Session X19: Minisymposium IV: Fluid Dynamics in Clinical Imaging

Aortic hemodynamics in health and disease: validation of fluid structure interaction simulations against 4D flow magnetic resonance imaging

Understanding the complex interplay between morphologic and hemodynamic features in the human aorta is critical for risk stratification and individualized treatment planning.  In this talk, we describe methods for patient specific fluid structure interaction (FSI) simulations in the healthy and diseased aorta. We validate simulations against in vitro 4D flow MRI in both cases.  First, we compare performance of the reduced unified continuum method for FSI against 4D-flow MRI using a compliant phantom of the healthy aorta with matched material properties. We extract high-resolution anatomical and hemodynamic information from an in vitro mock circulatory system. To accurately reflect experimental conditions, we implemented in-plane vascular motion, viscoelastic external tissue support and vascular tissue prestressing. Validation is demonstrated through close quantitative agreement in pressures, lumen area changes, pulse wave velocity, and early systolic velocities, as well as qualitative agreement in late systolic flow structures.  Second, we examine diseased aortic hemodynamics in patient specific models of type-B aortic dissection. We evaluate the effects of entry and exit tear size by comparing Abritrary Lagrangian Eulerian FSI simulations with in vitro 4D-flow MRI. A baseline patient-specific 3D-printed model and two variants with modified tear size (smaller entry tear, smaller exit tear) were embedded into a flow- and pressure-controlled setup to perform MRI. Results showed well-matched complex flow patterns between 4D-flow MRI and FSI simulations. Compared to the baseline model, false lumen flow volume decreased with either a smaller entry tear or smaller exit tear. True to false lumen pressure difference increased with a smaller entry tear and became negative with a smaller exit tear. This work establishes quantitative and qualitative effects of entry or exit tear size on hemodynamics in aortic dissection, with particularly notable impact observed on false lumen pressurization. In both studies, FSI simulations demonstrate acceptable qualitative and quantitative agreement with flow imaging, supporting its deployment in clinical studies.   

Techniques for estimation of model parameters in computational hemodynamics

A major challenge in constructing three dimensional patient specific hemodynamic models is the calibration of model parameters to match patient data on flow, pressure, wall motion, etc. acquired in the clinic. This work summarizes several formulations for model parameter and flow estimations in cardiovascular applications, using MRI, CT, and angiography.

In the context of MRI and CT data, we have developed a CFD parameter estimation framework that relies on the following fundamental contributions [1]. (i) A Reduced-Order Unscented Kalman Filter (ROUKF) model for data assimilation for wall material and simple lumped parameter network (LPN) boundary condition model parameters. (ii) A constrained least squares augmentation (ROUKF-CLS) for more complex LPNs. (iii) A “Netlist” implementation, supporting easy filtering of parameters in such complex LPNs. The ROUKF algorithm is demonstrated using non-invasive patient-specific data on anatomy, flow and pressure from a healthy volunteer. The ROUKF-CLS algorithm is demonstrated using synthetic data on a coronary LPN. These methods have been implemented as part of the CRIMSON hemodynamics software package [2].

In the context of X-ray angiography data, we have recently developed a fully automatic method to segment arteries, through a convolutional neural network, AngioNet [3]. The main innovation in this network is the introduction of an Angiographic Processing Network (APN) which significantly improves segmentation performance on multiple network backbones, with the best performance using Deeplabv3+. This APN enabled us to learn the best possible pre-processing filters to improve segmentation, including when using dynamic series. This is a key step towards automated assessment of flow using X-ray angiography.   

Flow Dynamics in Clinical Imaging – A Practical Review of Manifestations, Exploitations, and Specific Interrogations

Fluid dynamics plays a central role in several physiologic and pathophysiologic processes throughout the body, across multiple spatial and temporal scales. From the earliest applications of catheter angiography, medical imaging has at least qualitatively interrogated hemodynamics to improve diagnosis, and recent decades have seen dramatic advances in our ability to depict and quantitatively understand flow in-vivo. These advances have afforded a better understanding of imaging findings, improved diagnostics, a deeper comprehension of normal and disordered hemodynamics and their sequela, and importantly, have allowed the refinement of existing and creation of new therapeutics.

Cardiovascular flow and its effects are evident in most forms of clinical imaging. This talk will present key clinical examples where flow dynamics 1) manifest in anatomic imaging (not targeted to flow measurement), revealing features of locoregional hemodynamics, 2) cause imaging features that are exploited for either image formation or imaging interpretation, and 3) are directly interrogated for diagnosis of both flow-mediated and end-organ pathology in different vascular territories. The bulk of examples will concern computed tomography, magnetic resonance, and ultrasound imaging techniques that are in routine clinical use, introducing flow phenomena of prime clinical relevance and setting the stage for more detailed discussions on specific modern techniques for direct flow imaging.   

Cerebral Ischemic Injury during Extracorporeal Membrane Oxygenation Detected by Ultrasound Localization Microscopy of the Microvascular Flow

Affecting 8–50% of patients, ischemic brain injury is among the most frequent complications in patients undergoing pediatric Extracorporeal Membrane Oxygenation (ECMO). Non-invasive monitoring of brain perfusion during ECMO using contrast-enhanced ultrasound imaging could potentially improve this outcome. Here we utilize an in-house developed Ultrasound Localization Microscopy (ULM) procedure, which involves super-resolution deep learning with a subpixel localization accuracy of bubble centers, followed by Kalman filter-based bubble tracking, to map the cerebral circulation in a pediatric pig model during ECMO. Parameters, including the mean velocity magnitude in large (>1mm) and medium (0.2-1mm) blood vessels, as well as the microcirculation in micro (<0.2mm) vessels, are compared to histological data of the levels of brain ischemia. Results show significantly decreased micro perfusion in the cortex and thalamus with increasing ischemic injury level. The microvascular flows are more sensitive than those in the larger vessels, not only in detecting injury but also in evaluating the injury level. These findings suggest that the management of ECMO patients could be guided by and benefit from the non-invasive monitoring of cerebral microvascular flows.   

Topological Flow Data Analysis for Blood Flows Inside a Heart

Complex vortex patterns of blood flow in the heart play an important role in an efficient blood flow supply from the heart to the organs. Recent progress in medical imaging and computer technology such as echocardiography and cardiac MRI yields blood flow visualization tools recently. On the other hand, however, there are still few mathematical theories to clearly define the vortex flow structures such as size and location, or change over time in the main chamber in the heart. Although the function of the vortex blood flow inside the left ventricle is highly unstable and complex, we propose a new mathematical theory to extract topological features of the flows in the heart in terms of discrete graphs, called partially cyclically ordered rooted tree (COT) representations, thereby identifying well-organized vortex flow structures as topological vortex structures and characterizing healthy blood flows as well as inefficient flow patterns in the diseased heart. Developing an image processing software based on mathematical theory, we have conducted the topological classification of 2D blood flow patterns obtained by visualization tools. This realizes a new image processing characterizing healthy blood flow patterns as well as inefficient patterns in diseased hearts.   

Hemodynamics of Coarctation of the Aorta - 4D Flow MRI Based Modeling Approach.

Coarctation of the aorta (COA) is one of the most common congenital heart defects. It is characterized by a narrowing in the aortic arch or proximal descending aorta. Treatment procedures for COA include surgery, intravascular stent placement, and balloon angioplasty. However, the selection criteria for the optimal treatment strategy are not well defined, and interventionalists often rely on their preference rather than patient-specific conditions. 1 in 3 COA patients have post-interventional secondary complications, likely due to suboptimal treatment.

4D Flow MRI is a diagnostic imaging technique capable of measuring time-resolved 3D blood velocities within a volumetric acquisition region. Numerous studies have used 4D Flow MRI to evaluate patient-specific blood flow information in COA. 4D Flow MRI has high potential to be used as a tool for treatment planning, however, imaging on its own provides no predictive capabilities. Image-based computational fluid dynamics (CFD) and in vitro modeling can be used to complement medical imaging, such as virtually testing and simulating interventions. A major challenge hindering image-based CFD from achieving widespread clinical adoption is its high computational cost. This calls for reduced order modeling strategies.

In-vivo 4D Flow MRI was performed in  pediatric patients before and after COA repair following an IRB-approved HIPAA-compliant protocol. The thoracic aorta was segmented, and centerlines were generated using semi-automatic software Mimics (Materialise NV, Leuven, Belgium). Flow rates at the ascending aorta (AAo), brachiocephalic (BCA), left common carotid (LCCA), left subclavian arteries (LSA), and descending aorta (DAo) were obtained using EnSight (Ansys, Inc., Canonsburg, PA, USA). Information from in vivo 4D Flow MRI was used to generate anatomical models to perfrom CFD and to construct anatomically realistic models. Particle image Velocimetry (PIV) data was acquired in the fabricated models to validate the CFD and in vitro 4D Flow aquisitions. Diffirent flow dynamics metrics calculated with all methodologies were compared for all the cases. This study provides a comprehensive flow dynamics characterization of aortic coarctation.   

Machine learning-enhanced image-based blood flow modeling

This talk summarizes some of the recent advances in scientific machine learning and their applications in image-based cardiovascular fluid mechanics. Broadly, we will overview two classes of methods: sparse data-driven modeling and deep learning. We will discuss our recent work on different variants of deep learning such as autoencoders and physics-informed neural networks (PINN) as well as different sparse modeling approaches for super-resolution and denoising of 4D flow magnetic resonance imaging (MRI) data. We will also discuss challenges associated with using deep learning for robust surrogate modeling in patient-specific applications. We will specifically focus on interpretability and generalization issues and present a new explainable AI (XAI) approach based on functional data analysis for interpreting black-box deep learning models and enhancing extrapolation to unseen data.   

Validation of image-based direct numerical simulations of right ventricle against 4D flow

With recent advancements in medical imaging and high-performance computing capabilities, image-based computational modeling is becoming a powerful tool with diagnostic and predictive potential for cardiovascular disease. The high temporal and spatial resolution in direct numerical simulations (DNS) allows detailed quantification of flow patterns, especially within the right ventricle (RV), which has a more complex geometry and wall motion than the widely studied left ventricle (LV). Investigating intracardiac flow phenomena can eventually provide reliable biomarkers that can be used to monitor the evolution of postoperative morbidities and the timing of surgeries, especially in congenital heart disease (CHD). Despite these advantages, computational models may suffer from uncertainties due in part to kinematic reconstruction, boundary conditions, and geometric simplifications. Therefore, the validation of simulation results is of utmost importance. In this work, we will compare DNS of the flow inside the RV with conventional four-dimensional magnetic resonance imaging (4D flow). The comparison based on healthy control cases will include the kinematics of the RV through mass conservation, phase-averaged velocity and vorticity profiles, and volume and phase-averaged flow-related quantities such as kinetic energy and viscous energy loss.   

Learning Bayesian digital twins of flows from PC-MRI data

I formulate a Bayesian digital twin approach to the reconstruction of velocity fields from noisy and possibly sparse phase-contrast magnetic resonance imaging data. The method learns the most probable fluid dynamics model that fits the data by solving a Bayesian inverse Navier–Stokes (N-S) boundary value problem. This jointly reconstructs and segments the velocity data, and at the same time infers hidden quantities such as the hydrodynamic pressure and the wall shear stress, as well as their uncertainties. Using a Bayesian framework, I regularize the inverse problem by introducing a priori information about the unknown N-S parameters in the form of Gaussian random fields. This prior information is updated using the Navier–Stokes problem, an energy-based segmentation functional, and by requiring that the reconstruction is consistent with the data. I create an algorithm that solves this inverse problem, and test it for an experimental flow through a 3D-printed physical model of an aortic arch. I show that the method can successfully reconstruct noisy flow-MRI data in realistic geometries and high Reynolds numbers. Although this method has been developed for flow-MRI, it extends to other velocimetry methods such as ultrasound Doppler velocimetry, particle image velocimetry (PIV) and scalar image velocimetry (SIV). The reconstruction of in vivo cardiovascular and porous media flows are among the most interesting applications of this work.   

4D flow MRI error analysis and segmentation using principles of fluid dynamics

4D flow MRI measurements of blood flow are hindered by limited resolution, partial volume effects, and noise, affecting the accuracy of hemodynamic metrics related to vessel remodeling and vascular disease. Resolution limitations and noise also make the segmentation of 4D flow MRI data labor-intensive and challenging. Recently developed machine-learning methods require large training datasets and are limited to training-specific vasculature, disease conditions, and imaging systems. We develop automatic segmentation and error analysis methods for 4D flow MRI that are based on statistical inference, fluid dynamics principles, and imaging physics. In this approach, the flow-containing voxels are segmented by identifying net flow effects using the standardized difference of means (SDM) velocity. The error analysis of 4D flow-measured velocities is performed by assessing the local bias error of the measurement due to the intra-voxel distribution and quantifying the measurement uncertainty based on the local velocity error variance and error correlations. The algorithms are tested on synthetic 4D flow datasets, in vitro 4D flow data acquired in 3D printed phantoms, and in vivo 4D flow data in cerebral and thoracic vessels. 4D flow MRI error analysis enables quantitative comparison of datasets obtained in longitudinal studies, across patient populations, and with different MRI systems.