A5011050254- Cardiovascular Function and Risk Factors
- Model Reduction and Neural Networks
- Advanced Numerical Methods in Computational Mathematics
- Advanced Numerical Analysis Techniques
- Numerical methods for differential equations
- Cardiovascular Health and Disease Prevention
- Coronary Interventions and Diagnostics
- Computational Fluid Dynamics and Aerodynamics
- Advanced MRI Techniques and Applications
- Machine Learning in Healthcare
- Artificial Intelligence in Healthcare and Education
- Topic Modeling
- Lattice Boltzmann Simulation Studies
- demographic modeling and climate adaptation
- Heart Rate Variability and Autonomic Control
- ECG Monitoring and Analysis
- Numerical methods in engineering
- Cardiac Structural Anomalies and Repair
- Advanced Control Systems Optimization
- Elasticity and Material Modeling
- Mathematical Biology Tumor Growth
- Structural Health Monitoring Techniques
- Machine Learning in Materials Science
- Polynomial and algebraic computation
- COVID-19 diagnosis using AI
Stanford University
2022-2025
École Polytechnique Fédérale de Lausanne
2018-2021
Three-dimensional (3D) cardiovascular fluid dynamics simulations typically require hours to days of computing time on a high-performance cluster. One-dimensional (1D) and lumped-parameter zero-dimensional (0D) models show great promise for accurately predicting blood bulk flow pressure waveforms with only fraction the cost. They can also accelerate uncertainty quantification, optimization, design parameterization studies. Despite several prior studies generating 1D 0D comparing them 3D...
Bayesian boundary condition (BC) calibration approaches from clinical measurements have successfully quantified inherent uncertainties in cardiovascular fluid dynamics simulations. However, estimating the posterior distribution for all BC parameters three-dimensional (3D) simulations has been unattainable due to infeasible computational demand. We propose an efficient method identify Windkessel parameter posteriors: only evaluate 3D model once initial choice of BCs and use result create a...
0. Abstract Background The integration of large language models (LLMs) in healthcare offers immense opportunity to streamline tasks, but also carries risks such as response accuracy and bias perpetration. To address this, we conducted a red-teaming exercise assess LLMs developed dataset clinically relevant scenarios for future teams use. Methods We convened 80 multi-disciplinary experts evaluate the performance popular across multiple medical scenarios. Teams composed clinicians, engineering...
Red teaming, the practice of adversarially exposing unexpected or undesired model behaviors, is critical towards improving equity and accuracy large language models, but non-model creator-affiliated red teaming scant in healthcare. We convened teams clinicians, medical engineering students, technical professionals (80 participants total) to stress-test models with real-world clinical cases categorize inappropriate responses along axes safety, privacy, hallucinations/accuracy, bias. Six...
We are interested in a reduced order method for the efficient simulation of blood flow arteries. The dynamics is modeled by means incompressible Navier–Stokes equations. Our algorithm based on an approximated domain-decomposition target geometry into number subdomains obtained from parametrized deformation geometrical building blocks (e.g., straight tubes and model bifurcations). On each these blocks, we build set spectral functions Proper Orthogonal Decomposition large snapshots finite...
This work focuses on the development of a non-conforming method for coupling PDEs based weakly imposed transmission conditions: continuity global solution is enforced by finite number Lagrange multipliers defined over interfaces adjacent subdomains. The falls into class primal hybrid methods, which include also well-known mortar method. Differently from method, we discretize space basis functions interface spectral approximation independently discretization two domains. In particular, our...
The evaluation of cardiac contractility by the assessment ventricular systolic elastance function is clinically challenging and cannot be easily obtained at bedside. In this work, we present a framework characterizing left from readily available data, including systemic pulmonary arterial pressure signals. We implemented calibrated deep neural network (DNN) consisting multi-layer perceptron with 4 fully connected hidden layers 16 neurons per layer, which was trained data lumped model...
Reduced-order models (ROMs) allow for the simulation of blood flow in patient-specific vasculatures without high computational cost and wait time associated with traditional fluid dynamics (CFD) models. Unfortunately, due to simplifications made their formulations, ROMs can suffer from significantly reduced accuracy. One common simplifying assumption is continuity static or total pressure over vascular junctions. In many cases, this has been shown introduce significant error. We propose a...
Boundary condition (BC) calibration to assimilate clinical measurements is an essential step in any subject-specific simulation of cardiovascular fluid dynamics. Bayesian approaches have successfully quantified the uncertainties inherent identified parameters. Yet, routinely estimating posterior distribution for all BC parameters 3D simulations has been unattainable due infeasible computational demand. We propose efficient method identify Windkessel parameter posteriors using results from a...
Whole-body hemodynamics simulators, which model blood flow and pressure waveforms as functions of physiological parameters, are now essential tools for studying cardiovascular systems. However, solving the corresponding inverse problem mapping observations (e.g., arterial at specific locations in network) back to plausible parameters remains challenging. Leveraging recent advances simulation-based inference, we cast this statistical inference by training an amortized neural posterior...
Characterizing left ventricle (LV) systolic function in the presence of an LV assist device (LVAD) is extremely challenging. We developed a framework comprising deep neural network (DNN) and 0D model cardiovascular system to predict parameters function. DNN input data were systemic pulmonary arterial pressure signals, rotation speeds device. Output function, including end-systolic maximal elastance (E max,lv ), variable essential for adequate hemodynamic assessment LV. A system, wide range...
Reduced-order models based on physics are a popular choice in cardiovascular modeling due to their efficiency, but they may experience reduced accuracy when working with anatomies that contain numerous junctions or pathological conditions. We develop one-dimensional reduced-order simulate blood flow dynamics using graph neural network trained three-dimensional hemodynamic simulation data. Given the initial condition of system, iteratively predicts pressure and rate at vessel centerline...
This work focuses on the development of a non-conforming domain decomposition method for approximation PDEs based weakly imposed transmission conditions: continuity global solution is enforced by discrete number Lagrange multipliers defined over interfaces adjacent subdomains. The falls into class primal hybrid methods, which also include well-known mortar method. Differently from method, we discretize space basis functions interface spectral independently discretization two domains; one...