Data-driven cardiovascular flow modelling: examples and opportunities

Accelerated simulation methodologies for computational vascular flow modelling

Vascular flow modelling can improve our understanding of vascular pathologies and aid in developing safe and effective medical devices. Vascular flow models typically involve solving the nonlinear Navier-Stokes equations in complex anatomies and using physiological boundary conditions, often presenting a multi-physics and multi-scale computational problem to be solved. This leads to highly complex and expensive models that require excessive computational time. This review explores accelerated simulation methodologies, specifically focusing on computational vascular flow modelling. We review reduced order modelling (ROM) techniques like zero-/one-dimensional and modal decomposition-based ROMs and machine learning (ML) methods including ML-augmented ROMs, ML-based ROMs and physics-informed ML models. We discuss the applicability of each method to vascular flow acceleration and the effectiveness of the method in addressing domain-specific challenges. When available, we provide statistics on accuracy and speed-up factors for various applications related to vascular flow simulation acceleration. Our findings indicate that each type of model has strengths and limitations depending on the context. To accelerate real-world vascular flow problems, we propose future research on developing multi-scale acceleration methods capable of handling the significant geometric variability inherent to such problems.

A predictive surrogate model for hemodynamics and structural prediction in abdominal aorta for different physiological conditions

BACKGROUND AND OBJECTIVE

This study investigates the application of a Predictive Surrogate Model (PSM) for the prediction of the fluid and solid variables in the abdominal aorta by integrating Proper Orthogonal Decomposition (POD) and Long Short-Term Memory (LSTM) techniques.

METHODS

The Fluid-Structure Interaction (FSI) solver, which serves as the Full-Order Model (FOM), can capture the blood hemodynamics and structural mechanics precisely for a variety of physiological states, namely the rest and exercise conditions.

RESULTS

Detailed analyses have been conducted on velocity components, pressure, Wall Shear Stress (WSS), and Oscillatory Shear Index (OSI) variables. Firstly, the reconstruction error has been derived based on a specific number of POD bases to assess the Reduced Order Model (ROM). Notably, the reconstruction error for velocity components in the rest condition is one order of magnitude higher than that in the exercise condition, yet both remained below 10%. This error for pressure is even more minimal, being less than 1%.

CONCLUSIONS

The PSM is evaluated against rest and exercise conditions, exhibiting promising results despite the inherent complexities of the physiological conditions. Despite the inherent complexities of phenomena in the aorta, the predictive model demonstrates consistent error magnitudes for velocity components and wall-related indices, while solid variables show slightly higher errors.

A systematic review of clinical and biomechanical engineering perspectives on the prediction of restenosis in coronary and peripheral arteries

Objective

Restenosis is a significant complication of revascularization treatments in coronary and peripheral arteries, sometimes necessitating repeated intervention. Establishing when restenosis will happen is extremely difficult due to the interplay of multiple variables and factors. Standard clinical and Doppler ultrasound scans surveillance follow-ups are the only tools clinicians can rely on to monitor intervention outcomes. However, implementing efficient surveillance programs is hindered by health care system limitations, patients' comorbidities, and compliance. Predictive models classifying patients according to their risk of developing restenosis over a specific period will allow the development of tailored surveillance, prevention programs, and efficient clinical workflows. This review aims to: (1) summarize the state-of-the-art in predictive models for restenosis in coronary and peripheral arteries; (2) compare their performance in terms of predictive power; and (3) provide an outlook for potentially improved predictive models.

Methods

We carried out a comprehensive literature review by accessing the PubMed/MEDLINE database according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The search strategy consisted of a combination of keywords and included studies focusing on predictive models of restenosis published between January 1993 and April 2023. One author independently screened titles and abstracts and checked for eligibility. The rest of the authors independently confirmed and discussed in case of any disagreement. The search of published literature identified 22 studies providing two perspectives-clinical and biomechanical engineering-on restenosis and comprising distinct methodologies, predictors, and study designs. We compared predictive models' performance on discrimination and calibration aspects. We reported the performance of models simulating reocclusion progression, evaluated by comparison with clinical images.

Results

Clinical perspective studies consider only routinely collected patient information as restenosis predictors. Our review reveals that clinical models adopting traditional statistics (n = 14) exhibit only modest predictive power. The latter improves when machine learning algorithms (n = 4) are employed. The logistic regression models of the biomechanical engineering perspective (n = 2) show enhanced predictive power when hemodynamic descriptors linked to restenosis are fused with a limited set of clinical risk factors. Biomechanical engineering studies simulating restenosis progression (n = 2) are able to capture its evolution but are computationally expensive and lack risk scoring for individual patients at specific follow-ups.

Conclusions

Restenosis predictive models, based solely on routine clinical risk factors and using classical statistics, inadequately predict the occurrence of restenosis. Risk stratification models with increased predictive power can be potentially built by adopting machine learning techniques and incorporating critical information regarding vessel hemodynamics arising from biomechanical engineering analyses.

Towards Reduced Order Models via Robust Proper Orthogonal Decomposition to capture personalised aortic haemodynamics

Data driven, reduced order modelling has shown promise in tackling the challenges associated with computational and experimental haemodynamic models. In this work, we focus on the use of Reduced Order Models (ROMs) to reconstruct velocity fields in a patient-specific dissected aorta, with the objective being to compare the ROMs obtained from Robust Proper Orthogonal Decomposition (RPOD) to those obtained from the traditional Proper Orthogonal Decomposition (POD). POD and RPOD are applied to in vitro, haemodynamic data acquired by Particle Image Velocimetry and compare the decomposed flows to those derived from Computational Fluid Dynamics (CFD) data for the same geometry and flow conditions. In this work, PIV and CFD results act as surrogates for clinical haemodynamic data e.g. MR, helping to demonstrate the potential use of ROMS in real clinical scenarios. The flow is reconstructed using different numbers of POD modes and the flow features obtained throughout the cardiac cycle are compared to the original Full Order Models (FOMs). Robust Principal Component Analysis (RPCA), the first step of RPOD, has been found to enhance the quality of PIV data, allowing POD to capture most of the kinetic energy of the flow in just two modes similar to the numerical data that are free from measurement noise. The reconstruction errors differ along the cardiac cycle with diastolic flows requiring more modes for accurate reconstruction. In general, modes 1-10 are found sufficient to represent the flow field. The results demonstrate that the coherent structures that characterise this aortic dissection flow are described by the first few POD modes suggesting that it is possible to represent the macroscale behaviour of aortic flow in a low-dimensional space; thus significantly simplifying the problem, and allowing for more computationally efficient flow simulations or machine learning based flow predictions that can pave the way for translation of such models to the clinic.

Aneurysmal growth in type-B aortic dissection: assessing the impact of patient-specific inlet conditions on key haemodynamic indices

Type-B aortic dissection is a cardiovascular disease in which a tear develops in the intimal layer of the descending aorta, allowing pressurized blood to delaminate the layers of the vessel wall. In medically managed patients, long-term aneurysmal dilatation of the false lumen (FL) is considered virtually inevitable and is associated with poorer disease outcomes. While the pathophysiological mechanisms driving FL dilatation are not yet understood, haemodynamic factors are believed to play a key role. Computational fluid dynamics (CFD) and 4D-flow MRI (4DMR) analyses have revealed correlations between flow helicity, oscillatory wall shear stress and aneurysmal dilatation of the FL. In this study, we compare CFD simulations using a patient-specific, three-dimensional, three-component inlet velocity profile (4D IVP) extracted from 4DMR data against simulations with flow rate-matched uniform and axial velocity profiles that remain widely used in the absence of 4DMR. We also evaluate the influence of measurement errors in 4DMR data by scaling the 4D IVP to the degree of imaging error detected in prior studies. We observe that oscillatory shear and helicity are highly sensitive to inlet velocity distribution and flow volume throughout the FL and conclude that the choice of IVP may greatly affect the future clinical value of simulations.

Computer-aided shape features extraction and regression models for predicting the ascending aortic aneurysm growth rate

OBJECTIVE

ascending aortic aneurysm growth prediction is still challenging in clinics. In this study, we evaluate and compare the ability of local and global shape features to predict the ascending aortic aneurysm growth.

MATERIAL AND METHODS

70 patients with aneurysm, for which two 3D acquisitions were available, are included. Following segmentation, three local shape features are computed: (1) the ratio between maximum diameter and length of the ascending aorta centerline, (2) the ratio between the length of external and internal lines on the ascending aorta and (3) the tortuosity of the ascending tract. By exploiting longitudinal data, the aneurysm growth rate is derived. Using radial basis function mesh morphing, iso-topological surface meshes are created. Statistical shape analysis is performed through unsupervised principal component analysis (PCA) and supervised partial least squares (PLS). Two types of global shape features are identified: three PCA-derived and three PLS-based shape modes. Three regression models are set for growth prediction: two based on gaussian support vector machine using local and PCA-derived global shape features; the third is a PLS linear regression model based on the related global shape features. The prediction results are assessed and the aortic shapes most prone to growth are identified.

RESULTS

the prediction root mean square error from leave-one-out cross-validation is: 0.112 mm/month, 0.083 mm/month and 0.066 mm/month for local, PCA-based and PLS-derived shape features, respectively. Aneurysms close to the root with a large initial diameter report faster growth.

CONCLUSION

global shape features might provide an important contribution for predicting the aneurysm growth.

Feasibility of neural network metamodels for emulation and sensitivity analysis of radionuclide transport models

In this paper we compare the outputs of neural network metamodels with numerical solutions of differential equation models in modeling cesium-137 transportation in sand. Convolutional neural networks (CNNs) were trained with differential equation simulation results. Training sets of various sizes (from 5120 to 163,840) were used. First order and total order Sobol methods were applied to both models in order to test the feasibility of neural network metamodels for sensitivity analysis of a radionuclide transport model. Convolutional neural networks were found to be capable of emulating the differential equation models with high accuracy when the training set size was 40,960 or higher. Neural network metamodels also gave similar results compared with the numerical solutions of the partial differential equation model in sensitivity analysis.

4-dimensional local radial basis function interpolation of large, uniformly spaced datasets

BACKGROUND AND OBJECTIVE

Large, uniformly spaced, complex and time varying datasets derived from high resolution medical image velocimetry can provide a wealth of information regarding small-scale transient physiological flow phenomena and pulsation of anatomical boundaries. However, there remains a need for interpolation techniques to effectively reconstruct a fully 4-dimensional functional relationship from this data. This paper presents a preliminary evaluation of a 4-dimensional local radial basis function (RBF) algorithm as a means of addressing this problem for laminar flows.

METHODS

A 4D interpolation algorithm is proposed based on a Local Hermitian Interpolation (LHI) using a combination of multi-quadric RBF with a partition of unity scheme. The domain is divided into uniform sub-systems with size restricted to immediately neighbouring points. The validity of the algorithm is first established on a known 4D analytical dataset and a CFD based laminar flow phantom. Application is then demonstrated through characterisation of a large 4D laminar flow dataset obtained from magnetic resonance imaging (MRI) measurements of cerebrospinal fluid velocities in the brain.

RESULTS

Performance of the algorithm is compared to that of a quad-linear interpolation, demonstrating favourable improvement in accuracy. The technique is shown to be robust, computationally efficient and capable of refined interpolation in Euclidean space and time. Application to MR velocimetry data is shown to produce promising results for the 4D reconstruction of the transient flow field and movement of the fluid boundaries at spatial and temporal locations intermediate to the original data.

CONCLUSION

This study has demonstrated feasibility of an accurate, stable and efficient 4-dimensional local RBF interpolation method for large, transient laminar flow velocimetry datasets. The proposed approach does not suffer from ill-conditioning or high computational cost due to domain decomposition into local stencils where the RBF is only ever applied to a limited number of points. This work offers a potential tool to assist medical diagnoses and drug delivery through better understanding of physiological flow fields such as cerebrospinal fluid. Further work will evaluate the technique on a wider range of flow fields and against CFD simulation.

Machine Learning for Aiding Blood Flow Velocity Estimation Based on Angiography

Computational fluid dynamics (CFD) is widely employed to predict hemodynamic characteristics in arterial models, while not friendly to clinical applications due to the complexity of numerical simulations. Alternatively, this work proposed a framework to estimate hemodynamics in vessels based on angiography images using machine learning (ML) algorithms. First, the iodine contrast perfusion in blood was mimicked by a flow of dye diffusing into water in the experimentally validated CFD modeling. The generated projective images from simulations imitated the counterpart of light passing through the flow field as an analogy of X-ray imaging. Thus, the CFD simulation provides both the ground truth velocity field and projective images of dye flow patterns. The rough velocity field was estimated using the optical flow method (OFM) based on 53 projective images. ML training with least absolute shrinkage, selection operator and convolutional neural network was conducted with CFD velocity data as the ground truth and OFM velocity estimation as the input. The performance of each model was evaluated based on mean absolute error and mean squared error, where all models achieved or surpassed the criteria of 3 × 10−3 and 5 × 10−7 m/s, respectively, with a standard deviation less than 1 × 10−6 m/s. Finally, the interpretable regression and ML models were validated with over 613 image sets. The validation results showed that the employed ML model significantly reduced the error rate from 53.5% to 2.5% on average for the v-velocity estimation in comparison with CFD. The ML framework provided an alternative pathway to support clinical diagnosis by predicting hemodynamic information with high efficiency and accuracy.

Physics-Informed Neural Networks for Brain Hemodynamic Predictions Using Medical Imaging

Determining brain hemodynamics plays a critical role in the diagnosis and treatment of various cerebrovascular diseases. In this work, we put forth a physics-informed deep learning framework that augments sparse clinical measurements with one-dimensional (1D) reduced-order model (ROM) simulations to generate physically consistent brain hemodynamic parameters with high spatiotemporal resolution. Transcranial Doppler (TCD) ultrasound is one of the most common techniques in the current clinical workflow that enables noninvasive and instantaneous evaluation of blood flow velocity within the cerebral arteries. However, it is spatially limited to only a handful of locations across the cerebrovasculature due to the constrained accessibility through the skull's acoustic windows. Our deep learning framework uses in vivo real-time TCD velocity measurements at several locations in the brain combined with baseline vessel cross-sectional areas acquired from 3D angiography images and provides high-resolution maps of velocity, area, and pressure in the entire brain vasculature. We validate the predictions of our model against in vivo velocity measurements obtained via four-dimensional (4D) flow magnetic resonance imaging (MRI) scans. We then showcase the clinical significance of this technique in diagnosing cerebral vasospasm (CVS) by successfully predicting the changes in vasospastic local vessel diameters based on corresponding sparse velocity measurements. We show this capability by generating synthetic blood flow data after cerebral vasospasm at various levels of stenosis. Here, we demonstrate that the physics-based deep learning approach can estimate and quantify the subject-specific cerebral hemodynamic variables with high accuracy despite lacking knowledge of inlet and outlet boundary conditions, which is a significant limitation for the accuracy of the conventional purely physics-based computational models.

Modal Decomposition Techniques: Application in Coherent Structures for a Saccular Aneurysm Model

Aneurysms are localized expansions of blood vessels which can be fatal upon rupture. Studies have shown that aneurysm flows exhibit complex flow phenomena which consist of single or multiple vortical structures that move within the flow cycle. Understanding the complex flow behaviors of aneurysms remain challenging. Thus, the goal of this study is to quantify the flow behavior and extract physical insights into aneurysm flows using advance data decomposition methods, Proper Orthogonal Decomposition (POD) and Dynamic Mode Decomposition (DMD). The velocity field data were obtained by performing 2D Particle Image Velocimetry (2D PIV) on the mid-plane of an idealized, rigid, saccular aneurysm model. The input flow conditions were set to Rep=50 and 150 for a fixed α=2 using a precisely controlled piston pump system. POD was used to quantify the spatial features of the flows, while DMD was used to obtain insight on the dynamics. The results obtained from POD and DMD showed the capability of both methods to quantify the flow field, with the modes obtained providing different insights into the flow evolution in the aneurysm. The curve-fitting step of the POD time-varying coefficients, and the appropriate selection of DMD modes based on their energy contribution, allowed the mathematical flow models from POD and DMD to reconstruct flow fields at any given time step. This can be used for validation of numerical or computational data.

Application of Dynamic Mode Decomposition to Study Temporal Flow Behavior in a Saccular Aneurysm

Aneurysms are abnormal expansion of weakened blood vessels which can cause mortality or long-term disability upon rupture. Several studies have shown that inflow conditions spatially and temporally influence aneurysm flow behavior. The objective of this investigation is to identify impact of inflow conditions on spatio-temporal flow behavior in an aneurysm using dynamic mode decomposition (DMD). For this purpose, low-frame rate velocity field measurements are performed in an idealized aneurysm model using particle image velocimetry (PIV). The inflow conditions are precisely controlled using a ViVitro SuperPump system where nondimensional fluid parameters such as peak Reynolds number (Rep) and Womersely number (α) are varied from 50-270 and 2-5, respectively. The results show the ability of DMD to identify the spatial flow structures and their frequency content. Furthermore, DMD captured the impact of inflow conditions, and change in mode shapes, amplitudes, frequency, and growth rate information is observed. The DMD low-order flow reconstruction also showed the complex interplay of flow features for each inflow scenario. Furthermore, the low-order reconstruction results provided a mathematical description of the flow behavior in the aneurysm which captured the vortex formation, evolution, and convection in detail. These results indicated that the vortical structure behavior varied with the change in α while its strength and presence of secondary structures are influenced by the change in Rep.

Machine Learning for Cardiovascular Biomechanics Modeling: Challenges and Beyond

Recent progress in machine learning (ML), together with advanced computational power, have provided new research opportunities in cardiovascular modeling. While classifying patient outcomes and medical image segmentation with ML have already shown significant promising results, ML for the prediction of biomechanics such as blood flow or tissue dynamics is in its infancy. This perspective article discusses some of the challenges in using ML for replacing well-established physics-based models in cardiovascular biomechanics. Specifically, we discuss the large landscape of input features in 3D patient-specific modeling as well as the high-dimensional output space of field variables that vary in space and time. We argue that the end purpose of such ML models needs to be clearly defined and the tradeoff between the loss in accuracy and the gained speedup carefully interpreted in the context of translational modeling. We also discuss several exciting venues where ML could be strategically used to augment traditional physics-based modeling in cardiovascular biomechanics. In these applications, ML is not replacing physics-based modeling, but providing opportunities to solve ill-defined problems, improve measurement data quality, enable a solution to computationally expensive problems, and interpret complex spatiotemporal data by extracting hidden patterns. In summary, we suggest a strategic integration of ML in cardiovascular biomechanics modeling where the ML model is not the end goal but rather a tool to facilitate enhanced modeling.

Computational Methods for Fluid-Structure Interaction Simulation of Heart Valves in Patient-Specific Left Heart Anatomies

Given the complexity of human left heart anatomy and valvular structures, the fluid–structure interaction (FSI) simulation of native and prosthetic valves poses a significant challenge for numerical methods. In this review, recent numerical advancements for both fluid and structural solvers for heart valves in patient-specific left hearts are systematically considered, emphasizing the numerical treatments of blood flow and valve surfaces, which are the most critical aspects for accurate simulations. Numerical methods for hemodynamics are considered under both the continuum and discrete (particle) approaches. The numerical treatments for the structural dynamics of aortic/mitral valves and FSI coupling methods between the solid Ωs and fluid domain Ωf are also reviewed. Future work toward more advanced patient-specific simulations is also discussed, including the fusion of high-fidelity simulation within vivo measurements and physics-based digital twining based on data analytics and machine learning techniques.

Challenges in dynamic mode decomposition

Dynamic mode decomposition (DMD) is a powerful tool for extracting spatial and temporal patterns from multi-dimensional time series, and it has been used successfully in a wide range of fields, including fluid mechanics, robotics and neuroscience. Two of the main challenges remaining in DMD research are noise sensitivity and issues related to Krylov space closure when modelling nonlinear systems. Here, we investigate the combination of noise and nonlinearity in a controlled setting, by studying a class of systems with linear latent dynamics which are observed via multinomial observables. Our numerical models include system and measurement noise. We explore the influences of dataset metrics, the spectrum of the latent dynamics, the normality of the system matrix and the geometry of the dynamics. Our results show that even for these very mildly nonlinear conditions, DMD methods often fail to recover the spectrum and can have poor predictive ability. Our work is motivated by our experience modelling multilegged robot data, where we have encountered great difficulty in reconstructing time series for oscillatory systems with intermediate transients, which decay only slightly faster than a period.

Data Assimilation by Stochastic Ensemble Kalman Filtering to Enhance Turbulent Cardiovascular Flow Data From Under-Resolved Observations

We propose a data assimilation methodology that can be used to enhance the spatial and temporal resolution of voxel-based data as it may be obtained from biomedical imaging modalities. It can be used to improve the assessment of turbulent blood flow in large vessels by combining observed data with a computational fluid dynamics solver. The methodology is based on a Stochastic Ensemble Kalman Filter (SEnKF) approach and geared toward pulsatile and turbulent flow configurations. We describe the observed flow fields by a mean value and its covariance. These flow fields are combined with forecasts obtained from a direct numerical simulation of the flow field. The method is validated against canonical pulsatile and turbulent flows. Finally, it is applied to a clinically relevant configuration, namely the flow downstream of a bioprosthetic valve in an aorta phantom. It is demonstrated how the 4D flow field obtained from experimental observations can be enhanced by the data assimilation algorithm. Results show that the presented method is promising for future use with in vivo data from 4D Flow Magnetic Resonance Imaging (4D Flow MRI). 4D Flow MRI returns spatially and temporally averaged flow fields that are limited by the spatial and the temporal resolution of the tool. These averaged flow fields and the associated uncertainty might be used as observation data in the context of the proposed methodology.

Integrating multi-fidelity blood flow data with reduced-order data assimilation

High-fidelity patient-specific modeling of cardiovascular flows and hemodynamics is challenging. Direct blood flow measurement inside the body with in-vivo measurement modalities such as 4D flow magnetic resonance imaging (4D flow MRI) suffer from low resolution and acquisition noise. In-vitro experimental modeling and patient-specific computational fluid dynamics (CFD) models are subject to uncertainty in patient-specific boundary conditions and model parameters. Furthermore, collecting blood flow data in the near-wall region (e.g., wall shear stress) with experimental measurement modalities poses additional challenges. In this study, a computationally efficient data assimilation method called reduced-order modeling Kalman filter (ROM-KF) was proposed, which combined a sequential Kalman filter with reduced-order modeling using a linear model provided by dynamic mode decomposition (DMD). The goal of ROM-KF was to overcome low resolution and noise in experimental and uncertainty in CFD modeling of cardiovascular flows. The accuracy of the method was assessed with 1D Womersley flow, 2D idealized aneurysm, and 3D patient-specific cerebral aneurysm models. Synthetic experimental data were used to enable direct quantification of errors using benchmark datasets. The accuracy of ROM-KF in reconstructing near-wall hemodynamics was assessed by applying the method to problems where near-wall blood flow data were missing in the experimental dataset. The ROM-KF method provided blood flow data that were more accurate than the computational and synthetic experimental datasets and improved near-wall hemodynamics quantification.

Assessment and visualization of hemodynamic loading in aneurysm sac and neck: Effect of foam insertion

Shape memory polymer (SMP) foam is often proposed as the future alternative of coils in aneurysm treatment devices. Present work numerically investigates the unsteady, three-dimensional simulation of blood flow in a cerebral aneurysm filled with SMP foam. Simulations are conducted on patient-specific geometries with realistic blood velocity waveform imposed at the inlet while SMP foam is treated as a porous medium. The present study introduces a "loading risk map" that helps to visualize the hemodynamic effect of foam insertion on the aneurysm sac and neck. The loading risk maps suggest that while the SMP foam subdues the flow and wall shear pulsations in the aneurysm sac, the pressure distribution is minimally affected. The maps suggest that while the downstream lip is the most risk-prone site for both geometries, downstream vascular anatomy significantly influences foam efficiency in reducing pressure and wall shear stress loading.