WEBVTT 00:00.000 --> 00:03.600 Hi, my name is Wassim Brumschild and I have the pleasure of presenting 00:03.600 --> 00:07.160 a short introduction, short summary of this project entitled 00:07.160 --> 00:10.520 Developing Dynamic Digital Twins of the Human Heart. 00:10.940 --> 00:14.940 And this is a hugely and immensely interdisciplinary project together 00:14.940 --> 00:16.120 with my colleague Axel. 00:17.300 --> 00:20.900 Yeah, so let me start with the motivation for this problem. 00:20.960 --> 00:24.680 So there's a disease called atrial cardiomyopathy, which means that 00:24.680 --> 00:29.520 the atria, the two upper chambers of our heart, are affected by 00:29.520 --> 00:33.600 fibrotic infiltrations, which does lead to complications. 00:34.320 --> 00:38.600 And the most obvious one is that it increases the risk for atrial 00:38.600 --> 00:42.200 fibrillation, which is the most common sustained cardiac arrhythmia. 00:42.920 --> 00:47.560 Atrial fibrillation, in turn, increases the risk for stroke, so a 00:47.560 --> 00:51.180 lesion in the brain, with severe consequences. 00:52.180 --> 00:55.320 But nowadays there is increasing evidence that also atrial 00:55.320 --> 00:58.720 cardiomyopathy itself can increase the risk for stroke. 00:59.160 --> 01:04.200 So obviously there is a huge need to quantify atrial cardiomyopathy 01:04.200 --> 01:09.600 for risk stratification, both for atrial fibrillation and for stroke. 01:10.760 --> 01:16.320 The best way to diagnose atrial cardiomyopathy, and also to grade it, 01:16.680 --> 01:19.900 is to acquire what are called electroanatomical maps. 01:20.420 --> 01:24.360 For this, a catheter is inserted inside the heart, and the electrical 01:24.360 --> 01:27.680 signals of this tip of the catheter are recorded. 01:28.280 --> 01:33.320 The catheter is tracked in 3D space, and we can acquire a 3D 01:33.320 --> 01:37.780 anatomical model together with this functional information about the 01:37.780 --> 01:39.340 electrograms and their amplitude. 01:40.040 --> 01:44.960 And those parts of the atrium that exhibit a small amplitude, so 01:44.960 --> 01:49.000 -called low voltage areas, are directly affected by atrial 01:49.000 --> 01:49.820 cardiomyopathy. 01:50.350 --> 01:55.600 So that's a reliable way of staging atrial cardiomyopathy, but 01:55.600 --> 01:57.280 unfortunately it's quite invasive. 01:57.580 --> 02:00.420 It's minimally invasive, but still you need to insert a catheter 02:00.420 --> 02:01.200 inside the heart. 02:01.500 --> 02:04.820 But it's nothing you want to do on a regular basis, say every six 02:04.820 --> 02:06.000 months or every year. 02:06.520 --> 02:10.880 On the other hand, one can record body surface electrocardiograms. 02:10.940 --> 02:15.320 So the normal ECG that's acquired routinely without any invasive 02:15.320 --> 02:15.800 measures. 02:16.640 --> 02:21.540 And this could be done many, many times, but it only gives us a rough 02:21.540 --> 02:24.320 picture of what's going on inside the heart electrically. 02:25.020 --> 02:31.060 And our idea here is to start this digital twin of a patient, a 02:31.060 --> 02:35.020 virtual computer model of the individual patient from this invasive 02:35.020 --> 02:40.840 electron anatomical mapping procedure, and then use routinely acquired 02:40.840 --> 02:44.260 ECGs to update this digital twin over time. 02:44.660 --> 02:48.620 So to get to a longitudinal digital twin, which can then be used to 02:48.620 --> 02:53.500 assess, for example, the risk for atrial arrhythmia and also for 02:53.500 --> 02:58.900 markers that increase the risk for stroke using methodology that we've 02:58.900 --> 03:04.120 established here at the Institute for Atrial Fibrillation already. 03:04.980 --> 03:09.460 And this would allow us to build this detailed digital twin from the 03:09.460 --> 03:13.060 invasive data and then update it with the ECGs. 03:13.460 --> 03:16.320 And that's, of course, a very challenging problem mathematically. 03:17.060 --> 03:20.980 And that's where Sebastian Kumpsche's expertise comes in. 03:21.180 --> 03:25.120 As Axel already mentioned, it's a hugely complex problem, and 03:25.120 --> 03:28.220 unsurprisingly, it comes with many mathematical and computational 03:28.220 --> 03:28.780 challenges. 03:29.720 --> 03:34.280 For one, it's an inverse problem, meaning you want to infer patient 03:34.280 --> 03:39.580 -specific information from noisy, for example, ECG measurements. 03:39.920 --> 03:43.460 And in fact, you want to do this sequentially over a certain duration 03:43.460 --> 03:44.020 of time. 03:44.380 --> 03:48.460 These are known to be highly, highly complicated and challenging to 03:48.460 --> 03:48.720 solve. 03:49.320 --> 03:52.540 And this will be one of the key mathematical challenges we're going to 03:52.540 --> 03:53.520 deal with in this project. 03:54.140 --> 03:57.820 One way of addressing those problems is through a so-called 03:57.820 --> 04:00.860 statistical inverse problem, particularly Bayesian perspective. 04:01.720 --> 04:06.040 And these Bayesian perspectives offer so-called optimization-based 04:06.040 --> 04:06.600 procedures. 04:07.540 --> 04:12.640 But these optimization principles, they rely on non-convex principles. 04:12.980 --> 04:17.640 And this requires, therefore, advanced mathematical treatment to gain 04:17.640 --> 04:19.060 any useful insights. 04:19.460 --> 04:23.840 In fact, the insight that can be gained is also something we are going 04:23.840 --> 04:28.080 to address here, which is relating to the so-called identifiability 04:28.080 --> 04:29.160 question. 04:29.520 --> 04:34.380 Namely, can we give the, for example, local activation times of 04:34.380 --> 04:38.000 minimally invasive measurements together with the ECG measurements, 04:38.440 --> 04:43.640 can we actually identify the parameters that we're interested in? 04:43.940 --> 04:47.340 And if we can, so of course, the natural question is, how should we 04:47.340 --> 04:51.420 actually place sensors to gain the most valuable information? 04:53.140 --> 04:57.820 Lastly, we want to bridge the gap towards practitioners. 04:58.280 --> 05:01.760 So we want to perform this inference, that's our long-term version, to 05:01.760 --> 05:07.200 perform this inference at a scale that is of interest for real-world 05:07.200 --> 05:09.380 clinical settings. 05:09.860 --> 05:14.620 This means we want to have scalable inference with together with 05:14.620 --> 05:16.580 realistic uncertainty quantification. 05:17.100 --> 05:20.500 And secondly, we want a robust method, we want to develop a robust 05:20.500 --> 05:26.780 method that can handle also discrepancies in perturbed clinical data. 05:28.320 --> 05:32.160 As you may have already seen from this little outlook here, this is a 05:32.160 --> 05:34.600 highly interdisciplinary research project. 05:35.320 --> 05:38.780 As a candidate, we would therefore also be part of two research 05:38.780 --> 05:39.140 groups. 05:39.800 --> 05:43.040 On the one hand, there's this uncertainty quantification research 05:43.040 --> 05:47.160 group where we develop new mathematical methods for treating and 05:47.160 --> 05:51.480 assessing the effects of uncertainties at the interface of highly 05:51.480 --> 05:53.800 complex computation models and data. 05:54.120 --> 05:59.640 Our toolbox consists of tools from applied mathematics, scientific 05:59.640 --> 06:03.060 machine learning, but also we heavily rely on high-performance 06:03.060 --> 06:04.120 computing settings. 06:06.000 --> 06:11.640 But secondly, we also specialize in tailoring our methods to 06:11.640 --> 06:16.040 applications, to problems like the project together with Axel Hoeber 06:16.040 --> 06:22.260 on a digital twin, where we tweak and tune our methods to the problem 06:22.260 --> 06:22.760 at hand. 06:24.840 --> 06:29.580 The second group in which we would be integrated is the computational 06:29.580 --> 06:33.560 cardiac modeling group at KRT's Institute of Biomedical Engineering, 06:34.000 --> 06:37.100 where our mission really is to use engineering methods of 06:37.100 --> 06:40.840 computational modeling to ultimately improve patient care and 06:40.840 --> 06:41.140 medicine. 06:41.680 --> 06:47.060 We work towards this vision by developing methods, advancing methods 06:47.060 --> 06:51.380 on the one side, and then on the other hand, applying them often in 06:51.380 --> 06:54.840 collaboration with clinicians in computational studies. 06:55.200 --> 06:59.920 The way we do this is as open as possible, often in a very 06:59.920 --> 07:01.740 collaborative environment also. 07:02.440 --> 07:06.220 And as a group, we've defined a set of six values that are important 07:06.220 --> 07:06.740 to us. 07:07.260 --> 07:10.960 You can find more information, of course, on our websites and on 07:10.960 --> 07:11.600 social media. 07:11.600 --> 07:18.460 But most importantly, what's in it for you and what's the background 07:18.460 --> 07:19.500 we are looking for. 07:19.940 --> 07:23.120 I think it's going to be important to have a strong background in 07:23.120 --> 07:27.420 applied mathematics or a closely related field with some of the 07:27.420 --> 07:29.860 aspects that were mentioned before already. 07:30.560 --> 07:35.660 You should not be afraid of programming, as all of this research will 07:35.660 --> 07:40.920 be computational and this happens through programming. 07:41.880 --> 07:46.460 The second thing you should not be afraid of and actually being also 07:46.460 --> 07:51.180 curious to try is working in this interdisciplinary environment, where 07:51.180 --> 07:54.320 it's sometimes a little extra effort to find the same language. 07:54.800 --> 07:57.940 But from experience, I can tell that this is really worthwhile. 07:58.960 --> 08:03.120 And the computational cardiology aspect is something that's an 08:03.120 --> 08:06.880 advantage if you have already had contact with that, but it's by no 08:06.880 --> 08:08.080 means a requirement. 08:08.280 --> 08:10.720 This can also be learned along the way. 08:11.240 --> 08:15.460 If this sounds interesting for you, then yeah, don't hesitate to 08:15.460 --> 08:17.120 apply, to join us. 08:17.500 --> 08:20.540 And we are very much looking forward to your application.