Okay, so it's my honor to be here and talk to you about LIHRADS Future Direction. So LIHRADS is a dynamic system, and it will be expanded and refined as knowledge accrues in response to user feedback with updates every three to five years, next update expected in a couple of years. And it's an active dynamic system because it's led by a steering committee led by Professor Victoria Cherniak, who'll be hearing from in a minute, with multiple different working groups looking at different aspects of chronic liver disease and LIHRADS, and just a shout out to the members of the LIHRADS steering committee. So with that as an introduction, let's talk about LIHRADS Future Directions. This is sort of an entire vision of the future. I will not be talking about everything today in the interest of time. So let's start with something simple. One of the things we wanna do is just clarify some things. We wanna clarify three things. We wanna clarify what we call LIHRADS-M under the table. We wanna clarify that ancillary features do not apply to LIHRADS-M, and we wanna clarify that aggregate reporting is referred when there are innumerable observations on patients on systemic therapy. But I won't get into detail in all of these things. I'm just gonna talk briefly about LIHRADS-M under the table. So this is something that actually exists in LIHRADS right now, but the way we created it, we realize now in retrospect wasn't as clear as it should have been. So what is this LIHRADS-M under the table concept? Here is the diagnostic table, and what we would like to clarify visually is pointing out that if something makes it to the table but it's not a LIHRADS-5, and it has infiltrative appearance necrosis or marked diffusion restriction, it should be called a LIHRADS-M. Or perhaps just visualize something differently, if it ends up in the table as any one of these cells, and it has one of those features, infiltration, necrosis, or diffusion, marked diffusion restriction, call it LIHRADS-M. Okay, let's now talk about some simplifications. There's three different simplifications that we're interested in doing, reducing ancillary features, reducing the cells in the table, as well as simplifying the so-called dreaded diagonal cell. So let's go through these things. So first of all, we want to reduce the number of ancillary features. Shown here are the total number of ancillary features, and some of these features end up being either infrequent or being non-discriminatory, meaning that they are found pretty consistently, they are collinear with other features, and therefore don't have incremental added value. And some of the ones that have no added value, we may potentially eliminate from the algorithm to make it simpler. In addition, we want to reduce the number of cells in the table. So this is, again, the current table. And I think that the evidence may allow us to simplify it in the following way, by basically having a single column for lesions without AFI, but that will depend on actually having evidence. I also would like to get rid of this dreaded diagonal cell, which causes a lot of confusion, and which I personally have an emotional dislike of. So I feel emotional about this diagonal cell, and I'd like to get rid of it. And I think the evidence will show that things that are currently in the diagonal cell could be called LIRADS5 and still maintain a positive predictive value of over 90% for HCC. So after all those simplifications, the table might look something like this. Again, that's a prediction that will have to depend on actual evidence. We also want to improve the performance of LIRADS. And there's many things that we want to improve. In the interest of time, I won't talk about all of them today. I'll just talk about one of them, that we want to improve the sensitivity of LIRADS5 for HCC without sacrificing sensitivity. And basically, this is what we're trying to do. So we have an ROC curve. LIRADS needs to maintain 95% specificity for HCC. And so what we really want to do is without sacrificing specificity, we simply want to increase the size of that ROC curve. Now, it's not easy to improve the size of an ROC curve, so how can we do it? Well, there's many ways that we could potentially do it, and I'm only gonna focus on one, which is we want to optimize this diagnostic table. Earlier, I was talking about simplifications. Now, I want to talk about optimization. So first of all, when you look at this diagnostic table, it might look complicated, but it's actually a simplified visual representation of what it actually represents. So this is the actual diagnostic table if you spell it out in all of its detail. And this allows us an opportunity for optimization. So first of all, we'd like to identify those LIRADS4s that actually should be LIRADS5s because they're very likely to be HCC, and I think the evidence will allow us to say that these cells should be LIRADS5. Similarly, I think the size thresholds were initially arbitrary, and I think we could optimize them. In particular, I think we could lower some of the size thresholds, and that would also boost the sensitivity. Finally, we'd like to convert some LIRADS3s that are likely to be benign into LIRADS2s, and I think those lesions or observations I'm showing here could become LIRADS2s. We want to convert some malignant LIRADS3s to LIRADS4s, and what we're left with then is a diagnostic table that looks something like this. So I envision that the future table may look more like this than what it looks like now, possibly also with the simplifications I mentioned earlier. We want to expand the content. So there's many ways in which we want to expand the content. I'll only talk about a few of them right now. So in particular, we want to create new algorithms, new manuals, and we want to integrate quantitative imaging. So right now we have four algorithms, ultrasound contrast, enhanced ultrasound, CT, MRI for diagnosis and staging, and CT and MRI for local regional therapy. We would like to expand these algorithms to include, in addition to the ones we currently have, the following. So first of all, we'd like to develop an algorithm not only for ultrasound surveillance, but for abbreviated MRI surveillance. We would like to add an algorithm for contrast-enhanced ultrasound treatment response, and we'd like to expand the CT, MRI treatment response into radiation therapy, which you'll be hearing about later from Dr. Mandirata Lalla, as well as an expansion, eventually, of the CT, MRI treatment response algorithm to include systemic therapy. And then, I'm sure Dr. Lalla's working hard right now at developing, I'm sorry, so right now we have a manual for diagnosis, staging, and treatment response, and we would like to expand those manuals to include ultrasound abbreviated MRI and contrast-enhanced ultrasound. And one thing that I think is gonna be really exciting in the future is the following. Currently, when we use Lyrads, we require that patients either have chronic hepatitis B or that they have cirrhosis, and that puts us, A, in a conundrum, because we often, as radiologists, don't know who has cirrhosis and who doesn't. But the other problem is that that may not be the best way of doing it, because not all patients with cirrhosis actually have equal risk of developing HCC. So what I'm showing you here is a completely hypothetical model, but in the hypothetical model, I think that it will be depending on patients' underlying liver stiffness. So for example, I'm making this up now, but patients who have a stiffness, say, of greater than five or greater than four kilopascals on MR allostrography or equivalent stiffness levels on ultrasound allostrography would be suitable for the application of Lyrads, whether they have cirrhosis or not. Finally, we would like to expand our scope. So I'm not gonna talk about everything. I'm just gonna talk about a couple things. We'd like to expand beyond cirrhosis, beyond ordinal categories, and beyond diagnosis. So what do I mean by that? Well, right now, Lyrads, as I mentioned, applies to patients with cirrhosis, but not only that. Notice that it doesn't apply to all patients with cirrhosis. It applies to certain causes of cirrhosis only. And in terms of patients who don't have cirrhosis, it only applies to adults with chronic hepatitis B. Notice it doesn't apply to children at all. So we would like to expand beyond this, and I'll explain in a moment how I think that can happen. In addition, Lyrads currently has ordinal categories, whether we're talking about ultrasound surveillance, CEUS, or CT MRI diagnosis, or whether we're talking about treatment response. Everything is ordinal, and we would like to move beyond ordinal, and I'll explain in a moment how we might do that by focusing in on the diagnostic algorithms. And so this is sort of a vision for the future. So instead of having ordinal categories, what we envision will happen is that there will be an integration of imaging features of the observation itself, an integration of features in the background liver, potentially extrahepatic features, very importantly, integration of non-imaging clinical data, treatment history, and circulating biomarkers. Now, I'm not necessarily saying that we will have an algorithm that literally puts every single one of these things into the model, because I don't know which of these things will end up being truly predictive and which won't, but I do imagine a future in which key discriminatory features from this list over here will get factored in with imaging to create a very specific numerical probability of HCC and malignancy. In this particular case, this lesion over here, 92% probability of HCC with some error around it, 95% probability. And then this will allow clinicians to optimize their treatment in a truly precision individualized manner. Now, notice that in the future vision, LIRAD's imaging features are likely to be an important component, but not the entire component of the system. But even diagnosis is not enough. So again, what I was showing you in the last slide is that we think in the future we'll be able to provide continuous probabilities of HCC and malignancy, but we would like to move beyond diagnosis and in addition to predicting diagnosis, to predict prognosis. Not all HCCs are created equal, not all cholangiocarcinomas are created equal. It's one thing to make the diagnosis, but can we actually prognosticate and can we actually predict which lesions will respond to various treatments? This is, by the way, a 10 to 20 year horizon, this continuous probability. This is not something we're expecting to see in the next year or so. Now, how will we get there? I don't have time to talk about it. It's gonna require innovation, it will require unification, it will require collaboration and research both within the Americas as well as outside the Americas. And I think the development of registries is gonna be very important towards this development to allow evidence-based refinement. So in summary, in the last few minutes, I have provided an overview of where I think LI-RADS is going, and now the speakers to follow will tell you what LI-RADS actually is and its current present situation. Thank you so much for your attention. My name is Mustafa Bashir, I'm from Duke University. So I'm gonna start simply describing what this concept of pathomolecular features and prognosis is all about. If you're, I'm a country radiologist from North Carolina, when the words get really big, I can get a little bit confused, I think about imaging for making diagnoses and the prognostication thing gets a little bit strange. But I think it's a very important topic, something we need to be aware of and a place that we're going to be contributing a lot to the care of these patients in the future. It's something we're already doing without necessarily thinking about it that much. Then I'll talk about some of the pathologic subtypes of HCC that we can potentially predict as well as how to actually use imaging features based on what we already know for prognostication. So imaging has a huge role in HCC as you all know. We do surveillance, we perform diagnosis, we guide treatments and we do treatment response assessments but many of the things that we do are actually static assessments. We're using the imaging to look at features of a tumor and predict things about its histology, whether it's aggressive or not, whether there's invasion, things like that. So we're not really looking at things happening in the future, we're talking about what is happening today and we're trying to predict what's happening on a smaller scale than what we can see. And this contributes to the overall treatment of the patient in the following way. So we have patient level features, the patient's functional status, whether they have any impairments and then we use imaging to predict what's happening on a microscopic level and then we put that information together to derive a treatment. And right now there are particular things that we're trying to predict, how invasive is the tumor, how large it is, that sort of thing. But we consider an HCC to kind of generally be an HCC. We're taking a nodule and deciding it's an HCC or not, maybe it's a cholangiocarcinoma or not, that kind of thing and then the assessment kind of stops there. And it turns out that there's a lot more information in the imaging that might allow us to not just say, hey, this is an HCC but this is a really bad HCC or this is a really good HCC. And this should be a familiar concept. If you read a lot of liver imaging, you know that some of the tumors that you see are scary tumors. These are the ones that are gonna come to multidisciplinary conference, they're gonna get treated and you're gonna see that patient back in six months because they've recurred. And then there are some that are not a big deal. But we know this conceptually and we need to firm it up so that we can make actual predictions in much more specific ways than our gut feelings about, hey, this is a good tumor, this is a bad tumor. Now there are different ways to think about the concept of good tumor or bad tumor. The WHO has a subtyping system where there are eight or nine different subtypes of HCC, including conventional kind of standard HCC. And interestingly, most of the HCCs that we see are the conventional type. So we're trying to predict a relatively small minority that are these unusual subtypes. And these different subtypes are associated with better or worse prognosis. So both for counseling the patient about what's likely to happen with their tumor as well as treatment selection, it might be useful to know if a tumor is likely to have a worse prognosis where it needs more aggressive treatment or a better prognosis where it might need less aggressive treatment. Another way to think about subtyping tumors is based on immunologic stains and tumors can be divided into proliferative and non-proliferative subtypes. Proliferative means it's gonna grow and that's bad. And non-proliferative subtypes are the more quiescent ones. And these subtypes are, in fact, associated with different prognoses. If you look, for example, at macrotrabecular massive HCCs, which is one of the proliferative subtypes, and you look at survival curves and recurrence curves, the macrotrabecular massive HCCs, these shown by the red lines, have a distinctly worse prognosis, distinctly earlier and higher frequency of recurrence compared to non-macrotrabecular massive HCCs. So this is potentially useful. And we can identify these, not with a high level of accuracy but with a moderate level of accuracy based on the imaging. This is a typical example of a macrotrabecular massive HCC which is characterized by large size at the time of presentation and large amounts of necrosis. Reportedly, a proportion, a volumetric proportion of 60% necrosis or more is pretty strongly associated with this subtype. So if we see a tumor like this, we go, okay, this looks very different from the typical HCC that I see. There's a lot of necrosis here. We could predict that this may be a macrotrabecular massive type HCC. This is more likely to be an aggressive HCC than the standard one that we see. And on the less aggressive side, the steatohepatic HCC is an example. This is different from simply steatotic HCC that has fat in it. These, if you biopsy them, they actually look like steatohepatitis. There's not only fat deposition in the liver but there's inflammation and fibrosis in the tumor that looks like a NASH liver but it actually came from the tumor. And these are reported to have a better prognosis compared to conventional HCCs. The main imaging feature is the presence of steatosis. Now, that's all well and good. So we can try to predict histology from imaging. We're kind of doing the same job that we're already doing though, right? We're doing a static prediction. We're saying, okay, based on what I see on the imaging, what do I think is happening on a microscale? And then indirectly, we're trying to influence morbidity and mortality or predict morbidity and mortality based on what we think the histology is. But we have an opportunity to do even better than this and maybe try to use the tumor imaging features to more directly predict the patient's outcome without going through that waypoint of, I think this is the histology and therefore. Could we say, okay, this feature is present and therefore, and go straight to prognosis on that basis. And there's emerging literature that you can actually do this. If we think about the proliferative or non-proliferative paradigm for the subtypes of HCC, there are also imaging features that tend to be associated with proliferative or non-proliferative subtypes of HCC. And I've adapted this from a really great review article that Katie Fowler published in Radiographics last year. I really recommend that you read this article as a starting point on this topic. Now many of the features that you'll see here are fairly familiar. We've known for a long time that small is good and large is bad. We know that tumor in vein is bad. We don't need the histology. We don't need some intermediary endpoint for that. Tumor in vein is a bad thing. It's a poor prognostic finding. Infiltrative lesions are bad. But there are other imaging findings that are starting to emerge that are maybe a little bit less intuitive or less well-validated up to this point as predictors of good prognosis or bad prognosis. For example, if an HCC presents with Lyrad's M appearance or with rim arterial phase hyper-enhancement, this is a bad thing. This is a piece of work that Siobhan Kieran's published this year in JMRI looking at Lyrad's M lesions and showing that the hepatocellular carcinomas and the intrahepatic cholangiocarcinomas have very similar survival and recurrence outcomes to one another, even though they are distinctly different pathologically. One is an HCC, one is a cholangiocarcinoma. But at imaging, they look the same and they behave the same. So this is very interesting. These Lyrad's M HCCs don't behave the same as conventional HCCs. In fact, they behave more like the cholangiocarcinomas. And histology would not suggest this. Imaging would suggest this. Other features that are coming out as potentially predictive of worse outcomes include rim arterial phase hyper-enhancement, which is probably what drives the Lyrad's M categorization of these HCCs, as well as markedly restricted diffusion, another interesting feature that predicts worse prognosis. So to kind of summarize this just general thought process, we have even more of a role for imaging now than what we used to. We can potentially predict these tumor subtypes, which might be useful in prognosticating the patient's outcome. And we might potentially be able to do more than what we are already to predict outcome just simply based on the features. Now, you might be asking yourself, so what, who cares? I've talked a lot about predicting outcome, but those are the outcomes that the patient has. Can we do something to change those outcomes? And the answer is maybe. So if we think about how we treat HCC, this is a very simplistic way of thinking about it. But generally speaking, for early tumors, we might start with local regional therapy. We wanna get the patient to some sort of resection if we can, and that's their best chance at a cure. And then if they recur, we go into more local regional therapy or systemic therapy as a sort of salvage where we may not cure them after a recurrence, but we're trying to give them as much time as we can and fight the tumor as long as we can. And up till this point, the paradigm of multimodal treatment up front has not really existed outside of multimodal local regional therapy. But in almost all other tumors that we deal with outside of the liver, there is this idea of combining systemic therapy and local therapy, and that generally improves outcomes for a lot of different types of tumors. In HCC, this has been difficult to do before because up till a few years ago, the systemic treatments were very morbid and not very effective. And this has now changed. There are treatments that on their own for advanced tumors offer survival beyond two years, which has not been the case up till now for HCC. And so the thinking shifts to for these early stage tumors, which have a recurrence rate of something like 70% at five years after resection, can we improve that by applying systemic therapy early on? And in order to do that, we're going to have to carefully select which tumors we wanna do this for. Now, if trials show that everybody benefits from neoadjuvant systemic therapy, then we should do neoadjuvant systemic therapy for everybody. But there's a good chance that that won't be the case because there are indolent HCCs out there where you resect them and you're done for the patient's life. And so it's gonna be on us up front to try to figure out which of these HCCs can be treated simply with resection and which of these HCCs are likely to recur and need more therapy up front to increase their time to recurrence, possibly offer a cure through a multimodal treatment instead of just resecting them and then waiting for the recurrence to come back. So we can go even further beyond just prognosticating what's going to happen with these tumors and possibly, possibly determine what the optimal treatment is for these patients. And there are more options in terms of those treatments than ever before. So we need to play a very active role in trying to determine which patients are gonna benefit most from which types of therapy. So with that, I wanna thank you all for your attention. Okay, so we've heard a lot of interesting things so far. And now, for the next 12 minutes or so, I'll talk about the current gaps in knowledge we have in Lyraids version 2018. So as you know, Lyraids have been around since 2011 and I think we have done a tremendous job putting together this amazing system that works. But even though it works and it works great and we have a lot of scientific evidence to support that it works great, there are still some gaps that remain to be addressed. And I'm gonna focus on four things in the next couple of minutes of these gaps. The first thing is complexity. I know Dr. Serlin kind of alluded to it, but this is our algorithm right now. We have four steps and users who are just new to the system often get intimidated because we have eight categories. We have five major features. We have five LRM features. We have 21 ancillary features. We have rules. We have subtleties. We have a manual. We have a core. There's a lot. So it may be a little complex. But imaging of cirrhosis is complex if you think about it, right? We want to detect an HTC in a cirrhotic liver. But not only do we want to detect it, we have to distinguish that HTC from all other potentially premalignant, nonmalignant, benign treated lesions that occur in cirrhotic liver. And we want to do it with a nearly perfect specificity. An HTC as a tumor is complex, as we just heard from Dr. Bashir. All these pathomolecular differences exist. And as a result, the tumors, the HTC will fall into different categories. We really can't collapse them all into one. And as Dr. Stirling mentioned, we want to simplify the algorithm, but chances are we won't be able to simplify it too much because as Einstein said, everything should be made as simple as possible, but not simpler. And we do need a lot of evidence to support what modifications and simplifications we can do without loss of that performance that we have achieved today. Another gap is heterogeneity of Lyraids three and four observations. I know we've all seen this graph. This is a graph showing the percentages of HTC and overall malignancy per each of the diagnostic categories. It's based on a nice meta-analysis, a large number of patients. And what I want to highlight is that almost all observations that fall into LR5, TIV, or M category are going to be malignant. And nearly all observations, LR1 and LR2 are going to be benign. And I want to highlight that for LR3 and LR4 observations, in LR3 observations, about a third are going to be HTC. And in LR4 observations, about two thirds are going to be HTC. Why does it matter? Well, as we know, the categories drive management. They determine how the patients are treated. So for patients who have LR5, M, or TIV, or nearly 100% probability of malignancy, that's fairly straightforward. Patients will be sent for treatment options. And patients who have either no observations or almost certainly benign observations, that's also straightforward. They can go back to regular surveillance. But what do we do with these LR3 and LR4s, which fall into this indeterminate probability, not high enough to treat, not low enough to ignore? So the data we have for LR3 observations in terms of how they behave in the future is variable. And a large proportion of LR3 observations will continue an LR3 category over time. Also, a fairly large and wide range of LR3 observations can actually decrease in categories. And a minority, but important minority, will progress to either LR4, 5, or M categories. So some examples, here's an LR3, hyperbola phase hypointence nodule, unchanged for three years, remains LR3. Here's an example of a large T1, I mean, hyperbola hyperintence nodule, which is stable for five years and we can down-categorize it to LR2. And this is, again, what we want to determine, this small LR3 observation within two years will progress to LR5. For LR4 observations within six to 12 months of initial diagnosis, the data says a good proportion, about 44%, will remain in LR4 category. A chunk of them will progress to LR5 or M. And non-negligible minority can actually decrease in category. Just an example, again, most commonly, about a third to almost 40% of LR4 observations over time progress to LR5. If we look at cumulative incidence of HCC in LR3 and LR4 categories, within four years, just over 60% of LR3 observations will progress to frank HCC, and LR4 curve is kind of more linear, and more than 80, close to 90% of LR4s, by time, four years, will progress to HCC. Why is it a problem? Seems like we're doing great, but the problem is for LR3, right? We're recommending repeat imaging every three to six months, right? And so if we go back to this curve, this is all cumulative number of patients who will receive imaging every three to six months for four years, and will have no cancer, right? So it's a huge taxing on healthcare system, on patients, it's scary, it's difficult, anxiety is real, so this is an important thing. With LR4, it's a little less problematic, again, the probability of HCC is higher, but again, for some patients, you fall into, go with a very aggressive treatment, there's still probability, it's not HCC, so it still presents some treatment conundrums. We do have some data about which of these LR3 and LR4s are at higher risk for progression, they're listed here, but again, no study has ever shown a nice model that says, you know, with this combination of features, this lesion is more likely to progress versus if you don't have this, this is okay to not aggressively follow up for a few years. So clearly, we're missing a target with the LR3 and LR4s. So much so that NIH called, had a proposal call, which ended about a month ago, for research proposals, which would focus on these indeterminate lesions, and NIH called this a major unmet clinical need, ability to predict which of these indeterminate or LR3, LR4 lesions are going to be a problem. All right, our next puzzle piece that's missing is imperfect intermodality agreement. Currently, our diagnostic algorithm is equally applicable to CT, MR with extracellular conscious agents, and hepatobiliary conscious agents, why is it a problem? Here's an example, 23 millimeter observation on the outside study, has no arterial phase hyperenhancement, does have washout, so if we apply our diagnostic table folds into LR4 category, no problem. Patient came to us, and about 10 days later, we got an MR, and now, the same observation presents with, clearly we see non-rheum aphy, now we see nice little capsule, so now, if we apply our table, the same exact observation 10 days later, we call it LR5 category, what's happening? Well, it turns out that disagreement between CT and MR is not rare, and it can happen in up to 70% of cases for lesions imaged with both modalities, and it's not surprising, right, because we know that CT and MR have different sensitivities, some features are only accessible on MR and not CT, so that drives the differences. If we exclude LR1 category, MR tends to give a higher category than CT, but not always. And the ultimate solution, as Dr. Serlin alluded to, would look something like this, where we would incorporate, we would adjust for which modality was used in imaging the patient, put together all the imaging features of the lesion, background liver, incorporate patient factors, and come up with a very individualized diagnostic and prognostic probability. But as Dr. Serlin said, this is years and years, if not decades away, so what do we do now? Again, we have one study says LR4, one LR5, what do we do? Do we stay more conservative and stick with LR4, or do we become more aggressive and stick with LR5? Well, we have a data now based on a large meta-analysis, which is in press, and if you look at the graph, again, the same thing, percentages of HTC versus malignancy per each category, that looks very similar to the graph I showed you before. But if we separate these averages based on modalities, the graph looks something like this. It is a little busy, but let me highlight a few things. First of all, regardless of modality, percentage of HTC malignancy is nearly 100% identical between the modalities. For LR-TIV, almost all of them remain malignant, regardless of modality, and the same thing is true for LRM. So at least for our aggressive malignant categories, all modalities perform in an equal manner, which means that once we diagnose this as LR5, we can stick to LR5. This has 95% probability of being HTC, so we can proceed with treatment as planned. Our final gap in knowledge is imperfect inter-reader agreement, and the data I'm gonna show you is based on large meta-analyses. And actually, inter-reader agreement for major features is substantial for both CTNMR for all the features that we have. Similarly, for Lyraids category, inter-reader agreement is substantial, again, for both modalities. So substantial is great, but obviously not perfect. Why so? And one of the reasons is that we assess our features in a qualitative manner, and we compare to background. So for example, background is very heterogeneous. We can choose this part of liver to compare, in which case we would say there is no arterial face hyperenhancement, but if we choose this part, we would say there is. Similarly, depending on which part of the lesion we're gonna use to compare this or that one, you know, depending on what we choose, we may end up saying washout or no washout. So it's not perfect, but it may be as good as it can get. And if we look at the performance of Lyraids, which is, again, substantial, it really outperforms other RADs based on the MAR system. So substantial may be as well as we can do for variability for qualitative features. The ultimate solution may end up being AI-driven quantitative assessment. Okay, so in summary, we do have some gaps. We have algorithm complexity. We have heterogeneity of behavior of LR3 and LR4 observations. We have imperfect intermodality and inter-reader agreements. And as Dr. Sterling said, we really need a lot of data to fill these gaps and really move forward. With that, thank you very much for your attention.

LI-RADS

liver imaging

diagnostic performance

intermodality agreement

hepatocellular carcinoma

quantitative imaging

treatment strategies