Abstract:Spatially resolved simulation models of microcirculatory blood flow facilitate a detailed understanding of microcirculatory phenomena at the micrometer scale by capturing heterogeneity in blood flow. These models combine physical laws, empirical descriptions of the blood's complex rheological behavior, and in-vivo/ex-vivo imaging of the microvasculature. However, imaged areas often only partially represent self-contained tissue regions, leading to numerous vessels crossing boundaries and strongly influencing simulated blood flows through imposed boundary conditions. Selecting appropriate boundary conditions is challenging due to the heterogeneity of pressures and blood flows, resulting in significant uncertainties. This study addresses two key methodological aspects of spatially resolved blood flow simulations: selecting appropriate boundary conditions and quantifying the impact of boundary condition uncertainties on simulated hemodynamic variables. An adaptive method for assigning appropriate pressure boundary conditions is proposed and rigorously evaluated in extensive brain cortical networks against reference data from an established blood flow simulation model. A probabilistic approach is adopted to assess the impact of boundary condition uncertainties on blood flow simulations. The adaptive method is further integrated into a Bayesian calibration framework, inferring distributions over thousands of unknown pressure boundary conditions and providing uncertainty estimates for blood flow simulations. The adaptive method, which is straightforward to implement and scales well with extensive microvascular networks, produces hemodynamic simulations consistent with reference data, yielding depth-dependent pressure profiles and layer-wise capillary blood flow profiles consistent with previous studies. These phenomena are demonstrated to generalize also to biphasic blood flow simulation models incorporating in-vivo viscosity formulations. The uncertainty analysis further reveals a novel spatially heterogeneous and depth-dependent pattern in blood flow uncertainty. It is anticipated that the adaptive method for pressure boundary conditions will be useful in future applications of both forward and inverse blood flow modeling, and that uncertainty quantification will be valuable in complementing hemodynamic predictions with associated uncertainties.

Velocity Estimations in Blood Microflows via Machine Learning Symmetries

Machine learning in cardiovascular flows modeling: Predicting arterial blood pressure from non-invasive 4D flow MRI data using physics-informed neural networks

Application of machine learning in predicting blood flow and red cell distribution in capillary vessel networks

Predicting capillary vessel network hemodynamics in silico by machine learning

Physics-informed graph neural networks for flow field estimation in carotid arteries

Physics-informed neural networks for parameter estimation in blood flow models

A Meshless Solver for Blood Flow Simulations in Elastic Vessels Using Physics-Informed Neural Network

Stochastic Parameter Prediction in Cardiovascular Problems

Machine-learning fluid flow

AI based analysis of red blood cells in oscillating microchannels

Classification of Red Blood Cell Rigidity from Sequence Data of Blood Flow Simulations Using Neural Networks

Physics-Informed Graph Neural Networks to solve 1-D equations of blood flow

Physics driven reduced order model for real time blood flow simulations

Machine-Learning Identification of Hemodynamics in Coronary Arteries in the Presence of Stenosis

Physics-constrained coupled neural differential equations for one dimensional blood flow modeling

Machine learning for cerebral blood vessels' malformations

Machine Learning Identification Framework of Hemodynamics of Blood Flow in Patient-Specific Coronary Arteries with Abnormality

Image-Based Experimental Measurement Techniques to Characterize Velocity Fields in Blood Microflows

Blood flow simulation and uncertainty quantification in extensive microvascular networks: Application to brain cortical networks

226 Advancing Neurovascular Diagnostics for Abnormal Hemodynamic Conditions Through AI-Driven Physics-informed Neural Networks

Contributions towards Data driven Deep Learning methods to predict Steady State Fluid Flow in mechanical Heart Valves