Background:Aortic coarctation (CoA) is a major cause of arterial hypertension in young individuals, with recurrence occurring in up to one-third of patients throughout life despite guideline-directed management. Methods:We conducted a development and validation study at a single centre in Berlin, Germany, utilising routinely collected electronic health records, cardiovascular magnetic resonance (CMR), and mid-/long-term follow-up data from 218 visits (160 individuals with CoA receiving standard of care, guideline-based management) collected between January 2014 and April 2022. Machine learning (ML) models (CatBoost, XGBoost, random forest, support vector classifiers, neural networks, logistic regression, and K-nearest neighbours) were developed to predict three endpoints: re-coarctation requiring intervention (CoA-I), aortic surgery (CoA-S) as a subset of CoA-I, and persistent arterial hypertension. The dataset was divided by random stratified split into a training set (n = 159; for model development with five-fold cross-validation), and a hold-out test set (n = 59; for out-of-sample validation). Stratification was based on sex, age, and CoA-I status. We included a final set of 38 clinically relevant features, encompassing baseline characteristics, medication intake, echocardiography, CMR, electrocardiography (ECG), and treatment decisions. ClinicalTrials.gov Identifier: NCT02591940. Findings:Tree-based and support vector classifier models performed best after Bayesian hyperparameter optimisation, yielding high performance in a stratified validation cohort: area under the receiver operating characteristic curve (ROC AUC) 0.90 ± 0.01 for CoA-I, 0.90 ± 0.01 for CoA-S, and 0.84 ± 0.01 for hypertension. Shapley Additive exPlanations (SHAP) highlighted peak Doppler gradient, time since index visit, and ventricular size indices as key predictors for CoA-I. In inverse-probability-weighted analyses, antihypertensive medication was associated with a lower CoA-I probability (-17.3%; 95% confidence interval [CI], -28.2 to -6.4; p = 0.002), with concordant propensity-score-matched findings. An open-access research interface (https://icm.dhzc.charite.de/calc_coa) incorporates treatment thresholds and personalised risk estimates from an updatable ML framework. Interpretation:These findings suggest that patient-specific multimodal ML-based risk estimates may complement guideline-based care by identifying patients at increased risk of CoA-I or persistent hypertension, with the potential to support more tailored follow-up and reduce lifetime exposure to brachiocephalic hypertension. In adjusted cohort-level analyses, antihypertensive medication was associated with a lower probability of CoA-I. Funding:This study was supported by the European Commission's Seventh Framework Programme (FP7, project ID 611232). M.K. acknowledges support within the Charité Digital Clinician Scientist Programme funded by DFG. M.K. and T.K. have received funding within the CHAIN project (Project ID: 101314833), supported by the European Union's EU4Health Programme. T.K. and M.K. acknowledge support within the Collaborative Research Centre SFB 1470, funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation), project ID 437531118. M.K. has received funding from the Bundesministerium für Forschung, Technologie und Raumfahrt (BMFTR, Federal Ministry of Research, Technology and Space), VADYS-ME, grant number 01EJ2406A.

In vivo verification for the development of cardiac MRI and CFD simulations is limited by scan time and motion artifacts. We developed a subject-specific, MRI-compatible left ventricle (LV) phantom within a closed-loop circulation system driven by an MRI-safe pump to reproduce cardiac motion and provide robust data for high-fidelity model validation. The LV was fabricated from a PVA-based hydrogel to ensure MRI contrast, mechanical stability, and reusability. The complete setup fulfills key requirements for long-term leakage-free operation, close coil placement, and resembles in vivo soft-tissue contrasts. The setup reproduces healthy end-diastolic and end-systolic geometries, including physiologic contraction and papillary muscles. Integrated aortic and mitral valves approximate physiological opening and closing. High image contrast enabled time-resolved segmentation of the LV geometry. The LV shape matched the target anatomy well, despite minor deviations at the basal transition and in valve orifice, with mild aortic stenosis and mitral regurgitation. 4D flow MRI confirmed physiological flow patterns, including diastolic vortex ring formation and realistic systolic outflow. Velocity estimates confirmed the phantom’s applicability in an in vitro setting without scan time limitations. The acquired time-resolved MRI data enabled CFD simulations incorporating LV and valve motion, although segmentation accuracy remains a primary source of uncertainty. This LV phantom provides a stable, MRI-compatible platform for generating reproducible data to validate modeling of intracardiac hemodynamics. Its realistic anatomy, motion, and flow patterns support both simulation validation and MRI sequence development. Future work will focus on improving valve kinematics and LV motion fidelity.

Computational hemo dynamics can enhance image-based diagnosis and provide complementary insights to predict, understand, and monitor treatments. The high computational costs and the complexity associated with handling patient-specific settings remain a major challenge toward clinical applications. In this work, we propose a novel robust shape registration method for nonparametric aortic geometries, describing different applications for projection-based reduced-order modeling for the training of graph neural networks and for data assimilation. The registration approach is based on ResNet-LDDMM, trained with a dataset of synthetic shapes, generated from real ones with statistical shape modeling. The optimization is tailored to surface meshes and does not rely on a priori assumptions on domain parameterization. We employ a multigrid strategy during the training phase that allows handling realistic mesh sizes. The registration enables the definition of geometric encoding of different blood flow solutions on a single reference shape, as well as the design of projection-based reduced-order models. We use this geometrical encoding to improve the training of graph neural networks and to present potential applications in data assimilation problems, combined with a generalized parameterized-Background Data-Weak formulation. As a particular example of data assimilation problem, we address the reconstruction of velocity fields and wall shear stresses, as well as the estimation of pressure fields and pressure-related biomarkers, such as the pressure drop, from low-resolution velocity observations. We show various numerical tests based on synthetic data, comparing the proposed strategies with state-of-the-art estimators.

In the field of cardiovascular device development, new devices such as heart valves, stents or pressure probes for long term heart failure monitoring are subject to animal trials to evaluate their safety and efficacy. For such applications, swine and sheep are the animal models of choice owed to their similarities to humans with regards to heart size, weight and ventricular kinetics. However, clinical aspects regarding the choice of animal model revolve mainly around anatomical similarities as well as the ability to induce the desired pathology. In the case of pulmonary artery pressure sensors, both swine and sheep appear to be suitable candidates for animal trials since both animals have been used for pre-clinical evaluation. Hemodynamic aspects however, although equally important for device performance, appear rather underrepresented in current research and it remains uncertain whether anatomical similarities between humans and animal model in the region of interest translate to hemodynamic similarities. To provide insight whether pulmonary artery hemodynamics in large animal models are indeed comparable to those in humans, this work presents a computational fluid dynamics-based study on pulmonary artery hemodynamics for humans, swine and sheep. A total of 28 human, 41 porcine and 14 ovine transient simulations of pulmonary artery hemodynamics were performed based on subject-specific geometries reconstructed from computed tomography data. The distributions of wall shear stress (WSS) and oscillatory shear index (OSI) within the cohorts were then compared to assess hemodynamic similarity. Distributions of time averaged WSS were found to be similar between humans and sheep (median 1.2 vs. 1.5 Pa, interquartile range (IQR) 0.8 Pa vs. 0.6 Pa, Wilcoxon rank sum test p = 0.42) but were significantly different for swine (median 1.7, IQR 0.5, p < 0.05), whereas OSI was significantly different for sheep and swine (0.17 ± 0.04 vs. 0.14 ± 0.03 and 0.09 ± 0.02). between sheep and humans. In summary, pulmonary artery vessel wall stresses of both animal models appear broadly similar to humans, however, sheep seem to have a notable edge over swine in our study.

Implantable pulmonary artery pressure sensors (PAPS) might impose a flow-induced risk of thrombus formation in the pulmonary artery (PA). To assess this risk, an in silico study-enhanced animal study with 20 sensors implanted in 10 pigs had previously been conducted. In the in silico study, PAPS were virtually implanted mimicking real implantations, based upon data acquired by CT. This animal in silico study investigated changes in hemodynamics caused by PAPS using image-based computational fluid dynamics (CFD). However, porcine and human PA differ significantly in geometry and hemodynamics. To investigate the transferability of animal in silico study findings toward human conditions, we propose a parallel in silico human study. Based on a similarity analysis (L1 norm for 8 geometric features) human PA geometries with the least difference to 10 porcine PA were selected. PAPS were virtually implanted in human PA as close as possible, mimicking the implantation configuration of the animal study. Finally, a numerical flow analysis of the hemodynamic changes due to PAPS implantation was done. Comparing human and porcine PA, we found significantly larger left and right PA diameters in humans, whereas no differences were found for main PA diameters and bifurcation angle. Comparing hemodynamic boundary conditions, we found a significantly smaller heart rate and a significantly higher peak systolic main PA flow rate in humans, whereas no significant differences for cardiac output were found. The human in silico PAPS study found no relevant changes in hemodynamics increasing the risk of thrombus formation after sensor implantation. This is also valid for PAPS that were non-optimally implanted. Thus, despite differences between species, findings of the in silico animal study were confirmed by the human in silico study.

Background: Recent studies suggest that any degree of patient-prosthesis mismatch (PPM) increases morbidity and mortality after surgical aortic valve replacement (SAVR). We used computational fluid dynamics simulations to test the influence of prosthesis size and physical activity after SAVR. Methods: In 10 patients with aortic valve stenosis, virtual SAVR was performed. Left ventricular outflow tract stroke volume and flow direction information (4D Flow) were used, and an increase in stroke volume of 25% was chosen for simulating physical activity. Pressure gradients (DP max) across the aortic valve and blood flow profiles in the ascending aorta were calculated and predicted for three different valve sizes at rest and under stress in every patient. Results: Gradients across the aortic valve were significantly lower using larger valves; however, they were not normalized after SAVR (DP max [mmHg] norm/smaller/reference/larger valve = 6/14/12/9 mmHg, <0.01 compared to norm). Physical activity simulation increased DP max in all patients and across all valve sizes (DP max [mmHg] rest versus stress for the smaller/reference/larger valve = 14 vs. 23, 12 vs. 18, 9 vs. 14). Blood flow profiles did not normalize after SAVR and remained unaffected by physical activity. Gradients differed between mild and moderate stenosis between different therapy options and even showed moderate to severe stenosis under simulated physical activity. Conclusions: Prosthesis size and physical activity simulation have a significant influence on gradients across the aortic valve. Virtual therapy planning using patient-specific data might help to improve outcomes after SAVR in the future.

Surgical ventricular restoration (SVR) excludes scarred myocardium after myocardial infarction to restore shape and contractility of dilated, aneurysmal left ventricles (LVs). Detailed changes in intracardiac hemodynamics following the surgery are not fully investigated. In this study, digital replicas of the patient's LV were used to study the hemodynamic impact of successful SVR. The digital replicas were built based on pre-operative and post-operative cardiac computed tomography data of nine patients (3females, 60 ± 13years) who underwent successful SVR (significant reduction in heart failure symptoms). The computational framework was used to calculate LV morphology, dynamics, and intracardiac hemodynamics using image-based computational fluid dynamics (CFD). SVR successfully reduced the LV volumes. Morphological analysis showed restoration of myocardial wall thickness in aneurysmal regions (5.5 ± 2.0 vs. 8.6 ± 3.0 mm) and an increased end-diastolic sphericity (sphericity index 0.39 ± 0.07 vs. 0.46 ± 0.07). No distinct flow alterations could be linked thereto. CFD revealed a higher post-operative kinetic energy level (diastolic maximum 10.0 ± 7.6 vs. 16.8 ± 9.1mJ) and an improved global washout (29.5 ± 9.7 vs. 10.3 ± 6.4% after five cycles), which correlated to increases in volume-curve-derived diastolic energy gain and ejection fraction, respectively. Flow efficiency improved by means of an increased end-diastolic surface-averaged vortex strength (16.2 ± 5.1 vs. 30.0 ± 15.01/s) and a decreased normed diastolic energy loss (18.9 ± 3.9 vs. 15.0 ± 3.7%). The hemodynamic filling forces in diastole were aligned with the LV long axis before and after surgery and correlated with LV contractility. In summary, the digital patient replicas facilitated a detailed analysis and showed favorable flow changes with successful SVR.

Image-based, patient-specific modelling of hemodynamics can improve diagnostic capabilities and provide complementary insights to better understand the hemodynamic treatment outcomes. However, computational fluid dynamics simulations remain relatively costly in a clinical context. Moreover, projection-based reduced-order models and purely data-driven surrogate models struggle due to the high variability of anatomical shapes in a population. A possible solution is shape registration: a reference template geometry is designed from a cohort of available geometries, which can then be diffeomorphically mapped onto it. This provides a natural encoding that can be exploited by machine learning architectures and, at the same time, a reference computational domain in which efficient dimension-reduction strategies can be performed. We compare state-of-the-art graph neural network models with recent data assimilation strategies for the prediction of physical quantities and clinically relevant biomarkers in the context of aortic coarctation.

Transcatheter edge-to-edge repair (TEER) is an effective treatment for mitral valve regurgitation in patients with high surgical risk, but predicting the hemodynamic outcomes is challenging. Reduced-order models (ROMs) show promise for post-TEER hemodynamic outcome predictions, but personalization of the ROM equations is essential for accurate simulations of mitral valve blood flow rates and pressure gradients during both diastole and systolic regurgitation. While the mitral valve orifice area is a common parameter used for ROM personalization, other aspects of the mitral valve shape are usually not considered. In this in-silico study, we investigated the influences of mitral valve shape on transmitral hemodynamics using a combination of computational fluid dynamics (CFD) simulations and geometrical analyses. Mitral valves from ten TEER patients were analyzed at three valve states: early diastole (pre- and post-TEER) and systolic regurgitation (pre-TEER). The orifice-to-annulus area ratio and the orifice orientation were identified as key shape parameters impacting mitral valve hemodynamics. Based on these findings, we developed shape-based ROM equations that are personalized using routine echocardiographic data. The ROM estimates agreed well with CFD simulation results (mean relative differences <1 % and limits of agreements <13 % for both flow rates and pressure gradients). Application of the ROM equations to patient-specific data revealed distinct hemodynamic differences between the three valve states, aligning with expectations from both physiological and fluid dynamics perspectives. Our results suggest that incorporating mitral valve shape parameters into ROMs could improve the accuracy of patient-specific simulations, thus enhancing their potential for supporting TEER planning and predicting intervention outcomes.

Oxygenators are a lifesaving technology used for blood oxygenation and decarboxylation in case of acute respiratory failure, chronic lung disease, and during open-heart surgery. Devices typically consist of a bundle of thousands of fiber membranes in a housing, with gas flowing inside the fibers and blood flowing in the opposite direction outside the fibers. Both ends of the fiber membranes are attached with an adhesive to prevent direct contact between gas and blood. The shape of the volume through which the blood flows is determined by the housing of the oxygenator and the internal end surfaces of the bonded parts of the fiber-membrane bundle. The traditional potting process results in a volume shape that is associated with stagnation zones, which are known to promote thrombus formation. In this study, an adapted potting process is proposed which results in a blood compartment with beveled end faces of the glued bundle parts. Using a numerical study, we have demonstrated that the novel oxygenator design results in optimized flow conditions. ### Competing Interest Statement The authors have declared no competing interest. ### Funding Statement This study did not receive any funding ### Author Declarations I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained. Yes I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals. Yes I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance). Yes I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable. Yes All data produced in the present study are available upon reasonable request to the authors

Objective: Major challenges for clinical applications of in silico medicine are limitations in time and computational resources. Computational approaches should therefore be tailored to specific applications with relatively low complexity and must be verified and validated against clinical gold standards. Methods: This study performed computational fluid dynamics simulations of left ventricular hemodynamics of different complexity based on shape reconstruction from steady state gradient echo magnetic resonance imaging (MRI) data. Computed flow results of a rigid wall model (RWM) and a prescribed motion fluid-structure interaction (PM-FSI) model were compared against phase-contrast MRI measurements for three healthy subjects. Results: Extracted boundary conditions from the steady state MRI sequences as well as computed metrics, such as flow rate, valve velocities, and kinetic energy show good agreement with in vivo flow measurements. Regional flow analysis reveals larger differences. Conclusion: Basic flow structures are well captured with RWM and PM-FSI. For the computation of further biomarkers like washout or flow efficiency, usage of PM-FSI is required. Regarding boundary-near flow, more accurate anatomical models are inevitable. Significance: These results delineate areas of application of both methods and lay a foundation for larger validation studies and sensitivity analysis for healthy and diseased cases, being an essential step upon clinical translations.

Properties of the pulmonary artery play an essential role in the diagnosis and treatment planning of diseases such as pulmonary hypertension. Patient-specific simulation of hemodynamics can support the planning of interventions. However, the variable complex branching structure of the pulmonary artery poses a challenge for image-based generation of suitable geometries. State-of-the-art segmentation-based approaches require an interactive 3D surface reconstruction to prepare the simulation geometry. We propose a deep learning approach to generate a 3D surface mesh of the pulmonary artery from CT images suitable for simulation. The proposed method is based on the Voxel2Mesh algorithm and includes a voxel encoder and decoder as well as a mesh decoder to deform a prototype mesh. An additional centerline coverage loss facilitates the reconstruction of the branching structure. Furthermore, vertex classification allows for the definition of in- and outlets. Our model was trained with 48 human cases and tested on 10 human cases annotated by two observers. The differences in the anatomical parameters inferred from the automatic surface generation correspond to the differences between the observers’ annotations. The suitability of the generated mesh geometries for numerical flow simulations is demonstrated.

The utilization of numerical methods, such as computational fluid dynamics (CFD), has been widely established for modeling patient-specific hemodynamics based on medical imaging data. Hemodynamics assessment plays a crucial role in treatment decisions for the coarctation of the aorta (CoA), a congenital heart disease, with the pressure drop (PD) being a crucial biomarker for CoA treatment decisions. However, implementing CFD methods in the clinical environment remains challenging due to their computational cost and the requirement for expert knowledge. This study proposes a deep learning approach to mitigate the computational need and produce fast results. Building upon a previous proof-of-concept study, we compared the effects of two different artificial neural network (ANN) architectures trained on data with different dimensionalities, both capable of predicting hemodynamic parameters in CoA patients: a one-dimensional bidirectional recurrent neural network (1D BRNN) and a three-dimensional convolutional neural network (3D CNN). The performance was evaluated by median point-wise root mean square error (RMSE) for pressures along the centerline in 18 test cases, which were not included in a training cohort. We found that the 3D CNN (median RMSE of 3.23 mmHg) outperforms the 1D BRNN (median RMSE of 4.25 mmHg). In contrast, the 1D BRNN is more precise in PD prediction, with a lower standard deviation of the error (±7.03 mmHg) compared to the 3D CNN (±8.91 mmHg). The differences between both ANNs are not statistically significant, suggesting that compressing the 3D aorta hemodynamics into a 1D centerline representation does not result in the loss of valuable information when training ANN models. Additionally, we evaluated the utility of the synthetic geometries of the aortas with CoA generated by using a statistical shape model (SSM), as well as the impact of aortic arch geometry (gothic arch shape) on the model’s training. The results show that incorporating a synthetic cohort obtained through the SSM of the clinical cohort does not significantly increase the model’s accuracy, indicating that the synthetic cohort generation might be oversimplified. Furthermore, our study reveals that selecting training cases based on aortic arch shape (gothic versus non-gothic) does not improve ANN performance for test cases sharing the same shape.

To facilitate pre-clinical animal and in-silico clinical trials for implantable pulmonary artery pressure sensors, understanding the respective species pulmonary arteries (PA) anatomy is important. Thus, morphological parameters describing PA of pigs and sheep, which are common animal models, were compared with humans. Retrospective computed tomography data of 41 domestic pigs (82.6 ± 18.8 kg), 14 sheep (49.1 ± 6.9 kg), and 49 patients (76.8 ± 18.2 kg) were used for reconstruction of the subject-specific PA anatomy. 3D surface geometries including main, left, and right PA as well as LPA and RPA side branches were manually reconstructed. Then, specific geometric parameters (length, diameters, taper, bifurcation angle, curvature, and cross-section enlargement) affecting device implantation and post-interventional device effect and efficacy were automatically calculated. For both animal models, significant differences to the human anatomy for most geometric parameters were found, even though the respective parameters’ distributions also featured relevant overlap. Out of the two animal models, sheep seem to be better suitable for a preclinical study when considering only PA morphology. Reconstructed geometries are provided as open data for future studies. These findings support planning of preclinical studies and will help to evaluate the results of animal trials.

BACKGROUND:The study of blood trauma, such as hemolysis in blood-carrying devices, is crucial due to the high incidence of adverse events like alteration of blood function, bleeding, and multi-organ failure. The extent of flow-induced hemolysis, predominantly influenced by stress duration and intensity, is described by established model parameters based on the power law approach. In recent years, various parameters were determined using different Couette shearing devices and donor species. However, they have not been validated due to limited experimental data. METHODS:This study provides hemolysis measurements in a Couette shearing device and evaluates the suitability of different power law parameters. The revised Couette shearing device generates well-defined dynamic stress loads that are repeatedly applied to blood samples at a defined temperature. Human blood samples with an adjusted hematocrit of 30%, were tested with varying repetitions (20 to 80 times). The half-sinusoidal stress loads had amplitudes of 73 to 140 Pa and exposure times of 24 msec per repetition. The parameters of five common power law hemolysis approaches were then compared with the experimental data. RESULTS:The prediction with the power law model parameters C = 3.458 × 10-6, α = 0.2777 and β = 2.0639 showed a good agreement with the experimental results. CONCLUSION:The effect of multiple short-time stresses on hemolysis was investigated to validate the power law hemolysis model with the Couette shearing device of this study.

Numerical simulations of pulsatile blood flow in an aortic coarctation require the use of turbulence modeling. This paper considers three models from the class of large eddy simulation (LES) models (Smagorinsky, Vreman, σ -model) and one model from the class of variational multiscale models (residual-based) within a finite element framework. The influence of these models on the estimation of clinically relevant biomarkers used to assess the degree of severity of the pathological condition (pressure difference, secondary flow degree, normalized flow displacement, wall shear stress) is investigated in detail. The simulations show that most methods are consistent in terms of severity indicators such as pressure difference and stenotic velocity. Moreover, using second-order velocity finite elements, different turbulence models might lead to considerably different results concerning other clinically relevant quantities such as wall shear stresses. These differences may be attributed to differences in numerical dissipation introduced by the turbulence models.