14:30   Cardiac I Chair: Rik Vullings Predicting Ventricular Tachycardia, taking time into the equations Carlijn Buck, Marcel van 't Veer, Wouter Huberts, Frans van de Vosse, Lukas Dekker Abstract: Post-myocardial infarction (MI) patients have a risk of developing ventricular tachycardia (VT) years later. Primary risk stratification, based on left ventricular ejection fraction and symptoms, proves to be insufficient in identifying high risk patients. We postulate that Digital Twins (DTs) which integrate time-evolving pathophysiology through hybrid models possibly lead to better VT prediction. A hybrid approach of interest is the fast-scale-slow-scale model of Regazonni et al. (Int. j. numer. method. biomed. eng., 2021, 37(7)). Here, physics based models are used to predict diseases in the next seconds (fast time scale). However, over months or years (slow time scale), patient change, and thus the physics based models parameters remodel. This remodeling is unknown but could be learned by data-driven models. While fast-scale physics-based VT prediction models are established, understanding the slow-scale dynamics remains underexplored. In this retrospective study (1997-2023), laboratory data, vital functions and echocardiogram reports from 175 VT and 2703 control MI patients without VT were collected to explore the time-evolving trends. This resulted in 459 laboratory, 289 vital functions and 362 echocardiogram parameters with 1.33 million, 866950 and 270643 values respectively. Exploratory data analysis focused on control-VT differences at four time points: one day, three months and one year after MI, and one year before VT or last available parameter. Utilizing clinical data proved challenging due to its inherent sparsity, heterogeneity and occurrences of missing data, resulting in a steep decline in available parameters and patients over time. Despite these challenges, insightful trends showed in expected parameters (left ventricular volume-related parameters) and unexpected parameters (cholesterol-related parameters). This exploration into time-evolving trends demonstrates potential for identifying patients at risk for VT. Furthermore, it emphasizes the need to take time into account in DTs. Future steps include using linear mixed models and AI to potentially identify patients at risk for VT. Sequential Diffusion Models for Cardiac Ultrasound Image Reconstruction Tristan Stevens, Oisín Nolan, Jean-Luc Robert, Ruud van Sloun Abstract: Diffusion models have quickly risen in popularity for their ability to model complex distributions and perform effective posterior sampling. Unfortunately, the iterative nature of these generative models makes them computationally expensive and unsuitable for real-time sequential inverse problems such as ultrasound imaging. Considering the strong temporal structure across sequences of frames, we propose a novel approach that models the transition dynamics to improve the efficiency of sequential diffusion posterior sampling in conditional image synthesis. Through modeling sequence data using a video vision transformer (ViViT) transition model based on previous diffusion outputs, we can initialize the reverse diffusion trajectory at a lower noise scale, greatly reducing the number of iterations required for convergence. We demonstrate the effectiveness of our approach on a real-world dataset of high frame rate cardiac ultrasound images and show that it achieves the same performance as a full diffusion trajectory while accelerating inference 25×, enabling real-time posterior sampling. Furthermore, we show that the addition of a transition model improves the PSNR up to 8% in cases with severe motion. Our method opens up new possibilities for real-time applications of diffusion models in imaging and other domains requiring real-time inference. A fluid-structure interaction approach to distinguish between true and pseudo-severe aortic stenosis Sabine Verstraeten, Martijn Hoeijmakers, Frans van de Vosse, Wouter Huberts Abstract: Valve replacement therapy is recommended for patients with severe aortic stenosis (AS), defined as mean transvalvular pressure drop (Δp) > 40 mmHg, and aortic valve area (AVA) < 1 cm2 [1]. However, 50% of the patients [2] with AVA < 1 cm2, exhibit Δp < 40 mmHg (low-gradient AS). This low Δp may result from reduced cardiac output, preventing complete valve opening. Therefore, dobutamine is administered to increase flow. If AVA remains small, but Δp exceeds 40 mmHg, it is true severe AS, recommending valve replacement. If AVA exceeds 1 cm2 and Δp remains small, it is pseudo-severe AS, where valve replacement is ineffective. Unfortunately, 30% of low-gradient AS patients, do not respond to dobutamine, resulting in insufficient flow increase, leaving AS severity undetermined. Patient-specific fluid-structure interaction (FSI) simulations of the aortic valve [3] can impose higher flows independent of the patient’s cardiac function. This study aims to leverage FSI simulations to distinguish between true and pseudo-severe AS. FSI simulations were performed in LS-DYNA, using an aortic valve geometry of a “low-gradient” aortic stenosis patient. Blood was modelled as an incompressible Newtonian fluid. A flat, sinusoidal velocity profile (with negative values set to 0 Pa) with a cycle time of 0.91 s was prescribed at the inflow boundary. Several simulations were performed with maximum velocities ranging between 0.4 and 1 m/s. The outflow boundary pressure was set to 0 Pa. The leaflets were modelled as a hypoelastic material and the properties were assumed to be known. Forces and displacements were exchanged between fluid and mechanical solver through strong coupling. Mean Δp remained below 40 mmHg for all inlet velocities. AVA exceeded 1 cm2 at inlet velocities of 0.5 m/s and higher, indicating a pseudo-severe aortic stenosis. This FSI approach successfully differentiated between true and pseudo-severe AS, though a limitation was the assumption of known material properties. Future work must focus on determining patient-specific material properties. Currently, the FSI approach is validated against experiments with silicone valves. Once validated, patient-specific FSI simulations could significantly enhance clinical decision-making regarding valve replacement. The Frank-Starling Mechanism Characterized Using Photoplethysmography Roel Montree, Elisabetta Peri, Xi Long, Reinder Haakma, Lukas Dekker, Rik Vullings Abstract: The Frank-Starling mechanism describes the stroke volume of a heartbeat depending on the preload [1]. Clinically, the preload is estimated using the left ventricle end-diastolic pressure (LVEDP). As the LVEDP increases, the stroke volume increases linearly. However, this relationship saturates, and the stroke volume does not increase under circumstances of high preload. This is due to the maximum limit the heart muscle can extend and contract. Therefore, the Frank-Starling mechanism highlights the contraction force of the heart muscle based on the time since the last contraction. Because of this, it is often described as the force-interval relationship (FIR) [2]. A prime example of the Frank-Starling mechanism is during a premature ventricular contraction (PVC) and the beat following it that is enhanced by the post-extrasystolic potentiation (PESP) [3]. These types of beats have an exaggerated response able to help characterize the mechanism. However, these types of beats are unlikely to occur. This is alleviated by the continuous recording photoplethysmography allows. Photoplethysmography (PPG) is a non-invasive, non-obtrusive measurement able to measure blood volume under the sensor. It is often integrated in modern wearables, allowing continuous recording of an individual. A dataset consisting of ten individual patients for which both the PPG and ECG, as well as invasive blood pressure were recorded, for a total of 355 minutes of data, in which 27258 individual pulses were identified and categorized whether they were the result of normal rhythm, a PVC or the PESP, using the ECG signal to label them. Several parameters such as amplitude and maximum gradient are calculated and used to describe the FIR of a patient. Using the invasively acquired arterial blood pressure as a reference, it is shown that PPG can be used as a substitute to non-invasively characterize the Frank-Starling mechanism of the heart within an individual. Considerations need to be made, as the relationship is less obvious due to inherent sources of noise within the measurement and the non-invasive acquisition, but the inclusion of rare beat types enforces the characterization. [1] A. V. Delicce, & A. N. Makaryus (2023). Physiology, Frank Starling Law. In StatPearls. StatPearls Publishing. [2] D. J. Sprenkeler & M. A. Vos (2016). Post-extrasystolic potentiation: Link between Ca2+ homeostasis and heart failure?. Arrhythmia & Electrophysiology Review, 5, 20–26. https:doi.org/10.15420/AER.2015.29.2 [3] W. G. Wier & D. T. Yue (1986). Intracellular calcium transients underlying the short-term force-interval relationship in ferret ventricular myocardium. The Journal of physiology, 376, 507–530. https://doi.org/10.1113/jphysiol.1986.sp016167 Electrocardiographic Imaging Without Imaging Joël Karel, Tiantian Wang, Kurt Driessens, Job Stoks, Matthijs Cluitmans, Paul Volders, Pietro Bonizzi, Ralf Peeters Abstract: Electrocardiographic Imaging (ECGI) is a noninvasive technique that reveals the propagation of potentials on the heart surface. In the standard approach, imaging such as CT or MRI is needed to obtain a torso/heart geometry, limiting possibilities for large-scale cost-effective deployment. An approach was developed to estimate heart surface potentials solely from body surface potentials and a torso geometry, which can be easily obtained with e.g. a camera. This approach relies on projecting both the epicardial and torso potentials onto a rectangular image for each time instant. For the torso, this consists of a projection onto a cylinder, which is then opened up into a rectangular image. For the ventricular epicardium, it consists of a projection onto a bullseye plot, followed by an unwrapping to a rectangular image based on the polar coordinates of the bullseye plot. This approach allows for maintaining a consistent location of potentials in the two different plots. A deep learning architecture based on a Pix2Pix network is then used to perform an image-to-image translation for all time instants concurrently. In essence this is a sequential conditional Generative Adversarial Network (scGAN). The network was trained on 8 healthy subjects as well as 22 idiopathic ventricular fibrillation (IVF) patients and tested on 3 healthy subjects and 7 IVF patients. When compared with reconstructions from standard ECGI, this approach achieved an average Mean Absolute Error (MAE) for the heart surface potential maps (HSPMs) of 0.012 ± 0.011, and an average similarity index measure (SSIM) of 0.984 ± 0.026. For the electrograms (EGMs), the average MAE obtained was 0.004 ± 0.004, and the average Pearson Correlation Coefficient (PCC) 0.643 ± 0.352. The absolute mean time differences between estimated and reference activation and recovery times were 6.048 ± 5.188 ms and 18.768 ± 17.299 ms respectively. These results demonstrate a performance comparable to standard ECGI without the need for CT/MRI. This approach could allow for the deployment of ECGI for first screening or patient follow-up, where standard ECGI is not feasible. Echo-particle tracking velocimetry for quantification of high velocity flows Yichuang Han, Pritesh Ramya, Mathieu Pourquie, Selene Pirola, Johan G. Bosch, Jason Voorneveld Abstract: Vector flow measurements across cardiac valves provide diagnostic insights into valvular conditions, yet kernel-based blood flow estimators often underestimate peak velocities in regions with high spatial velocity gradients. Lagrangian particle tracking techniques offer a potential solution. This study explores the potential of echo-particle tracking velocimetry (echoPTV) in regions with high-velocity gradients within large vessels and heart chambers. Three flow models: 1) a steady parabolic jet (peak velocity 1m/s) entering a tank of quiescent fluid, 2) flow in carotid using computational fluid dynamics (CFD), 3) LV jet flow based on tomographic PIV measurement were simulated, with sparse scatterers distributed uniformly at the jet entrance. Ultrasound RF data was simulated and beamformed using the Ultrasound Toolbox. Hungarian tracking and Kalman-based tracking with a constant acceleration model, were compared. Results show echoPTV's capability in accurately estimating jet flow velocities, with Kalman tracking outperforming Hungarian tracking by recording longer tracks and achieving lower normalized velocity bias. However, Kalman tracking exhibited challenges during initialization, suggesting the need for improved filter state initialization strategies for further enhancement.

end %-->