The first talk is going to be on remote control, managing an enterprise cardiac MR practice. So over the next 15 minutes or so, I'm going to be speaking on some of the barriers that we face in expanding cardiac MR across an enterprise, which is becoming more and more relevant as our health systems are growing. And as we expand our services beyond the large medical centers. And then also show you some ways in which you might be able to improve your access to and standardize the performance of cardiac MR across a large enterprise. So as many of you are aware, cardiac MR is challenging. Part of the challenge in performing cardiac MR is due to patient related factors. The heart sits in a very oblique orientation within the chest. The anatomy can be variable from one patient to the next. And then there's a very wide variety of pathology that we're being asked to evaluate. And depending on the specific pathology that we're imaging, the sequences, the acquisitions that you're going to be using are going to vary. There are also scanning related factors that can make cardiac MR challenging. Specifically, we have many different sequences. They have different names on different vendors platforms. We also have to take into consideration how are we going to compensate breathing and cardiac motion artifacts. And then finally, there are differences from one scanner to the next in terms of what sequences may or may not be available clinically or commercially. And as our health practices grow and expand, we're often scanning patients with different vendors, scanners. And so we also have to become familiar with what is available on all these different platforms. So traditionally, cardiac MRI, because of its complexity and the many challenges that we face in performing it well, has been the domain of the large academic field or very large medical practices, sort of the White Tower model of healthcare in which the expertise in performing and interpreting the cardiac MRI studies were concentrated in a single center. Benefit of this approach is that you're frequently able to have access to the latest technology, which is particularly important in cardiac MRI, but it also presents a lot of barriers. And so ultimately, what we would like to do is make cardiac MRI available everywhere, not only in the large cities, metropolitan areas like Chicago, but even out into the smaller community practices wherever you may be. And so how are we gonna do that? And I think as you'll see over the course of the rest of these sessions, part of that I think is based in improving or optimizing our cardiac MRI protocols. I would encourage you to strongly consider moving from a comprehensive, let's do everything and everyone approach to a focused protocol. Another key component to expanding cardiac MRI everywhere is technologist training. This is probably the most important factor at this point, making sure that your technologists are trained in setting up the patients for cardiac MRI, being there hands-on at the beginning when everyone is learning how to do these types of studies. In the not too distant future, it's possible that a lot of this work will become automated. Some of this is possible already, but I think there's a lot of exciting developments taking place that will automate this even more and hopefully enable further expansion of cardiac MRI. So traditionally, cardiac MRI has been done by throwing every single sequence available in every patient, multiple orientations, no matter what the indication for the cardiac MRI. And this leads to very long scan times, at least an hour, if not an hour and a half, which a while ago was not an issue, but as MRI use becomes more and more prevalent from head to toe, it's becoming more and more difficult to find an hour available on the MRI scanners. And so this is really a big motivation for looking at what can we do to speed up our acquisition. And I came across this paper from 30 years ago that asked the question, when is one MRI sequence sufficient? Now they didn't include any cardiac MRI in this study. It was primarily neuro and MSK imaging, but they found that in the majority of cases, one sequence was sufficient to answer the clinical question. And there's really no reason why that isn't possible in cardiac MRI as well. Oftentimes the question is, did the patient have a myocardial infarction? Does the patient have myocarditis? And in many of the patients, if not most of the patients, you could answer that with one sequence. And so I would, again, we're gonna hear more about this in the next couple talks, but I am a big believer in using focused cardiac MRI protocols based on the recommendations of the SCMR that really relies on your Cine images for cardiac size and function and LGE. We pretty much do those two main sequences in everyone, and then everything else is supplementary or optional, depending on the indication. So advantages of a comprehensive approach to cardiac MRI is that it does lead to standardized imaging in all patients. It includes every sequence in all patients. So if there's something incidental, you probably have the data available to address anything that's not expected. However, scan times are very long, and most of the sequence do not add any value in that specific patient, or lead to very minimal additional value to the study. Focused protocols lead to shorter scan times. All of the sequences that you are acquiring contribute to your final impression to changes in patient care. Disadvantages is that there's variability from one patient to the next, depending on their indication. And then it does potentially require some additional work up front to figure out why that patient is coming to you for a cardiac MRI. And so there's a variety of focused protocols. I think each institution probably has a few different flavors of focused protocols, but in general, I would, if you haven't done this yet, or if you are considering implementing this approach, I would advise you, or strongly suggest, referencing the SCMR guidelines, because they have very good examples of these types of focused protocols. So we start with our routine SSFP sequences for size and function. We almost always acquire LGE imaging if we're giving IV contrast. And then everything else is kind of an optional sequence, depending on the specific indication. A couple years ago, when we were looking at how can we improve access to cardiac MRI, because we were at a point where there was increasing demand and patients were not able to get scheduled for a cardiac MR for at least a month, which really wasn't acceptable. So we looked at the causes for the delay in CMR access. And the problem was that cardiac MRI at the time had to be scheduled in a 90-minute slot, which was very difficult to find. We only used two MRI scanners in the main hospital center where the radiologists and cardiologists were located. So we then looked at how can we improve our cardiac MRI access. The first thing that came to mind was, well, let's get our scan times to under an hour, because 60-minute slots were much easier to find than 90-minute slots. And in order to do that, we had to start protocoling our studies before they got scheduled, so that the schedulers knew exactly how much time they would need on the scanner. Depending on the protocol, depending on the indication, you might be able to get that cardiac MRI done in 30 minutes. But there are some complex congenital heart disease cases that might take well over an hour, an hour and a half, especially if anesthesia is being used. And then, in order to expand our access beyond sort of the first shift availability, we had to train our technologists that are working on second shifts, and we had to start training our technologists that are working at the other sites that we had scanners. And so, when we did this, we found that almost 90% of our studies could be done in less than 90 minutes, most of them in 60 minutes, and about 30% in 45 minutes or less. This allowed us to decrease our scheduling time, increases our scanner availability for other studies, knees, brains, and then also decreases your cost to perform cardiac MRI. Now, one of my former fellows, who's now at the Mayo-Jacksonville, did something similar when she moved down there a little bit over a year ago. And similar situation, all of the studies scheduled for 90-minute slots. They acquired comprehensive studies in everybody. She implemented focused cardiac MRI protocols. She looked at the amount of scan time that was required compared to historical comprehensive control studies, and found that all of her studies were done less than 60 minutes. The average amount of time it took to perform the cardiac MRI was actually 40 minutes. So that's a significant improvement in efficiency, which is gonna improve your access and allow you to increase the amount of cardiac MRI that you're doing, expanding it out into the smaller community practices. Now, on a much larger scale, there's the rapid cardiac MRI group. This is a group based in multiple different countries around the world, where they have come up with 15-minute cardiac MRI protocols that are really focused to answer a specific clinical question. They have found that in 90% of their patients, they can answer the clinical question with this rapid cardiac MRI protocol, using this rapid 15-minute protocol, they can change patient management in almost 56% of patients. So talking a little bit about cardiac MRI training, as I mentioned, I think an essential component to expanding cardiac MRI access is to ensure that your technologists are trained. There's a very good link to educational programs, training programs on the SCMR website. The rapid CMR group also has a link to training programs in various places around the world. So that's another good place to go to find training programs. And then very briefly, sort of leading into the next couple talks, I think we've come a long way in the last 20 years in terms of enabling automatic cardiac MRI scanning through a variety of approaches. Over the last few years, this has been focusing primarily on using deep learning. There are now a couple vendors that have automatic scan planning available commercially. And then there's other vendors that are working on additional approaches to enabling AI approaches or deep learning approaches to automatically scanning. So in summary, a comprehensive, one-size-fits-all cardiac MRI approach in every study takes too long, reduces access to MRI scanners, and does not always translate into added value. Focused cardiac MRI protocols that answer specific questions will reduce the scan time, improve access to the MRI scanners, and optimizes the value of advanced imaging technologies. Technologist training is essential to successfully expanding cardiac MRI across your enterprise. And then automatic scanning using AI has the potential to further increase the ease and consistency of cardiac MRI. With that, I'd like to thank you for your attention. The title of my talk is Superspeed Towards Comprehensive Cardiac MR in 20 Minutes. And I'd like to thank the organizers for the opportunity to talk a little bit about this idea. And I think we're not quite there yet, but with what was mentioned in the previous talks, I think we have a good chance of getting there. So I will expand a little bit on this. In this talk, I will take you through choosing your protocols. It was already alluded to, but I think if you take one lesson away from this session, it is develop these focused protocols and use them. You cannot do everything in 20 minutes. I'll talk about some techniques to speed up the exam, and we'll come to some conclusions here. So with regard to these protocols, I think it's important to consider two position papers by the SCMR, one of which was already discussed. So this is the indications paper that recently came out. So is this a good patient to send to MR in the first place? That is what this paper looks at. And it's a comprehensive list of, let's say, indications, including the classes, the strength of the evidence for choosing them and sending patients to MR. So that's where it starts, I think. Then we have the standardized protocol document that Chris already mentioned. If you photograph the QR code, it'll take you to the free download. This is an open source published paper, and this is a fantastic paper. It has taken a lot of work to get this together from a lot of experts, but if you take a look in this paper, you can see it gives you very detailed outlines of the different sequences you can run, but also it gives you very good hints, a consensus-based hints on what you should include for each disease indication. And yeah, so it discusses both the technical aspects as well as the disease-specific protocols. And these are evidence-based, again, was reviewed by a big group of experts who reviewed the literature for evidence of efficacy, and yeah, broad experience in MR in these teams. Also representation from different parts of the world, as Chris also mentioned in his talk, as some in other parts of the world, they may have different needs or less time per patient, so we also needed to take that into account. And also importantly, for different vendors. So there was a rigorous review process, so this was, yeah, I think an excellent resource which is carefully vetted, and it's popular. It's been downloaded more than 50,000 times, so I hope we can count you among the people that start using this document. All right, so how do we go towards faster CMR? Well, first of all, we choose a tailored protocol. I advise you to design these protocols together with your referring clinician so they also know what they're getting when they order an exam for a certain clinical question. And then what you should do, I think, is examine the entire imaging chain for possible gains in efficiency, and a dedicated scanner is part of that, highly trained technologists. And what we're seeing now is a slow introduction of these, let's say, AI-guided acquisitions with automatic scan planning, things like that. But there's also a whole family of new sequences on the horizon that can be used to speed up examinations, and also some AI tools, for instance, that allow you to get better image quality so you can scan faster, the images are noisier, but you can then denoise them using AI techniques. So this is also a way to obtain a shorter exam. And finally, I'll show you some examples of high-element coil arrays that are in development. I won't have time to talk about post-processing, so we won't do that. Over the, let's say, past half-century, MR has come to an existence, but also a tremendous amount of innovations have taken place in the field, and this is a great paper to list these innovations quite nicely, so you can see what they can do. I mean, I'm showing a picture of a brain here, but yeah, this is a great paper if you wanna learn a little bit more about the technical foundations of why MR is now suddenly so fast and it can do so much. And all of these changes have basically resulted in changes, real efficiency gains in cardiac MR protocols over time. So in 2010, we had long breath holds, free breathing acquisitions of four to five minutes, lots of 2D, some 3D, single contrast, lots of repeats when things went wrong. And I think now we're in the era that these things are better, so we're coming down to shorter breath holds. Most of the free breathing sequences are now short, a couple minutes. We have transitioned some things to 3D acquisitions, and with better training and better knowing what we're doing, we're able to reduce the number of repeats significantly. So where's this going? I think in 2030, what we're looking at is completely free breathing, ECG-less acquisitions, volumetric as much as possible. Almost everything will transition from 2D to 3D, I think. And yeah, what you will also probably see is significantly less use of contrast agents because there are also now sequences that can depict some of the things that we're interested in without injection of contrast agent, which will also speed up the protocols again because we don't need to wait for that delayed enhancement, if that works, and several other technical advances. So let's take a look at some of these things. So Chris already showed automated scan planning. So this is a paper that recently came out, actually one month ago, that also discusses this. This is an AI-based description and planning of the entire cardiac MR examination, including AI-based shimming fully automatically. So they use, obviously, neural networks to get this done. There's a scout volume, and from this, automatically, all the different orientations that we're used to are prescribed without any human input. And this seems to work quite well in this study. What happens then is the AI network, or another AI network, basically is used to optimize the shimming, to optimize image quality, which can sometimes be a challenge and lead to repeated acquisitions. So this seems to be an interesting way to move forward here, and this will save time, ultimately. So I already spoke about AI and machine learning for denoising images once you have obtained them, and these are some slides that were given to me by my colleague, Jeremy Collins, who's looking into this at Mayo. And the idea here is that you reduce the size of your cave space, but if you do this, you get artifacts. So you need to filter that away and things like that. So if you reduce your cave space, you can run into artifacts. Obviously, we don't want that. So this is a traditional way of doing this, but if you use machine learning, you can actually probably do a better job at scanning faster but still getting good images. So here's the concept. You take the original cave space, and you don't just chop off a part of it on the outside, but you use a neural network to denoise that cave space, and then you reconstruct your image. So you scan the smaller cave space because you're scanning faster, which leads to more noise. You denoise this, and this is what you then get. This is from one vendor. So this is a standard image on the left, the accelerated image in the middle here, and then accelerated and AI denoised, which looks significantly better in just under two minutes. So that's quite nice. Here's an example from another vendor. So this is a parallel imaging approach on the left. So if you accelerate too fast, you get these noisy images. You can denoise them a little bit with compressed sensing, but the AI denoising works significantly better, as you can see here, and these are all similar acquisition times of seven seconds per slice with the same spatial resolution. This is not only suitable for delayed enhancement imaging, or T2, as you saw in the previous slide, but also for bright blood sequences, and there can also be significant time gains there, if you, as you can see. So here's another example for Cine imaging. So this is a highly accelerated sequence. You can see the noise breakthrough. This would probably be good enough for drawing circles and getting an EF, but the image on the right obviously looks much better, and perhaps we can even further accelerate and accept some noise and go to a faster acquisition again. So these are some of the things that are coming and are being developed by vendors and are becoming available, I would say, in the next year or so. Another concept is free-running acquisitions. So that means no ECG gating, no respiratory triggering, no breath-hold commands. One of the ways to do this is by optically monitoring physiological signals, so there's one vendor that has a camera built into the scanner that allows you to read the skin tone, which fluctuates ever so slightly by every heartbeat. So this signal can be read and put into the MR, let's say, reconstruction process, and you can obtain Cine images without cardiac triggering that way. So that's a really neat invention, I think. Here's some other examples using this approach, comparing VCG to contactless pulse and breathing measurements so the images look quite similar. Another way is to derive the physiological signals from the case-based data, and that's what's done in this approach here. So what you do is you put the patient in the scanner, you say, start at the diaphragm and at the thoracic aperture. That's the only thing you need to do, and then you press the start button. And this approach was developed by many people, but one of the groups who has done a lot of work in this is the group of Mathias Stuber in Lausanne. And this sort of came out of a frustration. You can go to Amazon, order anything with one click, but why can't we do a cardiac MR with one click? So that led to this approach. And yeah, what you do is you basically acquire a case base, which is a big data set of about 12 gigabytes, and then from there you derive the physiological signals and use that information to correct for respiratory motion as well as cardiac motion. And then you can create anything you like. You can create a 3D volume of data, as you can see on the left. You can create an image that looks at cardiac motion, but you can also create images that depict the respiratory motion. So this is a very interesting and powerful approach. This is obviously for balanced SSFP bright blood imaging. Here's another example, but you can also apply this approach to different contrasts such as flow imaging. If you encode the flow information as well, you can use this approach also to start looking at flow. And this is a collaboration between Mathias' group and the Northwestern group here in Chicago. So fantastic work. And the final approach I want to highlight is this one. This takes it one step further. So what you do here is you not only, this not only a free running approach for let's say one different parameter, but for all the different information you want to acquire. So also T1 and T2 maps can be acquired from these data. And this is called multitasking. So basically what you do is you encode the information in an image tensor that you then decompose for the different question or contrast or measurement that you want to do. And yeah, these approaches are pretty interesting. And I do know that some vendors are interested in this. So we might well see this coming into clinical practice in the coming years. One of the problems with these techniques is that these are huge data sets with lots of images, takes a long time to reconstruct. But Mattias Stuber told me that he's working together with a post-processing company, and they have managed to reduce the reconstruction time to less than 10 minutes now. So this seems to be clinically feasible then. Final thing I want to show you before I finish my talk is next generation coil technology. So we've been focusing on the, let's say, pulse sequence side of things, but also there are new developments in coil technology. This is a project that I'm involved in. Every company is working on this, but we got a grant from the Dutch Technology Foundation to try to build a next generation of coils. And this, what you see right there on the right side here is a prototype 196 element cardiac coil. And if you have so many coils, you can highly accelerate. So this is what we're investigating right now. Yeah, you can go up to compressed sensing factors of nine, which is significantly more, about three times as fast as we have today. So if you combine this with all of the previous innovations, I think we're in a pretty good position to reach that 20 minutes in the future. So with that, I'd like to come to my conclusions. And the most important thing that was also highlighted in a previous lecture, you cannot do everything in a 20-minute protocol. So take these guidelines and design highly tailored, focused protocols. Examine the entire imaging chain for possible gains in efficiency. That's something you can do today. And in the near future, you will be able to exploit new sequences, reconstruction techniques, and dedicated hardware to speed up cardiac MR and make it much more like we do with CT today, right? You just say start here and there, press the button, and it's done. This is where we're going with cardiac MR. Thank you for your attention. Okay, I'm gonna talk to you about flow imaging. And I'm gonna start by just doing a quick review of some of the technical aspects of flow. I'll then talk about some of our current clinical indications, how we use 40-flow MRI in the current clinical environment, and then finish up by talking a little bit about some of these acceleration strategies, how they can be incorporated into this rapid CMR protocol. So let's just start with some of the technical aspects of flow imaging. So flow imaging is based on the technique phase contrast MR. And phase contrast MR is essentially a subtraction technique where you acquire one spoiled green deco image with the polarity of the gradients in one direction. Then you take another image with the polarity reversed. You subtract one from the other. The signal that you see is gonna be the signal that results from any flow in the image, which is typically blood. We normally encode the flow or velocity in one direction in our routine clinical practice, but you can encode the velocity as shown here in three directions. So this is three-directional flow encoding where you get that magnitude image here, which looks like an anatomic image, and then these three different directions of flow in all the three axes. We don't really look at this routinely because we can't visualize the three axes, but we can post-process those images. But this is essentially what 4D flow is based on. So if you add in spatial encoding in the 3D direction, you get 4D flow MRI. So it's tri-directional velocity encoding plus the three-dimensional spatial encoding. And this is what it looked like here. You've got a 3D data set of magnitude or anatomic images, very similar to what you might see with cine-MR, perhaps, or MR angiography. And then you got the three-directional velocity encoding. This dramatically increases the number of images, and that's one of the challenges with 4D flow MRI. As seen here, you get anywhere from 2,000 up to 10,000 images. So this results in long acquisition times of five to 15 minutes in its original version. So that really throws the idea of the rapid CMR protocol out the window if you're looking at this unaccelerated approach. The other aspect of 4D flow MRI is that then you have to visualize this image data set. And this really involves very complex post-processing tools, which used to be done offline and took typically up to 30 minutes for an experienced student. Now it can be done in line in a smaller period of time. But when we do this post-processing, we can really visualize the flow, in this case, through the heart, looking at the inflow, the outflow, the flow in the ventricles. You can color-code it, and you can, of course, measure velocities and flow through the valve, because you also have the area of the anatomy. So with a standard 4D flow MRI, you get a three-dimensional view of the vastiture, so 3D MRI or MRA. You can also look at functional imaging. Now, in this case, we've added in a standard Cine Imaging, and that can be tagged into the protocol. And then you get your flow acquisition as well to measure velocity and flow. And now we really need to, as I mentioned, be able to post-process this data from the original 3D velocity and 3D anatomic acquisition. And there are now many post-processing tools to allow you to do that quickly. This is one of the vendors currently available, which will, in the space of a few minutes, give you an anatomic visualization. So this is a volume-rendered view of the anatomy. This is a patient with a residual co-art. And then you can go and post-process this further and look, in this case, at the 3D streamlines, which allow you to measure the velocity from any point, in this case, in the thoracic aorta by placing various sample planes throughout the aorta. And of course, you can color-code it to visually look at the areas of high velocity. So what are the current clinical applications that we use 4D flow MRI for? And I've listed them all here. I'm not gonna go through them other than to show you some examples. But of the main application areas, we are routinely now using 4D flow MRI in all of our routine protocols. You can also apply this in other areas. Valvular heart disease, I will show to you that this is beginning to enter into the main application area, but it's still somewhat emerging because it's complex. You can use 4D flow outside the heart, in the liver, for example, for portal venous flow, either pre- and post-tips, for example, or pre- and post-liver transplant, or cerebral blood flow in the setting of evaluating aneurysms or vascular malformations. So here's an example of a patient with aortic stenosis and a thoracic aortic aneurysm. And our typical protocol would include MR angiography, 2D single-direction phase encoding, and also 4D flow MRI. And you can see from here, there is an aneurysm in the ascending aorta. When you look at the 2D flow, there is some aliasing in the outflow tract. You can measure the velocity through plane, which would be fairly typical of a routine protocol. But then you can also visualize both the antegrade velocity and the retrograde velocity, in this case, using 4D flow MRI. And this is illustrated here. Here's my mouse here, but you can look at the antegrade velocity to measure the peak velocity from here. And you can really see the regurgitant jet very elegantly in this setting. In another example, then, where we can also use 4D flow in the intracardiac environment is looking at the mitral valve. And here's an example from Dr. Zhao in UCSD, where he used a 4D flow acquisition to look at a severe mitral regurgitant jet in greater detail. Now, I'll illustrate here that you're also getting the anatomy in great detail, which shows you the functional information of how the left ventricle is contracting. You can reconstruct the anatomic information to also show you your standard SIN-A orientations, for example. And then you can also look at the mitral regurgitant jet and measure the regurgitant fraction in greater detail. You can use, here's another example from a clinical case at our institution, where we looked at a pulmonary valve stenosis, for example, on the 2D view, we measured a peak velocity of 3.5, and on the 4D flow MRI view, you got a peak velocity of 3.45. So it also matches pretty well in clinical practice. More routinely, 4D flow MRI is used in the setting of congenital heart disease. In an example here from UCLA, this is from Paul Finn, they use what's called this music technique, which is essentially a 3D SIN-A acquisition of the anatomy, which allows you to visualize the entire anatomy in a functional SIN-A view. And you can see here the contraction of the left ventricle and the right ventricle. You can reconstruct the data in any orientation similar to your 2D SIN-A. And then you can encode the velocity information. As you can see here on the anatomic SIN-A view, there is a defect in the interatrial septum, and then you can very nicely see the corresponding flow jet from left to right across that AST. And of course, you can go back and retrospectively then analyze any part of that, measure the area of the defect, or measure the velocity of flow across the shunt. In another example of congenital heart disease from our own institution, here's a case of an incidentally-detected patent ductus arteriosus, seen very nicely on the high-res MRA view. You can also see it on the dynamic view. You don't get as good of a visualization of the flow across the shunt, but then on the 40-flow MRI, again, beautifully show the shunt. You can see the to and fro flow across the PDA, and then you can go in and measure the QPS by placing these planes in the aorta and in the pulmonary circulation to allow for a very comprehensive evaluation of the hemodynamics across that defect. You can do this, as I mentioned, with a cohort. Here's another case where you can visualize the flow in the cohort, but then using some of the standard post-processing tools that are now available, you can calculate the pressure gradient across the cohort. You can do this in de novo cohorts. You can also do this in stented cohorts, which is more frequently what we're asked to evaluate with cardiovascular MRI. And then another interesting application, this is an example from Brad Allen in our group at Northwestern. This is the so-called aneurysmal dissection. These are type B or sometimes type A dissections with residual type Bs where they're corrected surgically, but then they're followed over time, and frequently these aneurysms dilate. So the false lumen will dilate significantly. You end up with a large aneurysm which ultimately needs to be surgically prepared, but it's not clear what causes the aneurysm or which ones need to go to surgery. And even when you look at this MRA, it's really hard to get a good idea of the full extent of the aneurysm. This is the entire false lumen here, and this is a through lumen. But when you look at the 40-flow MR acquisition, you can see the reason for the ballooning of the false lumen. There's a fenestrum right here with rapid flow through the defect. And you can even take that a step further and measure the flow and the velocity across that jet by placing a plane right over the jet. And you can see here in the output that the flow through the fenestration here is approximately 20 to 25% of the entire stroke volume, so likely the cause of the expanding false lumen and aneurysm. So again, an application that could be used in practice by tagging it onto some of your existing protocols. So where are some of the future directions for 40-flow MRI? And I'm borrowing another one of Dr. Markle's slides here where he nicely summarized some of the current challenges. So long scan times, we talked about that. Well, there's a lot of focus on advanced acceleration techniques, which I'll show you a little bit about. And non-standardized analysis, so we need to optimize the workflow. And then are there some hemodynamic markers which may point to prognosis and outcomes? And talk a little bit about that in the last few minutes. So 40-flow MRI can be accelerated similar to some of the other techniques that you have heard about. And it's really based around compressed sensing. And this is a paper fairly recently published out of our group where they accelerated the acquisition up to the point where you could reduce the scan time down to just over two minutes. And here's the typical scan time of 10 minutes just with parallel imaging down to two minutes with compressed sensing without any significant reduction in image quality or without any significant inaccuracy in some of the quantitative information. You can apply compressed sensing to your entire coronal 3D acquisition. This is just an example of that. This is being acquired post-pheromoxidal, which is a blood pool contrast agent, but can actually be acquired after any contrast agent. The reason you might want to do that is to increase the signal in your anatomic images. Here's the anatomic cine image, again, with the flow data. And then when you reconstruct the entire data set, you really get, again, a really nice view of the entire hemodynamic pattern as well as the volumetric and functional pattern. But this is now done in a total scan time of five minutes. And remember, from this, you can reconstruct retrospectively the scan planes if you so wish. So this is the total acquisition in just a single acquisition, or total output in a single acquisition with essentially one click. Here's another example, again. This is from Lillian Ma. Again, using this acceleration strategy, now focused on 40-flow MRI of the thoracic aorta. This is reducing the scan time down to two minutes, but showing that there's almost a direct match between the quantitative velocity and flow that you get between the accelerated 40-flow MRI and the standard 40-flow MRI. We use this routinely in our clinical practice now, this accelerated acquisition. Two minutes added on to an MRA and some limited cine imaging for evaluating bicostal aortic valves and associated aneurysms. You saw a little bit about this from Tim Leiner. This is the work that arises out of the group from Lausanne, Shu and Lausanne from Matthew Stuber's group, and some targeted work by Lillian and Michael Markle to really accelerate the flow acquisition. So this is the single-click, single-push-button acquisition, gives you all the cine information, gives you the flow information. You can reconstruct this in any orientation you want. This is self-gated, self-respiratory-gated, self-ECG-gated, and then you can, from that, reconstruct the flow information that allows you to go back and retrospectively, sorry, that's not playing, but allows you to retrospectively visualize and also quantify all the flow and velocity information. Another interesting innovation that has also just been recently published is using 40-flow within the heart itself to look at the valves. And the challenge in the heart is, first of all, it's a long acquisition and we're beginning to address that now with compressed sensing, but also in the heart, the valves move and they typically move by a great degree as shown here. So your single acquisition plane, if you're placing it over the valve in diastole, as we would routinely do, that'll be in a different position to systole. So using this valve tracking analysis plane, you get this continuous acquisition or measurement plane, which is centered over the annulus. And this moves with the 40-flow quantification in the post-processing environment. And this group and others have been able to show that this is easy, it's quick, it gives very accurate and reproducible information about flow within the intracardiac valve, such as the mitral valve and the tricuspid valve. And in this other paper, it showed that the analysis times were cut in half. It improved the consistency of flow analysis with excellent intra-observer agreement and showed quite a significant number of regurgitant severity calculation discordances, which is not unexpected, given that the correlation would have been with a 2D phase contrast. AI and deep learning is not only integrated into the visualization and post-processing side, as well as the acquisition side, it's also across the entire value chain here. And again, this is work out of Michael Markle's group where they've introduced AI at various steps in the quantification process, which results in a reduction of the post-processing to the order of seconds or minutes compared to as long as I had mentioned, 20 to 30 minutes in a prior version. And this is gonna rapidly improve and significantly enhance our efficiency for 40-flow MRI. Finally, I just thought I'd show this slide from Dr. Elbaz, who uses 40-flow MRI to generate what he calls a virtual catheter. So this is really reconstructing the 40-flow visualization along the centerpiece or the central part of the vessel, which gives you the equivalent of almost a catheter, in this case, being passed across and withdrawn across a co-arc. And it allows you to almost in real time measure the peak velocity at different parts of the vasculature, much the same way as you might do this in an interventional setting where you pass a catheter and then you withdraw it across the stenosis. So just to finish up in conclusion, I will say that 2D phase contrast is still fairly routine for velocity mapping within the heart. 40-flow, however, is very powerful and has increasingly been integrated into most multivascular territories. There are currently numerous clinical indications, including congenital heart disease and pulmonary hypertension, which I didn't talk much about, but these are more advanced indications for 40-flow MRI. There are many promising new biomarkers, such as kinetic energy and wall shear stress, which are becoming available, which may provide information about prognostic information. And I think this newly accelerated 40-flow MRI will allow us to do a comprehensive exam with function, morphology, and flow in a single acquisition. I'd like to acknowledge all of these collaborators and investigators for input into this talk and for their slides, and thank you very much for your attention.

cardiac MRI

health enterprises

patient-related factors

technical issues

focused protocols

deep learning

4D-flow MRI

technology integration