13:30   Cardiovascular An In Silico clinical trial as an alternative for a traditional clinical trial: A feasibility study. Petrus L.J. Hilhorst*, Rajarajeswari Ganesan, Frans N. van de Vosse, Marcel van 't Veer, Pim Tonino, Wouter Huberts Abstract: Background: In Silico clinical trials have great potential for (partially) replacing clinical trials that aim to evaluate the safety, efficacy, and usability of new diagnostic tools or medical devices. In this study, we aim to demonstrate the feasibility of conducting In Silico clinical trials by generating virtual patients, and subsequently, reproducing a clinical trial in which the clinical benefit of fractional flow reserve (FFR) measurements was demonstrated (i.e., the FAME I study) for patients suffering from coronary artery disease. Here, we will present the strategy we envision to achieve our research goal. In addition, we will present preliminary results regarding data collection, processing, and model development. Approach: First, we have developed a one-dimensional pulse wave propagation model that is capable of computing the patient-specific FFRs observed in the FAME I study. Sensitivity analysis and uncertainty quantification is being used to determine the model parameters that are needed for model personalization. Geometric information can be extracted from angiograms, whereas unknown parameters such as the loss coefficient defining the pressure loss across a stenosis, can be estimated patient-specifically by using a machine-learning model that is previously trained using angiograms and pressure losses across stenoses. The latter is either based on actual measurements or 3D computational fluid dynamic simulations. Secondly, the derived parameters can be varied to generate virtual patients. Non-physiological virtual patients can be filtered out based on a priori selected filter conditions. In the future, the model output will be transformed into a clinically relevant output (i.e., mortality and morbidity) through a transfer function. Furthermore, we will evaluate our approach on an independent set of real clinical trial data (i.e., the FAME III study). Externally-induced shear waves in the right ventricular free wall during the cardiac cycle Luxi Wei, Rahi Alipour Symakani, Annette Caenen, Lana B.H. Keijzer, Daphne Merkus, Beatrijs Bartelds, Yannick Taverne, Antonius F.W. van der Steen, Hendrik J. Vos, Mihai Strachinaru Abstract: Increased right ventricular (RV) diastolic stiffness is linked to adverse prognosis in pulmonary arterial hypertension. Currently, a non-invasive method for measuring RV stiffness is lacking. In the left ventricle, shear wave (SW) speed is related to diastolic stiffness. We show for the first time that SWs in the RV free wall (RVFW) can be induced and imaged transthoracically with ultrasound. SW imaging was performed using an L7-4 array (ATL, USA) with a Vantage 256 system (Verasonics, USA) in a 5-weeks-old Yorkshire-Landrace pig (±120 beats/min). SWs were induced by a push beam (f0 = 4.5MHz, push duration = 800µs) on the RVFW and its propagation was imaged (f0 = 5.2MHz) using plane wave compounding (-12°, 0°, +12°) at 3 kHz framerate. Three acquisitions of one second were performed, where fourteen SWs were sequentially induced during each acquisition. SW propagation speeds were calculated using a semi-automatic pipeline that includes tissue velocity estimation along a manually traced spline on the RVFW, and Radon transform to estimate SW speed. At least 85% of the waves were tracked successfully for all acquisitions. Diastole and systole were identified using the ECG signal (Fig. 2a). The average SW speed was 0.6 ± 0.1 m/s at end-diastole (Fig. 2b). The measured speeds ranged from 0.5 ± 0.1 m/s during diastole to 1.9 ± 0.3 m/s during systole. The changing SW speeds correspond to the variation in muscle stiffness during the heart cycle as the RV relaxes and contracts. We demonstrate for the first time the induction and tracking of shear waves in the RVFW of a closed-chest pig. The possibility to noninvasively quantify RV wall stiffness opens a large field of translational research, with direct applications in pulmonary hypertension, congenital heart disease and heart failure in general. Pathological increase in stiffness should be further investigated in longitudinal case/control studies. Effects of cardiac resynchronisation therapy on right ventricular function: a model study Roel Meiburg, Jesse Rijks, Ahmed Beela, Justin Luermans, Smulders Martijn, Luuk Heckman, Frits Prinzen, Kevin Vernooy, Joost Lumens Abstract: Cardiac resynchronisation therapy (CRT) has shown to reduce mortality in patients suffering from heart failure. However, functional criteria for the implantation of a CRT device are currently restricted to left ventricular (LV) performance, such as LV ejection fraction, while right ventricular performance is typically left unappreciated. A similar trend is found in the definition of (non-)responders of CRT, which is often based on change of LV end-diastolic volume or degree of LV remodelling after device placement. This is at odds with current clinical knowledge, which indicates that right ventricular function is a strong predictor of outcome in heart failure with or without CRT. Furthermore, placement of the CRT leads - typically the RV apex and LV free wall - might effectively emulate the presence of a right bundle branch block. New lead placement strategies such as septal or left bundle branch (LBB) pacing have been suggested to improve CRT outcome, although these are also optimised with respect to the LV. In this study, we aim to better quantify the effect of the prevailing CRT strategies on the electrical and haemodynamic performance of the right ventricle. To simulate the placement of a CRT device in the heart, cardiac electrical activation maps were generated on a realistic 3D geometry of the LV and RV using a fast formulation of the Eikonal model. The geometry was segmented and average activation times for each segment were calculated, which were then used to inform the CircAdapt model of human cardiovascular mechanics and haemodynamics to calculate the effects of the different pacing strategies on RV pump function. This approach is computationally cheap compared to full 3D (Finite Element) simulations, whilst still providing results on a spatial resolution similar to data currently available in the clinical setting. While LBB pacing resolves the left intra-ventricular dyssynchrony, a significant right intra-ventricular and inter-ventricular dyssynchrony still remain, indicating there is room for improvement in RV function in CRT. Preliminary results indicate that non-selective LBB pacing, where both the LBB pathway and surrounding myocardium are activated, lead to the highest increase in LV and RV function. Hemodynamic performance of 16-20mm extracardiac Goretex conduits in adolescent Fontan patients Friso Rijnberg, Luca van 't Hul, Sasa Kenjeres, Jos Westenberg, Arno Roest, Jolanda Wentzel Abstract: BACKGROUND The synthetic extracardiac conduit used for the completion of the Fontan operation in single ventricle patients lacks growth potential and it is currently unknown if 16-20mm conduits remain adequately sized for adult Fontan patients. This study aims to determine TCPC hemodynamics using computational fluid dynamics (CFD) in adolescent Fontan patients with 16 to 20 mm conduits at rest and during simulated exercise and to assess the relationship between conduit size and hemodynamics METHODS Patient-specific respiratory cycle-resolved computational fluid dynamic (CFD) was performed in 3D reconstructions of the total cavopulmonary connection (TCPC) of 51 extracardiac Fontan patients (median age 16.2 years, Q1-Q3 14.0-18.2) with 16-20mm conduits (Figure 1). Power loss (VDR), pressure gradient and normalized resistance were quantified in rest and during simulated exercise, separated into the different respiratory phases. The cross-sectional area (CSA) of the TCPC vessels were determined and normalized for vessel-specific flow rate (mm2/L/min). Peak VO2 (ml/kg/min) and predicted peak VO2 (%) were assessed from cardiopulmonary exercise testing (CPET). RESULTS All CFD hemodynamics significantly increased from rest to simulated exercise (all p<0.001) and were highest during inspiration compared to expiration (p<0.001). A moderate-strong inverse non-linear relationship was present between normalized conduit CSAmean and CFD hemodynamics in rest and exercise (see figure 2). Pressure gradients of ≥1.0 at rest and ≥3.0 during simulated exercise were observed in patients with a conduit ≤45mm2/L/min. Furthermore, a negative correlation was present between normalized conduit CSAmean and the relative rest-to-exercise increase in hemodynamics, indicating strongest increases in patients with smallest conduits. Normalized TCPC resistance weakly correlated with (predicted) peak VO2. CONCLUSIONS Extracardiac conduits of 16 to 20mm have become relatively undersized in teenage adolescent Fontan patients, with most adverse hemodynamics observed in patients with a conduit ≤45mm2/L/min. Alternative surgical strategies should be explored to optimize future hemodynamics in adult Fontan patients. Time resolved multi-volume ultrafast ultrasound imaging of abdominal aorta mimicking phantoms Larissa Jansen, Hans-Martin Schwab, Richard Lopata Abstract: Abdominal aortic aneurysms (AAAs) have a risk of rupturing. Currently the aneurysm diameter is monitored and used to estimate this risk. However, it is not always adequate and can lead to fatal events and unnecessary treatments. Combining time resolved 3D ultrasound (4D US) imaging with mechanical modeling is a promising tool to support rupture risk assessment as it provides information about the shape and mechanical state of the aneurysm. However, imaging an AAA using conventional ultrasound imaging is challenging as the field-of-view is restricted and the temporal resolution is low (5-10 Hz), restricting geometry and strain analysis. Therefore, we propose to acquire multiple volumes with an ultrafast 4D US acquisition scheme in an experimental setup and temporally and spatially align these to obtain high volume rate volumetric images for geometry and strain assessment. Aorta mimicking phantoms were created by first mixing 10 wt. % polyvinyl alcohol (PVA) in a glycol ethylene demineralized water solution (ratio 2:3) that included 5 wt. % silicon carbide scatterers. Next the material was injected into a mould and underwent 5 freeze-thaw cycles, after which the aorta phantom was removed from the mould. This phantom was then attached to a circulatory mock-loop setup. A block of PVA with scatterers that underwent one freeze-thaw cycle was used to mimic the surrounding tissue. A flow pump was used to pump water into the phantom with a predefined pulsatile flow pattern. An ultrafast sparse aperture transmit-receive sequence was implemented for a 1024¬-element Vermon matrix probe connected to a vantage-256 Verasonics ultrasound research platform. Data was collected with a 60 Hz volume rate at different positions along the vessel length with 5 mm increments. Finally, a large multi-volume dataset was obtained by spatial and temporal registration of the data. With the approach proposed, multiple 4D US imaging datasets were obtained and after registration, these form large volumes from which the total aorta geometry can be obtained. Furthermore, the 60 Hz volume rate allows for local 3D strain measurements at different vessel wall locations. Future work will involve adding a second US probe to the setup. Jointly learned quantization and adaptive beamforming in ultrasound image reconstruction Ben Luijten, Nir Shlezinger, Ariel Amar, Yonina Eldar, Massimo Mischi, Ruud van Sloun Abstract: Digital ultrasound reconstruction relies on analog-to-digital conversion (ADC), mapping the continuous-time channel signals into discrete-time, and quantizing the continuous-amplitude signals in a finite number of quantization levels (or equivalently, bits). High bitrates minimize quantization distortion and maximize SNR, but also increase cost, power consumption, and the required bandwidth between probe and device. Recently, Shlezinger et al. showed that by considering the downstream task, one can significantly reduce the bitrate of ADCs while hardly degrading performance. This can be achieved by jointly learning the quantization levels along with the subsequent digital processing in an end-to-end fashion. In this work, we investigate the merit of task-based ADCs for low-data-rate ultrasound imaging. We propose to jointly learn optimized quantization levels, and adaptive beamforming by deep learning (ABLE), in order to significantly reduce data rates while maintaining high image quality. As such, our processing chain jointly optimizes the following within a predefined bit-budget: 1) Analog (linear) combining of channels, 2) digital quantization rules, and 3) adaptive digital beamforming by ABLE. To that end, the acquisition mapping is approximated as a differential layer, allowing its training along with the overall processing in the form of a deep neural network. For training, 1000 in-vivo frames of 16-bit RF channel data were recorded using the Verasonics L11-4v linear probe. As training targets, we employ Minimum Variance beamformed images using the full bit depth. A separate dataset was acquired for testing purposes. The framework is trained with different rates of analog combining (up to 2x), and bit-budgets (4-bit and 6-bit), yielding up to 8 times reduction in bandwidth. We compare against the original “uncompressed” 16-bit recordings by assessing mean-absolute-error and contrast-to-noise ratio. Results show that we can reduce the bitrate by a factor of 4 without compromising image quality, and that high contrast is retained even up to 8 times compression. Eventually we foresee that this method can facilitate the next generation of low-cost, wireless ultrasound probes.

end %-->