Hello, everybody. My name is Omer Berenfeld. I'm from the University of Michigan, and I will co-chair this session together with Anand Ganesan from Flinders University. We want to welcome everybody to this session on mapping of complex arrhythmias, auxiliary approaches to improve outcomes. And we'll immediately start with the first speaker, Team Michael Dickfield from University of Maryland. Good morning, and thank you so much for staying a day longer. I know half of our friends have left already last night, so appreciate it. I was tasked to talk about imaging with molecular challenges to enhance voltage mapping and ablation. This is my disclosure. Now, to start out, I have to say that there is not that much going on. There should be much more, and maybe today could be one of those impulses for the field to move a little bit forward. As we all know, we're using mostly structural imaging to image anatomy, to look for the anatomic basis of changes caused by a disease. And we're using MRI, CT, sometimes ultrasound, looking for fibrosis, fat, wall thinning, or hypertrophy. But there is this whole other part of functional imaging that potentially could be used, and that would rather look at disease-related changes in cellular, molecular, or physiological properties. And some of those imaging modalities would be PET and SPECT, looking, for example, at metabolism, perfusion, hibernation, or innervation. Now, why even worry about this? We have learned over the last few decades that there is more to arrhythmias and to the anatomic basis to that than just pure anatomically fixed borders. Eliud Antler showed many years ago in an ischemic pig model that reentrance circuits are largely determined by functional barriers with anisotropic conduction. And Katja Seppenfeld showed in 15 VT patients with prior MRI that areas that were completely transmural showed, in terms of voltage, a significant variability of only one-third of them really showing a low voltage of less than 0.5, and many of them were in the border zone, or even some of them were normal. And when looking at some of prior studies, there are certainly enough physiological plausible explanations why functional imaging provides some proarrhythmic guidance. In chronic porcine infarct models, chronic hibernation resulted in regional denervation and nerve sprouting, be it proarrhythmogenic. And in the large PARAPET study on more than 200 ischemic patients, denervation predicted sudden cardiac death better than ejection fraction alone. And so there is this tantalizing concept that maybe putting anatomic and functional information together would provide some synergy and make our operations better. We started this many years ago looking at FTG-PET. What FTG-PET does, it looks at the glucose transporter in the transmembrane and transport system and the cytoplasmic hexokinase-mediated phosphorylation, and this frequently gets combined with information about the blood flow with rubidium PET and then results in those different combinations of either totally normal or scar where the FTG and the blood flow is matched, or when we have a mismatch pattern indicating some diseased myocardium with hibernation or stannic. Our lab and other labs have looked at the correlation of the voltage maps to those metabolic defects, and they correlated well in really all of those four main studies that were kind of at the foundation of that field, correlating mostly with PET uptake of 40 to 50% when correlating it to the voltage map. We tried to gain additional information from that. We were looking at 56 patients with ischemic cardiomyopathy and correlating about 23,000 points, and the obvious part is that a decrease in FTG-PET correlated to a decrease in voltage. We wanted to concentrate on three specific patterns that we found interesting. One of them was mismatch of the metabolism with the voltage, so an area of low PET activity with preserved voltage, and metabolic channel, so this is a little bit more akin to what we're using in our CT reconstructions with somewhat preserved wall thickness within wall thinning, and then also an area of a rapid transition zone where we have a rapid drop off of metabolic activity that looks in an integrated map similar to what our ILAM map would look like. When we looked at that, we saw that the metabolism of voltage mismatch areas occurred in about one-fifth of the patients and harbored the VT channels in about 40% of patients, and maybe more interesting were the areas of rapid transition zone that occurred in about 60% of patients, and we found the ablation sites in that area or in the proximity in most of the patients. When we look at those metabolic channels that we found in about 45% of patients and in terms of orientation length, pretty similar to the actual electrically mapped channels, we also found a good correlation with the ablation sites, about a third of those were in the channels and two-thirds in close proximity. An additional area of our interest was if this could help with prediction of survival, and so we looked at 35 patients with scar-related ventricular tachycardia, and we found that when we looked in a PET-defined transition area, so not the totally normal area that had an uptake of 70% or the scar area under 50, but in this transition zone of metabolically somewhat alive tissue that conferred an additional 31% mortality over an 18-month follow-up, and so patients with a larger metabolic transition zone had a higher mortality. What about hibernation? Because animal models have shown that hibernation would provide a pro-arthmogenic milieu. We know that sympathetic activation precedes sudden cardiac death in hibernating swine myocardium in multiple studies, and that hibernation results in partial sympathetic denervation and increased nerve sprouting. When we looked in a 61-patient cohort, we found that 13 of those did exhibit some hibernation, and we found that the VT channels co-localized in 43%. Interesting enough for us, there was some significant voltage heterogeneity, suggesting some anisotropy in that area. Now, moving from our PET imaging to innervation imaging, we've performed this mostly with PET imaging spec'd, and the 1,23-midiodobenzylguanidine is a norepinephrine analogue and has been correlated in chronic heart failure with mortality and ventricular arrhythmias. There is some pathophysiological plausibility to that. Indeed, nerves under fibers are more sensitive to ischema than myocardial tissue, so one would get a larger nerve defect with ischema than in terms of just myocardial necrosis. In those areas, we have remodeling that is a functional remodeling with neurotransmitter accumulation in the synapses. We have structural remodeling with nerve sprouting and denervation hypersensitivity, and electrical remodeling with prolonged affective refractive period, all of those are prorhythmogenic. When we tried this out in a patient population of 15 ischemic VT patients, we saw that, as we expected, our innovation defect was significantly larger, about more than twice as large as the area of low voltage, and over a third of the successful ablation sites were in this area of denervation, but in areas of preserved voltage. Only a few other centers have started to look at that. I want to show you some of the PISA data. They looked at an area that has preserved perfusion and decreased innervation. They labeled this the perfusion innervation mismatch. They assessed this in 16 patients undergoing ablation, and they found a decreased voltage in those PIM, in those mismatched areas. They found an increased prevalence of lava, and in patients that were successfully ablated, they found a reduction of PIM at follow-up scans. Our lab was interested in the innervation pattern and potential dynamic changes, and we looked at 13 patients with ischemic and non-ischemic cardiomyopathy, looking at an area that is totally denervated, which is in red, an area that is in the transition zone, which is the yellow bar, and then in green, an area with normal denervation. We saw that in ischemic cardiomyopathy, there was a larger area of abnormal innervation compared to the non-ischemic patients, and when we re-scanned those patients six months later, we saw that those categories had partially shifted, as well in the ischemic, as well in the non-ischemic group, suggesting that ongoing neurological remodeling or innervation remodeling does take place in those patients. That leads us to the question of, one minute, that's good, that leads us on this kind of my main final slide, when we combine our different imaging modalities. This is a work from one of our post-docs, looking at a voltage map MRI and MIVG, and when we've seen our Venn-Dahm diagrams, we can find areas that are only abnormally one category, two categories, or triply abnormal. We found our successful ablation sites all co-localized to the area that was abnormally innervated, so it was an area of MRI scar, but it was only in the area that was also abnormally innervated. And all those data, of course, could be much better looked at. We are all sure with AI, and so our hope is with Dr. Trajanova that we can look at this at some point in digital twin modality, and we're all curious to see what else we can find out, and make sure that they don't run over time. This is my summary slide, just saying that functional imaging is possible, but definitely there are too few people working on that, and it's a rich area, and just a thank you to all the collaborators that made this possible. Thank you. We have time for only one brief question from the audience. Can you say something about extrapolation of those technologies to the atria, because everything you showed is eventual? Yeah, so it's tough, because the spatial resolution is even worse. So we looked with some other centers as well at innovation, and that is investigative at best, but the reproducibility is just hard. For the PET, we have a spatial resolution of about five, six millimeters, and the atrial walls are too thin, and then we look at SPECT, we're really looking at one centimeter. It's just not there. You know what there is always work and technical advances, but it's not tomorrow. It's going to be in five, ten years. Thank you very much. Thank you. Let's move on to the second speaker, who's my co-chair, Anand Ganasen, will talk about topological approaches to the mapping of atypical atrial flutter. Thank you, Dr. Barenfeld. Now while the slides load up, I'll just add that I'm really stepping in on behalf of Nellie Van Der Sickle, who was unable to attend the meeting, but I've thought about these approaches and I've used these approaches clinically, and I think they're very powerful and significant for our field. Thank you. So these are my disclosures. So I'll start with a clinical case, actually. This is a patient of mine who presented in flutter at a rate of 155 beats a minute, and I didn't think the F waves looked normal. And I planned for 3D mapping. And he was a guy who had presented with spontaneous left atrial flutter to clinic. He had had no prior surgery with a sustained tachycardia at 155 beats a minute. I mapped it in the laboratory. And the question becomes, should we entrain, try and perturb the tachycardia to identify the optimal site of ablation? And what sites should be targeted? And I think this is a really or sometimes quite challenging and very difficult question. And those of you who attended the live cases would have seen how difficult this can be. So what we're talking about here today is atypical atrial flutter. And that is defined as a macrorentin atypical tachycardia that's independent of the cavotrichosporidismus. It often involves complex circuits around anatomical scars, such as the valves or veins or scars within the atrium. And they're rising in incidence because of atrial fibrillation ablation, particularly linear lesions, and cardiac surgery. And it poses a significant diagnostic and therapeutic challenge. And unfortunately, our long-term outcomes are not as good as what they should be. Now the conventional approaches that we use to mapping atypical flutter are really complex and developed over a series of last two decades. But in general, most people use a combination of activation. We're trying to really identify the sequence or entrainment mapping, where we pace or perturb within the circuit to kind of identify the critical areas. But the problems in practice are that there can be often extensive scars or low-voltage areas which can obscure the signals or make it difficult to annotate the timing. There can be ambiguity in distinguishing critical isthmuses from passive bystander tissue. And there can be difficulty in interpreting the entrainment responses. So putting all of this together, there is definitely a need for a more robust framework that's mechanistically plausible to understand and treat atypical atrial flutter. Now a new concept in this field is the application of topology to the atria, which is relatively new and really only emerged in the last two or three years. So what is topology? Well, it's a branch of mathematics studying the properties of systems that are preserved under deformation, so stretching or bending but not allowed to tear. So the classic example of topological equivalence is the coffee cup with a torus. And so how does that apply in the context of circuits and reentry? Well, what we're looking at really is that when we look at the reentry, the key thing that I think we need to take into account is that the surface of the atrium is actually an orientable surface. We can distinguish clockwise from anticlockwise. So in topological terms, what that means is if we have a circuit that's proceeding in an anticlockwise direction, we could call that topologically quantized as a plus one. And a circuit in the negative direction, in the clockwise direction, it's a minus one. And these circuits must be around non-conductive regions. So they can be anatomical, the valve annuli, veins, or the coronary sinus ostium, or acquired such as surgical scars or in the context of ablation. So one of the key principles that has to apply because mathematically is that the topological charge within the whole chamber has to sum to zero. And that is quite significant because a single isolated rotational activities are actually topologically impossible. Right? So what that means in practice is that there must be two wave fronts rotating in clockwise and counterclockwise directions around two separate but critical boundaries. So where does this come from? So how can this be? It's a bit strange to suddenly start talking about theorems, but it's actually very old, right? So the principle of topological charge conservation goes back to the time of Winfrey, who identified the linkage of circuits due to the ciphered surfaces in 1987, and was extended by Davidson and Caprao to boundaries in 2004. I even made an attempt to prove this theorem myself here, which I'm not going to go through today. But the proof idea is basically that based on the triangulation of integrals of the surface phase integral on the surface. So how does it apply in the context of these circuits? Well, for index theorem in this case, and the index theorem is that it's around because this is a special kind of thing. It's not a vector field, but it's a phase field that we see in the heart. So the sum of all the indices must be zero. So to give an example, this is here from Nelly's paper, which was published in CircuitP last year. You can see that in this simulation, you have a counterclockwise circuit here around one boundary, and it balances with a clockwise circuit here. And then, so the topological charge sums to zero. And it still applies in the case where the circuits are asymmetrical, which can be quite important, both clinically and mechanistically. So the interesting thing is that two separate papers that were published last year identified this principle separately, but they didn't realize that they had done it. So this paper was published clinically last year by the group of Santucci. And they actually demonstrated in this paper that in 128 cases, they could identify essentially what they were identifying is a clockwise and a counterclockwise circuit, and that you had to have separate but equal conduction boundaries, and you need to burn between those to terminate the flutter. And so what they showed was that in 128 cases, in 121 patients, that basically you could identify these two conductive boundaries in all of the cases and terminate the tachycardia in 96% of cases. And in the cases you couldn't terminate, it was basically because you couldn't block the isthmus. Now, I think Matthias Deutscheber, working with Nelly, have done the same thing. And they looked at it and made a classification of 131 cases and showed that there's a variety of pairing structures involving scars and valves and vein pairs. So I've done this now several times myself in the last two years or three years, because we've started noticing this pattern, right? So in this case in particular, what you can see, if you follow it through, is that there's a counterclockwise circuit that goes up and around here on the posterior wall and down here, and then the second one that comes around the left veins in the back. And so the spot to burn is right there, and this patient terminated with a single burn in this spot, right? Similarly, this is another case, and this is an example of a case where the difficulty of interpreting the maps can sometimes be there, but having an underlying construct can be helpful, right? So this case illustrates a scar that's on the anterior surface of the atrium. And what you can see, you could think it's perimitral, right, by this, but the activation here into the right pulmonary veins is somewhat confusing. But recognizing that then there must be a paired circuit, I think I realized that what I needed to do was to burn here on the anterior wall of the atrium, and we terminated at the meeting point of where we postulated there would be a second. So even though we didn't have a complete visualization of the second counterclockwise boundary, we were able to terminate the tachycardia in the correct spot. So the application of topological principles is the very exciting frontier that has implications for all arrhythmias, and in particular, I think it has implications for atrial fibrillation, because I think the principle of topological charge preservation occurs by the principle of continuity. And we've known the principles of topological preservation for many, many years. This is old data, but we know the rules of topological, and that phase lines can't intersect, phase singularities are joined by isophase lines, and phase singularities form and terminate as oppositely rotating pairs. So these rules are well known and have been known for many years. And those rules account for the statistical stability of phase singularity formation and destruction and their Poisson distribution of the phase singularities, principles that are remarkably seen in nearly all published data, and the same statistical signatures of topological defects apply through multiple systems throughout nature. So what does this all suggest? It all suggests to me that perhaps the atrial flat flutter, atrial fibrillation relationship is a topological phase transition. Perhaps it links what we see in the heart to most heaps in other natural systems. So what are the conclusions and take-home messages? Well, atypical atrial flutter is a challenge. It's complicated to ablate, it takes a long time and can be difficult, but topology offers a powerful theoretical framework based on boundaries and paired rotation. And topology-guided ablation shows a high acute success and provides a more fundamental approach to a mapping complex re-entry, because it takes into account the orientability of the surfaces that we're dealing with. Topological approaches like this are very exciting because perhaps we may be able to develop a more comprehensive framework encompassing atrial fibrillation too. Thank you. This presentation is open for discussion. Thanks, Anand. It's such a nice presentation and I'm happy to see that I spent a lot of time doing my PhD studies, trying to relate nonlinear dynamics to experimental studies. And what I observed in animal studies is that topological charges can teleport from one location to another, and sometimes they are not stable, like sometimes they disappear after one rotation. And when they form, they tend to be kind of correlated based on the spatial distance. So when you look for them in the atrial activity, do you always detect them in pairs? To call them face singularity, how many rotations do you have to observe to confirm they are stable? Or what do you think? Are they anchored to any anatomical structure, or do you think that's simple like dynamical formation? That's an excellent question. So to answer your question, my personal view is that these entities are on a continuum, right? So there's no specific significance of a single rotation or multiple rotations, that you have partial rotations, which are more frequent, and you have longer rotations, which are in the same spot, and they're biologically on a continuum. To answer your... I think you had a second component to your question. Well, I think you're right about your observation about the pairing. I think that makes sense. It seems logical by the principle of continuity that you will see pairing in your experimental observations. It seems logical. And I think in general, in natural systems, a topological charge is paired. So I think it's a general property of other systems in nature. In theory they're paired, but experimentally it's hard to see that. Sometimes they are not. Yeah, yeah, that's right. I think, well, the thing about it is, I think an important intuition, I think, is that even if the singularities themselves are not visibly paired, the overall topological charge in the system has to still be zero, right? I think that's a very important concept to keep in mind, right? I think it took me some time to get my head around this, but once you have that, it gives you a construct to be able to approach clinical cases. Thanks. I would like to have a comment on the question. I mean, the person who asked, please identify yourself when you ask questions. The constraints about the topological charge equals zero is pretty broad, and actually there is no constraint on the stability and the frequency of each of the pairs that form the topological charge, the total zero. So therefore, one may be stable, the other may be brief, and on top of that complication, phase singularity points can travel faster than the velocity of light because they are basically phase. So that makes it very complicated to track. There is one condition that also we have to keep in mind is the fact that an intramural re-entry that forms a filament in the form of a ring also has a topological charge of zero, and it's plausible. What do you say about that? Yeah. So I think that's an interesting question. So I think Winfrey already dealt with this, right? So I think that's an interesting point. I think you raise an important point. I think if we go back to the old papers of Winfrey turbulence, I think this issue of a sideways filament is a very interesting and thoughtful issue, but essentially what we have is a soup of turbulence in three-dimensional space, and so that soup of turbulence is still topologically. I think by the principles of continuity, when we take into account the boundaries and also the three-dimensional spaces, some of the topological charge is still zero, and the reason for that is that fundamentally the cells are communicating with each other. So except for areas of singularities, the manifold is differentiable even if it's differentiable in three-dimensional space. So I think that's where that comes from. I think we have other ways of cross-validating that. I think the statistical properties of the defects that we can see on the surface are also reflective of a three-dimensional formation and destruction process. But it's an interesting debate to have, and I think potentially reflective of our clinical experience. One more brief question, please. Brief one. Thank you. I've got lots of new phrases I can go home and impress my colleagues with now, but was this essentially what we just say is dual loop re-entry, first of all? And then secondly, the plus one minus one cancelling out, can you apply that to both atria, or is it just one atrium, one chamber? When you draw your space, how do you start distinguishing between the different anatomical chambers? Yeah, that's a very good question. So it's over the circuit. So the thing is, in general terms, when you're applying this principle in clinical cases, which is probably a worthwhile thing if you want to talk about how I apply this principle during clinical cases, there are some tricky situations to deal with in practice. So one is the incomplete loop phenomenon. You'd have to take this into consideration. But the thing is, there can be partial penetration by one loop of another loop, which you can identify with the mapper. I've seen this, and it takes intuition to see this. The other situation that is a complex situation is a biatrial flutter, and dealing with biatrial flutter, which is a very complex problem when identifying the boundaries. A third particular clinical problem that comes up is annotation. So annotation of signals within the circuit is a specific difficulty. So the power of the topological approaches, though, is that I think this is a general principle that we can apply to mapping, really, is that in general as a field we have been, and you come to the HRS, everyone's focused on the local electrogram. I think you can come and you meet the masters, you spend all day looking at the individual local electrogram, but actually that is often the hardest spot to annotate within the circuit. There's information, the power of these approaches is that it applies information which is distant from the areas that are difficult to annotate. So where the electrogram is easy to kind of, the local activation, so I think bearing in mind some of those clinical limitations, I think like, it's a practically applicable tool once you start using it. Thank you, Dr. Berenfeld. So it's now my pleasure to introduce Dr. Vadim Fedorov from the Ohio State University, to talk about substrate imaging and adenosine-challenged mapping in experiments in clinical atrial fibrillation. Thank you so much, it's really a pleasure to be there, thank you colleagues, and thank you so much for the invitation, it's a wonderful session. I'm so glad that we, I'm sorry, I really need to get my glasses, now we should go out. Here's my disclosure, so we have been working with Abbott Tapira for many years in term of validation in improvement multi-electrode mapping. So after all these meetings, we're very familiar that atrial fibrillation is very complex process, it's years taken to get remodeling before IF will start, it will never start in normal healthy human patient heart. And as so far, the only current, available, doable, as well as actually reliable method, it's pulmonary vein. There are multiple other methods exist to showing potential sources outside of pulmonary vein, who else are driving atrial fibrillation after we'll avoid and completely insulate now, completely insulate with PFA, practically destroy all this area, but it's still driving. And no clear methods, how exactly we can unmask, and this was the reason, like two years ago, we met in Iceland, beautiful place together with many of you present here, organized by a lot, Dr. Elad, Natasha DeGroot, so Omar was there, Natalia, so it was wonderful place when they actually discuss how exactly we have to proceed further, because as we need to clarify all specific definitions and mechanism of atrial fibrillation, that we will speak on the same language. So the whole aim was to actually provide advancement, where we're going with IF research, and where we're going, how we can improve atrial fibrillation treatment. So that's exactly what we do at our lab at Ohio State University, when for the last 14 years, we have been studying ex vivo human hearts, both donors and diseased hearts, when we actually have ability to completely optical map with infrared dye, image them with high resolution, both MRI, CT, and now PET scans, and actually have wonderful clinical studies. Because we are able, of more than 200 hearts, as well as many patients which we have, clinical large animal models, to see more clearly what exactly going on in our hearts, how we can figure out what exactly atrial human substrate, who actually creating and supporting atrial sources, driving atrial fibrillation outside of pulmonary vein. Because if we can define the sources, IF driver, source of localized electrical activity, critical to maintenance, it's such an ablation of this area, have to slow down at least, and then convert to ET, eventually terminate atrial fibrillation. The point is, we may have several of different drivers, and it's not easy to find them in clinical settings. So make sure I will acknowledge my team, because all this effort, it's enormous effort of many of students, post-docs, Ning Li, I'm glad that she's there, my former student Brian Hansen, Dr. Hammel is our main clinical electrophysiologist, our wonderful leadership, Peter Moeller, and of course, my buddy, my D. Chao, who actually been working on many different computational models and data analysis, which you probably already saw just today or previously. So with him together, we actually analyzed optically mapped human hearts, and then we scanned them with highest resolution possible at this moment, MRI up to 100 microns, and we can visualize atrial wall, he processed and analyzed atrial wall in their highest ever been details. As you can see, atrial, human atrial wall, it couldn't be really considered 2.5 millimeter, which is standard for example, data MRI analysis, because it's variable, both right and left and it's variable and can be extended up to 10-12 millimeters. And generally, it's in average, we see it's about 5 millimeter and array and about 4.4 mLA. With again, variability, which all showing this bundles, my architecture of atrial human wall, actually obstacle for multiple electrode mapping, any electrodes from both high-res APNN would still be suffering to visualize what we have in transmural activation, which we can do with infrared optical mapping, which allow us to see from both sides up to 8-1 centimeter into the depths. This advantage, leading that we more understand how exactly conduction would happen between APNN and the cardiome, how exactly it's traveling. Even during pacing, we already know in from clinical studies that transmural activation difference between conduction are significant, could be like 10-15 milliseconds and all significantly determined by anisotropy of conduction, by fibrillation, fibrosis. And that's the reason why during a fib, it's way more pronounced, it could go up to 100 milliseconds, both seen by this is a great figure, which make with Natasha Degroot in this review, figure 2 in vivo, x-vivo, intramural re-entry seen only from one side and only tiny bundles, which we actually define as there by high-resolution MRI. We can visualize transmural activation delay there and if you put electrodes right here, right on this bundle, we still actually can't see that. Meaning sometime even electrodes, even in vivo, we may see potential rotation, potential half of re-entry, but in vivo, this optical x-vivo, in vivo optical, we can see complete re-entry circuit. I would like to clarify that what we see is not a rotor, but it's actually specific micro-anatomic re-entry. Micro-anatomy means they have common paths and usually 2, 3, it could be returning loops. The reason for that, we have specific structure in the human heart, which predetermined its wall thickness variation, where we will get this driver substrate. Fibrotic tissue and AP endo-myofiber orientation is very different on endo versus AP and when we have exit, when we have transition between endo and AP, we actually very often see this myofiber twist transmural myofiber twist. The common observation not only in x-vivo, here it's in vivo optical mapping of canine, high-resolution during in vivo open chest, we can see consistent presence, even during when we have exit from other driver, presence of this line of the block, it's common re-entrant track with 2, 3 rotational back loop. It's very similar to what we will see in VT cases, when we have common track and it's nicely shown by Bill Stevenson in this review. Here's a common path entrance, we have several different loops, which lead into multiple type of visualization in ventricle tachycardia from AP or endo, but in vivo and x-vivo optical mapping, manage and merge with high-res MRI, we can see that actually what we see, it's again very similar situation. We just have several different drivers. And how can we map them? So it's a great study, which I'm going under review in European journal, together with Miguel Rodriguez, who's model our optically mapped drivers, simulate how exactly this, what type of resolution of electrode, what spacing, what contact, what combination must be done to see this tiny bundle, to see this re-entry. And it's here example, one centimeter resolution can visualize only, if you actually touch in closer. If you have move away, meaning poor contact, you already will see focal type of activation. And it's omnipolar showing similar stuff. We actually see combination of omnipolar and unipolar, what we probably the best option in clinical settings for high-res catheters. However, when you have high-res catheter, you don't see the full structure, you don't see the full picture, because the drivers are spatially stable, but they're temporarily unstable, because they have, we have quite oftenly seen optical mapping in our, in many patients too. We have competing drivers, present only one at the time, and they re-entrain each other, which leading to a question of how we can see in any real in vivo situation with a high-res, which is not possible to do simultaneously across both atria in vivo and patients. Such we have to look more on structure, we have to look more on specific structure. Here's example, it's persistently Fib hard, when we have three drivers, competing drivers, primary driver is 56% stability was present in inferior, it's shown by dominant frequency, the secondary drivers. And yes, you do have face singularity around this conduction block, but they're both of them not specific predictors. When we have high-res MRI, it's showing that all going, all the tracks, the main common track going along the fibers, along the fibers in fibrotically insulated area, as well as with wall thickness transition. So, with again C-Ciao, we are analyzed what exactly features we will have in this optically mapped driver region, and we see they are different between left and right region, because in the left drivers, they actually have thinner area compared to non-drivers on the left, but in right drivers, it's opposite. As such, I hope you saw his presentation in the morning, when he show now new approach, when he combining all possible feature, which we could extract from this X-Vivo optical map hard with this high-resolution images and model how the combination predict where exactly the source would be. Because it would be very important, if you know where the source, we can actually have targeted ablation, that's how we show the first time 10 years ago with Brian Hansen, we can define the source, ablate the track, then we have to of course extend it, and to prevent initiation of atrial tachycardia. Here's again figure 3, great study. Thank you, Natalia, for your patience with me regarding these figures and beautiful study by Natalia, showing similar general we can use in vivo MRI and define where the source, define predict where the source and then again do kind of this targeted line ablation. Well, in vivo it's great, but we both agree that we need more, we need transmural understanding of atrial wall, where we can do so far right now in X-Vivo and that's again the child work, when we have the driver optically mapped and he's showing what exactly feature atrial wall, variance, transmural, my fiber all together predict where the source and we can target and do smart ablation. So, how we can translate to in vivo, we have great tools both X-Vivo and then we have also in vivo, that just paper published by the child in expert system with application, when we actually can use artificial intelligence approach, helping us to get very similar outcome between X-Vivo, helping validate in vivo. Now we have the same wall thickness in using clinically available MRI scans. With X-Vivo validation, now we have combination of both wall thickness as well as transmural fibroids, that's actually patient showing where we have more clearly see the fibrotic substrate, because now we have AP and middle cartransmural. In the past, we just initially when we work with Dr. Hamill, we go in smart clinical trial, recruiting patient, we do MRI before and left atrial driver, we could clearly have correlation between fibrotic area and the driver source when Dr. Hamill later ablate, but we couldn't see by standard MRI in right atrials. We have to work to improve this to actually significantly increase visualization of the fibrosis, which was done recently and again with work of by the group, we could have now clinical MRI, AP and intramural analysis. We see that actually in males versus females, we have different distribution as well as where we have recurrence or not and the fibrotic tissue are higher in females and by analysis of 96 patients, it's by Dr. Hamill who will study, we can analyze what drivers we have, how they're distributed. It's interesting that female higher than 63 have higher right atrial drivers, which actually associated with endocardial fibrosis. As such, now we have to look not only in X-Vivo when we can do all this transmural wall, but also in Vivo, how we can find and predict where driver and how clinician to improve this. When I mentioned about six, it's important to mention that the sex always been equation, sex difference between male and female sexual demorphism predicted what type of arrhythmia we will have. We have more effip on female, more in male, but more tachycardia in females. It's one we actually studied. We just published this paper in circulation arrhythmia, electrophysiology, using their all are now gene approach as well as structure, cell distribution. We can predict what type of arrhythmia we'll have in male and female. We also found that important consideration pathway in right atria in female, we have more fibrotic profile, which predetermine why we have in female more subendocardial fibrosis during effip. But in males, we have more information, pathway factor already in control, healthy donor hearts when we have significant increase in male are prolonged through this pathway to get fibrosis, to get eventually remodeling. Where female, if you get effip, you'll get more fibrosis. Here's nice paper showing a combination of non-pulmonary focus coming mostly from superior vena cava in females rather than males. Again, because we have prone, we found this predisposition to autonomic activity using actually sexual demorphism. It's primarily T-box, 3 HCN channels. Where in male, we have more epicardial fat, more inflammatory pathway. That's why the actual sex have to be considered, because we may have different substrate between male and females. In addition, of course, we not only can have consider fat fibrosis, that new approach, FDG, and it was nice presentation, several from University of Washington group. We also have approach FDG scan. It's showing that together we need to consider inflammation, fibrosis, and fibrosis transmural atrial wall thickness altogether, which can help us target drivers, which clinically available, clinical available, may be multi-electrode, may be require additional structural substrate. In the last, I briefly can show that the question regarding adenosine. We're using adenosine all the time during now clinical scans. Adenosine is a great tool, which activate temporally action, ACH, IK2 channel, hyperpolarized membranes, wall down, even nodal conduction, sinus rhythm, as well as shortening APD. And we show in past that in humans, we can actually provoke vis-a-vis easily atrial fibrillation. It's heterogeneous shortening and blocked by a specific blocker. Where we actually apply during AFib, we actually modify dynamics. Importantly, if in cases we don't see well by one only surface, for example, epicardial complete re-enter circuit, during ADO we can improve stability because we unmask more common paths, we diminish numbers of returning paths to visualize more clearly where we need to do ablation. The same we've done in translate in clinics. It's important that adenosine also helping to, in clinical settings, diminish an amount of R-wave. You can see that due to AV block, temporal AV block, during this period we have not only stabilization of activity and full complete re-enter circuit, but as well as full pulse because it's easy to map using unipole electrodes. Here example where before it wasn't possible in a clinical setting visualize, it was the period, the same basket, 64 electrodes. But then we unmask and successfully ablate this patient still after five years free from atrial fibrillation. But it's not only could improve in case we already have stable driver. Adenosine can destabilize, which also makes sense by clear consistent re-enter mechanism. Such you have to consider if you need improvement, you can use adenosine to see it better. And that's indeed useful too in addition to all structural imaging preceding multi-electrode mapping. So in conclusion, I would like to emphasize that we always have to consider patient-specific characteristics, sex, age, hormone, as well as our 3D structure when we're planning targeted ablation. So it must be combination of multi-modal imaging, now great approaches, multiple group developing AI. But it's when we're using multi-modal imaging, we have to resolve atrial wall thickness, transmural and myofibrillation because they are key predetermined where we'll have driver substrate. In the case here, we actually consistently helping in vivo approaches using X-vivo, which allow us to improve consistent and validate any new imaging AI approaches which we have to treat our atrial fibrillation patients. Thank you so much. I really appreciate all your time and all colleagues. Thank you. Thank you. We're a little bit past time, but we might have time for just one very quick question. Thanks for such a rich presentation. I'll try to be short, two quick questions. With optical mapping that you applied on explanted human hearts, I think you showed that it was done on the right atrium, but typically with explanted hearts, we don't get much of the atria. So how do you manage that to do it? So we have great access to donor's hearts, which are intact. And one of the reasons why we're receiving these donor hearts is because they have atrial fibrillation. So in fact, we have 40% of donor hearts which we receive, they have different stage of atrial arrhythmia, and specifically FIB. That happened as soon as diagnosed in the heart, like 30, 40 years, females, sometime we had getting new, they actually have a FIB that wouldn't be used for transplant. As such, we do have complete intact with pulmonary vein. Whatever I presented today, it was all intact hearts. But yes, in the failing hearts, we wouldn't have pulmonary vein. But our surgeons preserve whole right atria based on all appendage, Bachman bundle, except pulmonary vein region, posterior wall, we have entire intact heart. Thanks. I'll have to convince my students to do that. Sorry, sorry, we might have to move on. Yeah, thank you so much. Our next speaker is Dr. Carolyn Roney from Queen Mary's College in London, who will be talking to us about computational modeling for mapping of atrial fibrillation. Great. Yeah. So thanks very much for the invitation to speak today. I've had some really great talks, so. Great. Yeah. So I'll be speaking about computational modeling for the mapping of atrial fibrillation. So first, a bit of motivation for why we use computational modeling for personalizing AF therapy. So we're working with increasingly large electrical data sets and also imaging data sets. And we're working at both the population level to compare treatments across large in silico or virtual trials, and also at the patient-specific level. So to personalize treatment approaches for an individual. And we then aim to combine these approaches at the population level with the patient-specific models with the aim of personalizing treatment approaches. Now, there are multiple challenges associated with mapping AF, which computational modeling might help us to overcome. These include that we're working with electronic mapping data from catheters with a limited spatial resolution or spatial coverage, which can make it hard to map AF, which is spatiotemporally chaotic. Also, the atria displays complex restitution properties, meaning that it's hard to translate sinus rhythm data to AF dynamics. We also have a large heterogeneity across the atria and also between patients. Our measurements are uncertain in their location and signal. Finally, we've got a large amount of data that we would like to combine to predict long-term outcome for AF patients. So we'll now discuss how computational modeling can help us to overcome these challenges. So here, I'll focus on four areas, starting with how we construct personalized atrial models. So we have multiple steps involved in creating models or digital twins updated over time. So the data that was used for constructing and calibrating these models include CT, MRI, electro-anatomic mapping, and wearable data. These data are processed to create models and used to develop personalized therapies for mechanistic studies and also for large insulocode clinical trials. Now, the first step here is to create an anatomical model, which could be a single layer. It could be coupled endocardial and epicardial layers, or it could be volumetric, as we've just seen. Now, the recent STACON Bioatrial Segmentation Challenge, led by Juchow, provided a large data set for training deep learning algorithms to automatically segment the left and the right atria with our proposed model using an ensemble of five residual encoder unit models. So once we've got our segmentations, we can then clip them, we can add fibrosis to them, and we can also add atrial regions and fibers. Once we've got the anatomical model, we then have to think about how to calibrate the model to match clinical electrophysiology data. So these models could be calibrated to local activation time maps or voltage maps, measures at different pacing rates. So what Kat's done here as part of her PhD is she's investigated how the model predictions depend on the data that you've used for the calibration. And what she found was that we got more rotational activity in the models when we calibrated to a pacing rate of 250 milliseconds, so closer to AF, than if she calibrated to a local activation time map from 600 milliseconds, so closer to sinus rhythm. And she didn't see this effect when she calibrated instead to bipolar voltage. Models can also be calibrated to late gadolinium enhancement, as we've seen. And so here what Mahmood has been looking at is how do our models change if we calibrate them to late gadolinium enhancement compared to AF data, compared to coronary sinus data, coronary sinus paced data with either unipolar, bipolar, or omnipolar voltages and activation time maps. And these maps capture different aspects of the AF substrate, with the late gad likely capturing more structural remodeling rather than functional. And then from the electron atomic mapping, we also have fixed and functional remodeling depending on the rate. Interestingly, what we found is that when we compared the model calibrations, Mahmood found that the number of areas of rotational activity and wave break was higher in the models that he calibrated to conduction velocity data compared to using voltage or late gadolinium enhancement. Now all of this calibration needs us to incorporate the effects of uncertainty in electric location and signal through uncertainty quantification methods. So what Kat was doing here was we were using uncertainty quantification methods to calibrate models to atrial fibrillation data. Now obviously, AF data has the advantage that incorporates restitution effects, also direction-dependent effects, but it's also a lot more complicated to use for model calibration. So we've really been looking here at how do we calibrate the AF cycle length and the AF conduction velocity. So now that we've got our personalized models, we can use them to test patient-specific therapy responses. So one use of computational models is identifying metrics that can be measured clinically for guidance therapy. So as an example, here we created 50 patient-specific left atrial models, and we compared different types of ablation. We then performed a Shapley analysis for determining which features were most important in predicting patient outcome to the different ablation approaches. We also focused on pulmonary vein isolation outcome, and we used modeling to test the effects of catheter resolution and size on the ability of flow and face singularity metrics to predict PVI outcome. Models can also be used to test the effects of the catheter recording resolution on the observed AF dynamics and for evaluating different post-processing algorithms in this context. So we can simulate different catheter arrangements in different locations with either unipolar, bipolar, or omnipolar electrogram modalities. We can also use these models to help us mechanistically understand the effects of patient-specific fixed and functional remodeling on AF dynamics, including the locations of rotational drivers. So as well as evaluating these different catheter ablation therapies, modeling can be used to assess the likely AF burden for an individual patient with different antiarrhythmic drugs. We've got an example pipeline shown here. Now, within silico clinical trial approaches, you should first try and reproduce clinical trial results before new treatments can be evaluated under the same framework. So what Sam has done here as part of her PhD, she's compared antiarrhythmic drug success rates to the ESE guidelines to demonstrate that the models that incorporated the fibrotic remodeling may represent a suitable trial cohort. So in parallel to patient-specific modeling studies, large clinical in silico trials may guide clinical trial design, reduce patient numbers, and capture more heterogeneity. So we can use these virtual cohorts for in silico trials. So we've got an example here where we were motivated by the DCAF2 trial where left atrial fibrosis ablation with pulmonary vein isolation didn't improve AF outcome. But what we wanted to see is, are there a cohort of patients where actually biatrial fibrosis ablation could improve the AF ablation outcome? So we did this as a large virtual in silico trial. So we compared the following ablation types, so pulmonary vein isolation, left atrial fibrosis ablation, right atrial fibrosis ablation to biatrial fibrosis ablation. And we've got an example here to show you more or less what we found, which is that we've got one with low right atrial fibrosis at the top, where AF terminates when we apply the left atrial fibrosis ablation. But if we have the same model, but instead we have an increased amount of fibrosis in the right atrium, we see that AF sustained to the left atrial ablation, but it terminates when we had the biatrial fibrosis ablation. And we then quantified this across the 4,000 cases as a virtual trial. How about if we try to use in silico trials to test new ablation approaches that utilize novel AF mapping methods? So what Ovais has done here as part of his PhD research is he started with an iterative substrate-based ablation approach before he automated the process to test this across a large virtual cohort. This is in silico trial-based approach and motivates a clinical trial into this ablation approach. So specifically, we're proposing the novel electro-optic flow mapping and ablation strategy that combines electro-physiologic flow and optic flow analysis to optimize this ablation site selection while trying to minimize the ablated tissue area. So Ovais developed a consensus approach that accounts for the fact that AF dynamics are changing over time to combine targets from multiple AF episodes into a single consensus map. He then developed an automated process to apply these ablations across the cohort and to determine the acute outcome and also to test the inducibility. Finally, the in silico trial results are shown here where the consensus electro-optic flow mapping targets resulted in the largest ratio of AF non-inducibility to ablated tissue area. And the other approaches that he considered led to a lower non-inducibility and a higher ablated tissue area, so highlighting the advantage of this approach. So this shows that computational modeling can be used to develop new techniques for AF mapping and novel ablation approaches. And our next steps are to extend this work to different catheter arrangements. So finally, we'll look at combining these patient-specific models with population-based approaches to look at predicting AF burden and patient trajectories over time. So our aim here is to predict patient-specific long-term AF recurrence after ablation using large cohorts. So because we found that clinical patient history data alone, imaging data alone, or simulation data alone didn't predict AF recurrence accurately, we used an approach of combining these different data sources. Specifically for each patient-specific model, we ran a series of simulation stress tests with different model setups to represent the uncertainty we had in our data. And we then trained machine learning classifiers across the population to predict whether AF recurred or not. These models were constructed from late gadolinium enhancement data for 100 AF patients. And our simulations were run using OpenCarp, which is a fantastic open resource. So in the spirit of open resources for the community, we released our pipeline for creating atrial models at scale as a cardiovascular use case for the European Digital Twins in Healthcare project. We included instructions and considered the requirements for a clinical end-user. Now to move our framework to the clinic, predictions need to be on clinical timescales. So with this motivation, Alexander developed this deep learning approach to predict AF outcome from biophysical simulations. The approach combined different anatomical, structural, and functional tissue feature maps. And you can find out more about Alexander's post and also our recent publication. So finally, training these deep learning approaches and performing in silico trials requires a large number of different virtual patients. So to overcome the limitations of clinical data privacy and data set size, we... Sorry, I... Okay. Just getting a warning about time. I think maybe we've gone over the time for the slot. Okay. Okay. Okay. Sorry. So yeah, we compared to synthetic distributions to our real distributions. Okay. And we... As we know, we can also apply these approaches for ventricular tachycardia therapy. I'd like to thank everyone in our lab and also all of our collaborators. Thank you. Thank you for your time. Thanks, Carolyn. That was an excellent presentation. Are there any questions from the audience? Do you have any sex difference? You mentioned about 100 models. Do you see any sex difference in right, left fibroids? We haven't looked into it in enough detail. We had it as a predictor in the model, whether male or female, but we didn't look into... We haven't looked into that yet, but we definitely should. Thank you. Yeah, thank you. I've got a question, which is, which do you see is the closest clinical application? Something that is about to be used or we would use, actually, in clinical practice out of your work? We'd love that some of the ablation approaches that we're suggesting here are now tested in prospective trials. At the moment, we are applying them on retrospective data, but we're now hoping to move them to prospective. But I think, yeah, Professor Trenove is really kind of leading the way on that, so we're hoping to get there soon. If there's no other questions, then it's my pleasure to thank all of my co-speakers and thank Dr. Barenfeld for putting this session together, and thank you all for coming.

complex arrhythmias

atrial fibrillation

advanced imaging

computational modeling

multifunctional imaging

ablation therapies

ischemic cardiomyopathy

topological approaches

adenosine stabilization

personalized AF treatments