arXiv subjects

Michael Neidlin

Publications and source records attributed to Michael Neidlin.

8 recordsLinked to original sources

arXiv2025

Machine Learning Framework for Thrombosis Risk Prediction in Rotary Blood Pumps

Thrombosis in rotary blood pumps arises from complex flow conditions that remain difficult to translate into reliable and interpretable risk predictions using existing computational models. This limitation reflects an incomplete understanding of how specific flow features contribute to thrombus initiation and growth. This study introduces an interpretable machine learning framework for spatial thrombosis assessment based directly on computational fluid dynamics-derived flow features. A logistic regression (LR) model combined with a structured feature-selection pipeline is used to derive a compact and physically interpretable feature set, including nonlinear feature combinations. The framework is trained using spatial risk patterns from a validated, macro-scale thrombosis model for two representative scenarios. The model reproduces the labeled risk distributions and identifies distinct sets of flow features associated with increased thrombosis risk. When applied to a centrifugal pump, despite training on a single axial pump operating point, the model predicts plausible thrombosis-prone regions. These results show that interpretable machine learning can link local flow features to thrombosis risk while remaining computationally efficient and mechanistically transparent. The low computational cost enables rapid thrombogenicity screening without repeated or costly simulations. The proposed framework complements physics-based thrombosis modeling and provides a methodological basis for integrating interpretable machine learning into CFD-driven thrombosis analysis and device design workflows.

cs.LG↗

arXiv2025

Bayesian Parameter Inference and Uncertainty-Informed Sensitivity Analysis in a 0D Cardiovascular Model for Intraoperative Hypotension

Computational cardiovascular models are promising tools for clinical decision support, particularly in complex conditions, such as intraoperative hypotension (IOH). IOH arises from different mechanisms, making treatment selection non-trivial. Patient-specific predictions require calibration, typically performed using deterministic approaches prone to parameter non-identifiability and lacking uncertainty quantification, hindering clinical translation. Consequently, Bayesian approaches are needed that facilitate parameter inference, sensitivity analysis, and uncertainty quantification in cardiovascular models. We utilize Bayesian Markov chain Monte Carlo (MCMC) to estimate parameter distributions of a cardiovascular lumped parameter model (LPM) across different IOH scenarios. We demonstrate parameter non-uniqueness and its impact on sensitivity indices. We improve parameter reliability by incorporating clinical knowledge and measurement uncertainties. We enable continual learning of the model using sequential parameter updating as new patient data become available. We introduce an uncertainty-aware sensitivity analysis and compare it with a deterministic approach. Deterministic calibration yielded many local solutions for IOH, with notably different sensitivities. MCMC distinguished different hypotension scenarios, such as those induced by impaired contractility or hypovolemia. Propagating uncertainties from MCMC through sensitivity analysis provided tighter credible intervals, resulting in more stable parameter rankings than the deterministic approach. The Bayesian approach revealed differences in model behavior and treatment suggestions across patient conditions. Combining Bayesian inference with sequential updating and sensitivity analysis improves the reliability and identifiability of parameter estimates, enhancing the clinical utility of LPMs for therapy guidance.

physics.med-ph↗

arXiv2025

Characterizing Intraventricular Flow Patterns via Modal Decomposition Techniques in Idealized Left Ventricle Models

Understanding the formation, propagation, and breakdown of the main vortex ring (VR) is essential for characterizing left ventricular (LV) hemodynamics, as its dynamics have been linked to the onset and progression of cardiovascular diseases. In this study, two idealized LV geometries, a semi-ellipsoidal chamber and a more rounded configuration, are analyzed using computational fluid dynamics (CFD) simulations under physiological conditions, with the aim of investigating the fluid mechanisms that govern VR evolution during diastole. Modal decomposition techniques, specifically proper orthogonal decomposition (POD) and higher order dynamic mode decomposition (HODMD), are employed to identify dominant coherent structures and track their temporal behavior. To the authors' knowledge, this is the first time such an analysis is conducted with the explicit goal of unraveling the physics of vortex ring dynamics in idealized ventricular chambers. The comparative approach reveals that geometric morphology plays a central role in modulating the flow: in one case, early interaction between the VR and the ventricular wall, driven by the chamber's shape, triggers strong nonlinear interactions and a more intricate dynamic evolution. In the other, the vortex ring propagates more freely toward the apex before dissipating, resulting in a more organized flow pattern and simpler spectral content. These findings advance the understanding of flow-based indicators relevant to early diagnosis and treatment planning in cardiovascular disease. Moreover, they illustrate how the choice of ventricular geometry can influence not only the simulated hemodynamics, but also the effectiveness of data-driven analysis tools, depending on the clinical context under study.

physics.flu-dyn↗

arXiv2024

Towards Robust Hemolysis Modeling with Uncertainty Quantification: A Universal Approach to Address Experimental Variance

Purpose: The purpose of this study is to address the lack of uncertainty quantification in numerical hemolysis models, which are critical for medical device evaluations. Specifically, we aim to incorporate experimental variability into these models using the Markov Chain Monte Carlo (MCMC) method to enhance predictive accuracy and robustness. Methods: We applied the MCMC method to an experimental hemolysis dataset to derive detailed stochastic distributions for the hemolysis Power Law model parameters $C$, $\alpha$ and $\beta$. These distributions were then propagated through a reduced order model of the FDA benchmark pump to quantify the experimental uncertainty in hemolysis measurements with respect to the predicted pump hemolysis. Results: The MCMC analysis revealed multiple local minima in the sum of squared errors, highlighting the non-uniqueness of traditional Power Law model fitting. The MCMC results showed a constant optimal $C=3.515x10-5$ and log normal distributions of $\alpha$ and $\beta$ with means of 0.614 and 1.795, respectively. The MCMC model closely matched the mean and variance of experimental data. In comparison, conventional deterministic models are not able to describe experimental variation. Conclusion: Incorporating Uncertainty quantification through MCMC enhances the robustness and predictive accuracy of hemolysis models. This method allows for better comparison of simulated hemolysis outcomes with in-vivo experiments and can integrate additional datasets, potentially setting a new standard in hemolysis modeling.

physics.med-ph↗

arXiv2024

Quantifying the Influence of Combined Lung and Kidney Support Using a Cardiovascular Model and Sensitivity Analysis-Informed Parameter Identification

Combined extracorporeal membrane oxygenation (ECMO) and continuous renal replacement therapy (CRRT) pose complex hemodynamic challenges in intensive care. In this study, a comprehensive lumped parameter model (LPM) is developed to simulate the cardiovascular system, incorporating ECMO and CRRT circuit dynamics. The model is used to analyze nine CRRT-ECMO connection schemes under varying flow conditions. Using a robust parameter identification framework based on global sensitivity analysis (GSA) and multi-start gradient-based optimization, we calibrated the model on 30 clinical data points from eight veno-arterial ECMO patients. Our results indicate that CRRT has a significant impact on the cardiovascular system, with changes in pulmonary artery pressure of up to 202.5 %, highly dependent on ECMO flow. The GSA proved to be a powerful tool to improve the parameter estimation process. The established parameter estimation framework is fast and robust without the need for hyperparameter tuning and improves the parameter estimation process with an R^2>0.98 between simulation and experimental data. It uses modeling methods that could pave the way for real-time applications in intensive care. This open-source framework provides a valuable tool for the systematic evaluation of combined ECMO and CRRT, which can be used to develop standardized treatment protocols and improve patient outcomes in critical care. In addition, as a digital twin, this model also provides a good basis for addressing research questions related to mechanical circulatory and respiratory support.

physics.med-ph↗

arXiv2024

A one-dimensional model for aspiration in blood vessels

This work introduces a reduced one-dimensional model for aspiration in blood vessels that accounts for the elasticity of both the vessel wall and the catheter. The inclusion of vessel wall viscoelasticity transforms the governing equation for the flow rate into a parabolic form, enabling accurate resolution of the sharp pressure gradients near the catheter tip that a purely hyperbolic formulation cannot capture. A simplified catheter equilibrium approximation is proposed that reproduces the fully elastic catheter model with high accuracy while reducing computation time significantly. The numerical treatment is based on a relaxation of the hyperbolic subsystem that yields a Lax-Friedrichs-type finite volume scheme and facilitates nodal solvers, enabling efficient coupling between catheterized and uncatheterized vessel segments, including bifurcations and the catheter tip. An implicit-explicit splitting strategy ensures that the viscoelastic terms incur only negligible additional computational cost relative to the purely hyperbolic model. The model is validated against three-dimensional CFD simulations and reference data from the literature, including a suction-force-suction-distance analysis. Numerical experiments investigating the role of catheter elasticity and suction force on the hemodynamics are presented, and an uncertainty quantification study demonstrates the suitability of the framework for efficient parameter studies.

math.NA↗

arXiv2021

Spline-PINN: Approaching PDEs without Data using Fast, Physics-Informed Hermite-Spline CNNs

Partial Differential Equations (PDEs) are notoriously difficult to solve. In general, closed-form solutions are not available and numerical approximation schemes are computationally expensive. In this paper, we propose to approach the solution of PDEs based on a novel technique that combines the advantages of two recently emerging machine learning based approaches. First, physics-informed neural networks (PINNs) learn continuous solutions of PDEs and can be trained with little to no ground truth data. However, PINNs do not generalize well to unseen domains. Second, convolutional neural networks provide fast inference and generalize but either require large amounts of training data or a physics-constrained loss based on finite differences that can lead to inaccuracies and discretization artifacts. We leverage the advantages of both of these approaches by using Hermite spline kernels in order to continuously interpolate a grid-based state representation that can be handled by a CNN. This allows for training without any precomputed training data using a physics-informed loss function only and provides fast, continuous solutions that generalize to unseen domains. We demonstrate the potential of our method at the examples of the incompressible Navier-Stokes equation and the damped wave equation. Our models are able to learn several intriguing phenomena such as Karman vortex streets, the Magnus effect, Doppler effect, interference patterns and wave reflections. Our quantitative assessment and an interactive real-time demo show that we are narrowing the gap in accuracy of unsupervised ML based methods to industrial CFD solvers while being orders of magnitude faster.

cs.LG↗

arXiv2021

Possible Contexts of Use for In Silico trials methodologies: a consensus-based review

The term "In Silico Trial" indicates the use of computer modelling and simulation to evaluate the safety and efficacy of a medical product, whether a drug, a medical device, a diagnostic product or an advanced therapy medicinal product. Predictive models are positioned as new methodologies for the development and the regulatory evaluation of medical products. New methodologies are qualified by regulators such as FDA and EMA through formal processes, where a first step is the definition of the Context of Use (CoU), which is a concise description of how the new methodology is intended to be used in the development and regulatory assessment process. As In Silico Trials are a disruptively innovative class of new methodologies, it is important to have a list of possible CoUs highlighting potential applications for the development of the relative regulatory science. This review paper presents the result of a consensus process that took place in the InSilicoWorld Community of Practice, an online forum for experts in in silico medicine. The experts involved identified 46 descriptions of possible CoUs which were organised into a candidate taxonomy of nine CoU categories. Examples of 31 CoUs were identified in the available literature; the remaining 15 should, for now, be considered speculative.

physics.bio-ph↗