Uncertainty Quantification and Sensitivity Analysis for Computational FFR Estimation in Stable Coronary Artery Disease

Abstract

We present a framework for the estimation of the Fractional Flow Reserve index based on blood ow simulations that incorporate clinical imaging and patient-specic characteristics. The process of model design implies making choices in order to build a suitable mathematical model, e.g. simulating a 3D domain versus a 1D domain, modeling of peripheral resistances, determining the regions of interest, etc. Here we thoroughly evaluate the impact of such choices on FFR prediction accuracy by reduced-order models with respect to more complete models by means of uncertainty quantication and sensitivity analysis. Moreover, we assess the uncertainty of FFR predictions based on our framework with respect to input data, and further determine the most inuential inputs with sensitivity analysis, aiming at increasing the clinical usability of predictions by providing information on the reliability of model output on a per case basis. Analysis is carried out for a population of 13 patients for which 24 invasive FFR measurements are available. Our analysis conrms previously observed sources of uncertainty and provides insight into aspects to be improved in any model-based non-invasive FFR estimation method.

Fractional Flow Reserve Evaluated as Metric of Coronary Stenosis — A Mathematical Model Study

“…The dependence of FFR upon the degree of narrowing α is found by first the substitution of Equation (14) in Equation (1), i.e.,…”

“…Also based on CFD and using angiographic images it was demonstrated (13) that sensitivity analysis for physiological lesion significance was influenced less by coronary or lesion anatomy (33%) than by microvascular physiology (59%). Using a reducedorder model for the estimation of FFR (rather than 3D) based on blood flow simulations that incorporated clinical imaging and patient-specific characteristics, others found that model errors were small, and that uncertainty related to the factor by which peripheral resistance is reduced from baseline to hyperemic conditions proved to be the most influential parameter for FFR predictions, whereas uncertainty in stenosis geometry had greater effect in cases with low FFR (14). Similarly, 296 lesions were studied (15) and the authors compared (by linear regression) various clinically relevant measures, including diameter stenosis (R = 0.565), lesion length (R = 0.306), reference vessel cross sectional area (R = 0.195), and the myocardial supply area subtended by the coronary vessel under study (R = 0.504).…”

Cardiovascular models for personalised medicine: Where now and where next?

“…If the coronary artery can be segmented from medical image data, and an estimate of the personal myocardial resistance can be made, then FFR can be computed. This requires the coupling of a local three-dimensional model (or, indeed, a one dimensional model [56][57]) of personal coronary anatomy coupled to a personalised model of the distal resistance. Morris et al [58] review the challenges and limitations of the computation of FFR, including the question of the estimation of distal resistance in an individual.…”

Multiple Aneurysms AnaTomy CHallenge 2018 (MATCH): uncertainty quantification of geometric rupture risk parameters

“…Uncertainties caused by different image segmentation methods will result in uncertainties in geometric and hemodynamic parameters used for rupture risk prediction. These uncertainties are major factors preventing successful introduction of image-based methods in clinical practice [20, 21].…”