal TDLUs tend to be parallel, but the cancers that arise from the enter and posterior TDLUs will be taller than wide. So AABs, at least some percentage of AABs is going to be taller than wide. We talked about how DABs arise in the center of the mammary zone and the ducts, and the ducts are horizontally oriented. So the DABs are always going to be wider than tall. So what this means is that if we see taller than wide, it ought to be an AAB and probably a grade one AAB or a luminal A AAB. So again, we showed how a small AAB from an anterior TDLU looks like a tennis rack with head up to handle down lies in the superficial half of the breast. From posterior TDLU looks like an upside down tennis rack with the handle up in the head down in the deep half of the breast, but some terminal TDLUs are going to be wider than tall. They'll never be taller than wide. We can see that as a lesion gets larger, its chances of being taller than wide falls off because, you know, larger lesions represent either DABs that were never taller than wide or they represent combination AABs and the DABs where the DAB component is making the lesion wider than tall. So how does the data look on that? Well, we can see that grade threes do have a lower percentage of non-parallel orientation in grades one or grade two, but it was not statistically significant. The P value was .09. Triple negatives do have a lower percentage of non-parallel orientation of luminal A subtypes. And so that did have a significant P value of .002. One thing important here is that in the byrads, addition five, round is considered not parallel. And so many triple negatives are not parallel by virtue of being round. So that sort of the definition that you should define a round lesion is not parallel is the correct thing to do, but it makes this correlation weaker, I think. So if we look at a non-parallel orientation versus grade, we can see that indeed grade one, which goes with luminal As, is 1.7 times more likely than grade three. And if we look at the molecular subtypes, the one thing that sticks out is that the HER2 is only one-third as likely as luminal A to be taller than wide. And that makes sense because we know the HER2s often have a large DAB component or maybe entirely DAB, and they correlate with linear casting calcifications, which are in a horizontally oriented duct, so it does make sense. So here we're seeing a taller than wide lesion that should be an AAB. It looks like it's arising from an anterior TdLU. It has normal sound transmissions, probably going to be a grade two AAB, luminal A. And then the one on the right is round and has advanced through transverse mass triple negative. Here we have a taller than wide lesion, same lesion I showed you previously. Again, that's going to be a lumalase single TdLU cancer arising from an anterior TdLU. And here we have multifocal disease with a very wide distribution, much wider than tall. I mean, this is a very classical pattern for multifocal HER2. Now, what about the size of ducts? So in Byred's edition five, this would be considered part of associated findings. We noticed in the past that larger ducts tend to correlate with grade three. A DCIS components are in site two components, and smaller with grade two and intermediate size with grade, smaller with grade one and intermediate size with grade two. We know that the mechanisms of duct enlargement in DAB or DCIS are four components. You have tumor cells within duct lumen. You either have secretions or necrosis within the lumen. You can have variable amounts of perioductal lymphocytic reaction and perioductal desmophlegia. And we know that all four of those tend to be most common in grade three DABs, less common in grade two, and certainly grade ones may have only tumor cells with or without secretions, but generally known necrosis. So if we look at these two ducts side by side, on the left is a normal duct with some minimal duct hyperplasia. On the right is a DAB grade three, and what I want to focus on is the widest part of the duct just to the right edge of the histologic slide and the outlined rectangle shown by the arrow. And at the point where the arrow is, is the widest point of the duct. There we can see purple stuff in the lumen. That's the micropapillary tumor cells. The white stuff is the secretions that are secreted by the micropapillary DAB. The pink stuff is the perioductal desmophlegia on the walls of the ducts. And then the purple stuff outside of the, is the, is the perioductal lymphocytic reaction. But the reason the duct is so large there is that all four components are contributing. So we would think that when you get large ducts in the surrounding tissue, it's going to suggest more likelihood of all four components contributing and more likelihood that it's grade three and more likelihood that it's going to be her two. Now, how good are we as seeing the large ducts? Well, if it's grade three, we found that you can have four, six, eight, 10, even 12 times normal size ducts. So it's not difficult to detect and large ducts caused by grade three. Nuclear grade two, you may get two to four times normal size. So if it's two times normal size, it could be tough to pick up. But at four size, four, you can probably pick it up. So we're intermediate in our ability to pick up grade two. We're pretty good at grade three. Grade one frequently does not enlarge the duct enough to efface the surrounding loose stromal tissues within the surrounding the ducts. And so we often can't even detect grade one. So big fat juicy ducts like this on the left would tend to go with grade three. Intermediate size ducts, as we're showing in the middle, would tend to go with grade two. And small ducts that are hard to tell from curved speculations tend to go with grade one. Same with microlabulations. We know that microlabulations are often clusters of enlarged ducts. So when we have large microlabulations, up to say two, three millimeters, this tends to go with grade three. Very small ones tend to go with grade one and intermediate sized ones with grade two. So if you look at large ducts in the surrounding tissue and their correlation, you can see that there are multiple correlations. Now what's interesting is that HER2 should be the best correlation, but we didn't get statistical significance. I think there's an artifact of small numbers. Unfortunately, during the study that was funded by a company, the research coordinators at the sites did not realize that a 2 plus HER2 IHC required a fish. And so the study got closed before it was evident that there was missing data on the HER2s. So almost all the, I shouldn't say almost all, but a large percentage of the 2 plus IHC HER2s did not have the data for the fish entered. They got the fish. Somebody knows it at the site, but it never got put into the central database. So I really think the HER2 should be the best correlation, but it isn't because of small numbers. There were only 22 total HER2 cases, which was about one-fourth as many as triple negatives and less than a tenth of what we had for limelays or limelabies. Now, if we look at actual numbers, even though this wasn't statistically significant, if you had enlarged ducks in the surrounding tissues, it made it just about two times more likely that it was HER2 positive than HER2 negative. So it's kind of what we predict, just not significant p-values due to two small numbers. Now, if we look at grade again, we said that if we see enlarged ducks, it ought to be more likely grade three. You can see that that's true. If there are enlarged ducks, it makes it 2.4 times more likely to be grade three than grade one. Now, I also mentioned when we talked about molecular subtypes that the types that were often associated with extensive DCIS or DAB components are limelabie and HER2. You can see that indeed, if we have enlarged ducks, it makes it 2.9 times more likely to be HER2 than triple negative, and 2.3 times more likely than Illuminal A. So indeed, enlarged ducks tend to go with grade three, and they tend to go with Illuminal A over two. So on the left, we have some moderately enlarged ducks surrounding an invasive cancer that's multifocal. That turned out to be Illuminal B, with grade two DAB components. On the right was a mass with much larger ducks in the surrounding tissues. This turned out to be a HER2 grade three. Now, here's a case where we have lots of very prominent ducks in the surrounding tissues, lots of small lesions, and lots of calcations in the ducks. This, the calcations in the ducks, definitely tends to go more with HER2. And so this, in fact, was a grade three HER2 lesion. Now, what about the relative thickness of the echrogenic rim? I showed you some recent articles showing the thickness of the rim, and speculations tended to go with Illuminal A in grade one. So what evidence do we have to support this? Well, on the left, we have a small taller than wide lesion with a very thick echrogenic halo, and very long spicules. This is a 38 millimeter probe, and the spicules are essentially 3.5 centimeters wide. Even though the width of the lesion is about 4 millimeters. That's very typical of Illuminal A, or grade one, and or grade one. Now, the right is a roughly round lesion with enhanced through transmission, and it has an echrogenic rim, but it's relatively less thick, and the spicules present are less prominent. So if we look at the relative thickness of the echrogenic rim to the hypococentral nitus, you can see a big difference, and the same relative length of the spicules versus the width of the internal rim, much less with Illuminal A. I mean with triple negative, sorry. So here I'm just showing you ratios. So the white measurement would be the width of the hypococentral nitus. The pink would represent the width of the lesion, including its echrogenic rim, and if I were taking a ratio of pink divided by white, you can see that the pink divided by white is much bigger for the grade one Illuminal A on the left than it is for the grade three triple negative on the right. So when we plot this out, and we look at the relative thickness of the boundary zone to the hypococentral nitus, you can see that indeed it's greater for grade one than grade two than grade three. Unfortunately, there's a little bit of overlap. You can see that the 95th percentile, or the upper confidence interval for grade three just very slightly overlaps that for grade one, and greatly overlaps grade two. If we look at the echrogenic rim thickness versus molecular subtype, we can see that the confidence interval of triple negative overlaps Illuminal B on her two, but does not overlap Illuminal A. So we can tell Illuminal A from triple negative, but not necessarily Illuminal A from Illuminal B, Illuminal A from her two, or triple negative from her two, or Illuminal B. So we can tell the opposite ends of the spectrum from each other, but not the fine gradations in the middle. Now if we look at the length of the spicules versus the hypococentral nitus, again, we can see that grade three has significantly shorter spicules compared to the hypococentral nitus compared to grade one and grade two, and in fact, doesn't overlap the confidence interval for either grade one or grade two. So you can tell grade three from grade one and grade two, but you can't tell grade one from grade two. Now, if you look at molecular subtypes, again, triple negative can easily be distinguished from Illuminal A and Illuminal B. Illuminal A and Illuminal B can even be distinguished from each other, but you can't tell triple negative from her two, and you can't tell Illuminal B from her two. But what this data is showing is that indeed, the Illuminal A is most likely to be speculated and have a thick hypoallergenic halo that's very thick compared to the hypococentral nitus. The spicules are very long compared to the hypococentral nitus, whereas triple negative is very likely to have a thin or absent hypoallergenic rim with absent or short spicules in the surrounding tissue. So basically, we're showing that both boundary zone and peripheral zone features can distinguish Illuminal A from triple negative, not necessarily from Illuminal B or her two. Now, am I troubled that we do a better job of telling grade one from grade three than we do from grade one from grade two? Not at all. There are several studies now published where they've done genetic assays and showed that, and this is very interesting. I'm going to tell a little side story. A friend, a pathologist mind, told me that there is no such thing as grade two. There's just grade one and grade three, and he calls it grade two, and he can't decide whether it's grade one or grade three. So he meant that as a joke. He wasn't being serious, but in effect, there's some truth to that. So these are genetic assays, and what have been shown is that grade one and grade three are markedly different from each other genetically. But there are really two types of grade two. There's what they call a G21 and a G22, or G23, I'm sorry. So about two-thirds of grade twos genetically are similar to grade one, and about one-third of grade twos are genetically similar to grade three. Now, what that means is that whatever imaging biomarkers we see for grade two should be more similar to grade one than they are to grade three, because a larger percentage of grade twos are more genetically similar to grade one. They've shown the exact same thing with proteomics, and the exact same thing with non-coding RNAs, which are the key things that determine epigenetics. So genetically and epigenetically, grade twos are a mixture of lesions, about two-thirds of which are similar to grade one, and one-third of which are similar to grade three. So I think the data here show this. You can see that in terms of percentages and actual scores, grade one and grade two are more similar to each other, and then grade two is to grade three. Now, this is all very complicated. I've shown you innumerable features that correlate. So here I'm showing various grades versus each other, and again, we've used the same color coding. If I give it an aqua color, it's better than 0.05. P value less than 0.05. And if it's green, the P value is less than 0.001. So you can see there are lots of correlations. And then if we look at luminal A and luminal B and triple negative and heart two, so if we look at my leg, our subtypes again, lots of different correlations. And what this means is it's very complicated. How's our mind going to figure all this stuff out? Now if we look at a luminal A and what its most common feature score is for the various categories, and the five categories are shape, echo texture, sound, rend, mission, margin, boundary zone, and peripheral zone, what I've shown in yellow is different between the two. So really the only thing that's similar between triple negative and luminal A is the presence of an echogenic rim or echogenic halo. And yet it's not like it's 100% and 0%. And we've got 51% versus 31%. 43% versus 52, 65 versus 57. So these percentages are not staggering and be large. So how's our mind going to deal with this? Well, it's very difficult. Now if we look at less significant features, what I'm looking at is various shape scores, echo texture scores, so they're color coded by the five different categories. And if I made it bold in yellow, it means that there's a 3 to 1 difference in the percentage of cases that are triple negative versus luminal A. So if you look at oval shaped, triple negatives have 8.9% or oval shaped, where it's only half a percent of luminal A. So that's an 18 to 1 ratio. If you look at oval or round shape, it's 11.4 versus 2.1, a 5 to 1 ratio. But these are small percentages. It's not like 100% of triple negatives are that and 0% of little ways are that. So again, even though these ratios are greater than 3 to 1 differences, it involves only a subset of the cases. And so it's difficult for our mind to put all this together. So here's this taller than wide lesion with shadowing at thick halo and very long specules. This should be luminal A. And this is a oval shaped or circumscribed lesion with very thin equigenic rim compared to the central nitrous and very short spicules. This is triple negatives got enhanced through transmission. I mean, these are classics. So they jump out at you, but not every cancer is a classic. And a lot of the things in the middle, the herd twos and the little bees are tough to pick out. So we could use some help there. So we're going to talk about this in the next talk. What I'd like to do is get machine learning to help us. AI machine learning. So what we could do is put in many of the things we know correlate with prognostic features such as age, race, BR status, mammogram features, MR feature size, universal multifocalody, large lymph nodes, the ultrasound feature scores, and just have the machine figure out what the likelihood of each malignant subtype is. So we're going to talk more about that in the next talk. But in summary, what I've shown you is that we can and we probably should ask more of our salt cell. And if we want to really deliver personalized medicine, there's a tremendous amount of information on ultrasound that we're just throwing away and not using. There are numerous ultrasound features of breast masses that are significant prognostic biomarkers, but it's really difficult for our puny human brains to put them all together without the help of AI. So in the next talk, I'm going to show you how I would visualize this could work in the real world. Thank you.