Model-based assessment of cardiovascular health from noninvasive measurements

Leveraging the potential of machine learning for assessing vascular ageing: state-of-the-art and future research

Vascular ageing biomarkers have been found to be predictive of cardiovascular risk independently of classical risk factors, yet are not widely used in clinical practice. In this review, we present two basic approaches for using machine learning (ML) to assess vascular age: parameter estimation and risk classification. We then summarize their role in developing new techniques to assess vascular ageing quickly and accurately. We discuss the methods used to validate ML-based markers, the evidence for their clinical utility, and key directions for future research. The review is complemented by case studies of the use of ML in vascular age assessment which can be replicated using freely available data and code.

Noninvasive Cardiac Output and Central Systolic Pressure From Cuff-Pressure and Pulse Wave Velocity

GOAL

We introduce a novel approach to estimate cardiac output (CO) and central systolic blood pressure (cSBP) from noninvasive measurements of peripheral cuff-pressure and carotid-to-femoral pulse wave velocity (cf-PWV).

METHODS

The adjustment of a previously validated one-dimensional arterial tree model is achieved via an optimization process. In the optimization loop, compliance and resistance of the generic arterial tree model as well as aortic flow are adjusted so that simulated brachial systolic and diastolic pressures and cf-PWV converge towards the measured brachial systolic and diastolic pressures and cf-PWV. The process is repeated until full convergence in terms of both brachial pressures and cf-PWV is reached. To assess the accuracy of the proposed framework, we implemented the algorithm on in vivo anonymized data from 20 subjects and compared the method-derived estimates of CO and cSBP to patient-specific measurements obtained with Mobil-O-Graph apparatus (central pressure) and two-dimensional transthoracic echocardiography (aortic blood flow).

RESULTS

Both CO and cSBP estimates were found to be in good agreement with the reference values achieving an RMSE of 0.36 L/min and 2.46 mmHg, respectively. Low biases were reported, namely -0.04 ± 0.36 L/min for CO predictions and -0.27 ± 2.51 mmHg for cSBP predictions.

SIGNIFICANCE

Our one-dimensional model can be successfully "tuned" to partially patient-specific standards by using noninvasive, easily obtained peripheral measurement data. The in vivo evaluation demonstrated that this method can potentially be used to obtain central aortic hemodynamic parameters in a noninvasive and accurate way.

Subject-specific pulse wave propagation modeling: Towards enhancement of cardiovascular assessment methods

Cardiovascular diseases are the leading cause of death worldwide. Pulse wave analysis (PWA) technique, which reconstructs and analyses aortic pressure waveform based on non-invasive peripheral pressure recording, became an important bioassay for cardiovascular assessment in a general population. The aim of our study was to establish a pulse wave propagation modeling framework capable of matching clinical PWA data from healthy individuals on a per-subject basis. Radial pressure profiles from 20 healthy individuals (10 males, 10 females), with mean age of 42 ± 10 years, were recorded using applanation tonometry (SphygmoCor, AtCor Medical, Australia) and used to estimate subject-specific parameters of mathematical model of blood flow in the system of fifty-five arteries. The model was able to describe recorded pressure profiles with high accuracy (mean absolute percentage error of 1.87 ± 0.75%) when estimating only 6 parameters for each subject. Cardiac output (CO) and stroke volume (SV) have been correctly identified by the model as lower in females than males (CO of 3.57 ± 0.54 vs. 4.18 ± 0.72 L/min with p-value < 0.05; SV of 49.5 ± 10.1 vs. 64.2 ± 16.8 ml with p-value = 0.076). Moreover, the model identified age related changes in the heart function, i.e. that the cardiac output at rest is maintained with age (r = 0.23; p-value = 0.32) despite the decreasing heart rate (r = −0.49; p-value < 0.05), because of the increase in stroke volume (r = 0.46; p-value < 0.05). Central PWA indices derived from recorded waveforms strongly correlated with those obtained using corresponding model-predicted radial waves (r > 0.99 and r > 0.97 for systolic (SP) and diastolic (DP) pressures, respectively; r > 0.77 for augmentation index (AI); all p—values < 0.01). Model-predicted central waveforms, however, had higher SP than those reconstructed by PWA using recorded radial waves (5.6 ± 3.3 mmHg on average). From all estimated subject-specific parameters only the time to the peak of heart ejection profile correlated with clinically measured AI. Our study suggests that the proposed model may serve as a tool to computationally investigate virtual patient scenarios mimicking different cardiovascular abnormalities. Such a framework can augment our understanding and help with the interpretation of PWA results.

Patient-specific parameter estimation in single-ventricle lumped circulation models under uncertainty

Computational models of cardiovascular physiology can inform clinical decision-making, providing a physically consistent framework to assess vascular pressures and flow distributions, and aiding in treatment planning. In particular, lumped parameter network (LPN) models that make an analogy to electrical circuits offer a fast and surprisingly realistic method to reproduce the circulatory physiology. The complexity of LPN models can vary significantly to account, for example, for cardiac and valve function, respiration, autoregulation, and time-dependent hemodynamics. More complex models provide insight into detailed physiological mechanisms, but their utility is maximized if one can quickly identify patient specific parameters. The clinical utility of LPN models with many parameters will be greatly enhanced by automated parameter identification, particularly if parameter tuning can match non-invasively obtained clinical data. We present a framework for automated tuning of 0D lumped model parameters to match clinical data. We demonstrate the utility of this framework through application to single ventricle pediatric patients with Norwood physiology. Through a combination of local identifiability, Bayesian estimation and maximum a posteriori simplex optimization, we show the ability to automatically determine physiologically consistent point estimates of the parameters and to quantify uncertainty induced by errors and assumptions in the collected clinical data. We show that multi-level estimation, that is, updating the parameter prior information through sub-model analysis, can lead to a significant reduction in the parameter marginal posterior variance. We first consider virtual patient conditions, with clinical targets generated through model solutions, and second application to a cohort of four single-ventricle patients with Norwood physiology. Copyright © 2016 John Wiley & Sons, Ltd.

On a sparse pressure-flow rate condensation of rigid circulation models

Cardiovascular simulation has shown potential value in clinical decision-making, providing a framework to assess changes in hemodynamics produced by physiological and surgical alterations. State-of-the-art predictions are provided by deterministic multiscale numerical approaches coupling 3D finite element Navier Stokes simulations to lumped parameter circulation models governed by ODEs. Development of next-generation stochastic multiscale models whose parameters can be learned from available clinical data under uncertainty constitutes a research challenge made more difficult by the high computational cost typically associated with the solution of these models. We present a methodology for constructing reduced representations that condense the behavior of 3D anatomical models using outlet pressure-flow polynomial surrogates, based on multiscale model solutions spanning several heart cycles. Relevance vector machine regression is compared with maximum likelihood estimation, showing that sparse pressure/flow rate approximations offer superior performance in producing working surrogate models to be included in lumped circulation networks. Sensitivities of outlets flow rates are also quantified through a Sobol׳ decomposition of their total variance encoded in the orthogonal polynomial expansion. Finally, we show that augmented lumped parameter models including the proposed surrogates accurately reproduce the response of multiscale models they were derived from. In particular, results are presented for models of the coronary circulation with closed loop boundary conditions and the abdominal aorta with open loop boundary conditions.

Assessment of cardiovascular function by combining clinical data with a computational model of the cardiovascular system

OBJECTIVE

A sufficient understanding of patients' cardiovascular status is necessary for doctors to make the best decisions with regard to the treatment of cardiovascular disease; however, it is often not available because of the limitation of clinical measurements. The objective of this study was to examine whether cardiovascular function can be assessed quantitatively and for specific patients by combining clinical data with a computational model of the cardiovascular system.

METHODS

Seven consecutive patients undergoing off-pump coronary artery bypass grafting were enrolled in this study. The clinical data were collected both during the preoperative diagnosis and during the operation. Sensitivity analysis was performed to select the major model parameters most relevant to the measured data. The major model parameters were then estimated through a data-fitting procedure, enabling a patient-specific quantitative assessment of various aspects of cardiovascular function.

RESULTS

The results revealed the prevalence of left ventricular diastolic dysfunction in the patients, although the severity of dysfunction exhibits significant interpatient variability (the estimated left ventricular passive elastance varies from 194% to 540% of its reference value). Moreover, 4 of the 7 patients studied had impaired left ventricular systolic function.

CONCLUSIONS

The current study demonstrates the feasibility of assessing cardiovascular function quantitatively by combining clinical data with a cardiovascular model. In particular, the assessment utilizes the measurements already in use or available in clinical settings, enhancing the clinical potential of the proposed method.

Continuous and less invasive central hemodynamic monitoring by blood pressure waveform analysis

Blood pressure waveform analysis may permit continuous (i.e., automated) and less invasive (i.e., safer and simpler) central hemodynamic monitoring in the intensive care unit and other clinical settings without requiring any instrumentation beyond what is already in use or available. This practical approach has been a topic of intense investigation for decades and may garner even more interest henceforth due to the evolving demographics as well as recent trends in clinical hemodynamic monitoring. Here, we review techniques that have appeared in the literature for mathematically estimating clinically significant central hemodynamic variables, such as cardiac output, from different blood pressure waveforms. We begin by providing the rationale for pursuing such techniques. We then summarize earlier techniques and thereafter overview recent techniques by our collaborators and us in greater depth while pinpointing both their strengths and weaknesses. We conclude with suggestions for future research directions in the field and a description of some potential clinical applications of the techniques.

Cardiovascular simulation toolbox

A toolbox for Matlab Simulink (trademark of Mathworks corp. etc.) was developed to simulate various models of flow in the cardiovascular system and study effects of different pathological conditions. The toolbox was based on well-known analog lumped models of blood flow in vessels, the varying elastance heart model, blood flow through vessels, shunts, and valves as well as models of oxygen exchange at lungs and tissue. The toolbox is modular providing the basic building blocks of the cardiovascular system. Parameters for the individual components may be set by the user to adapt the component to the simulated system. Several examples are shown. This modeling system is described and is also available for downloading as an open source for free use. The authors see this as the basis for wide collaboration and standardization in modeling. A web site will be available for accepting contributions from other researchers and to create a free exchange.

Mathematical modeling of cardiovascular system dynamics using a lumped parameter method

This work reviews the main aspects of cardiovascular system dynamics with an emphasis on modeling hemodynamic characteristics by the use of a lumped parameter approach. The methodological and physiological aspects of the circulation dynamics are summarized with the help of existing mathematical models. The main characteristics of the hemodynamic elements, such as the heart and arterial and venous systems, are first described. Distributed models of an arterial network are introduced, and their characteristics are compared with those of lumped parameter models. We also discuss the nonlinear characteristics of the pressure-volume relationship in veins. Then the control pathways that participate in feedback mechanisms (baroreceptors and cardiopulmonary receptors) are described to explain the interaction between hemodynamics and autonomic nerve control in the circulation. Based on a set-point model, the computational aspects of reflex control are explained.