Uncertainty Quantification and Sensitivity Analysis for Non-invasive Model-Based Instantaneous Wave-Free Ratio...

Area of Science:

• Cardiovascular Physiology
• Computational Fluid Dynamics
• Medical Imaging

Background:

• Accurate estimation of coronary haemodynamics is crucial for diagnosing and managing stable coronary artery disease.
• Patient-specific computational models offer a non-invasive approach to assess haemodynamic indices like the instantaneous wave-free ratio (iFR).
• Uncertainties in input parameters can affect the reliability of these computational predictions.

Purpose of the Study:

• To validate a 1D-0D unsteady solver with a distributed stenosis model for patient-specific estimation of resting haemodynamic indices.
• To assess the sensitivity of iFR predictions to uncertainties in input parameters, including boundary conditions and vascular geometry.
• To compare 1D-0D model results with 3D models and invasive measurements.

Main Methods:

• Development and validation of a 1D-0D unsteady solver incorporating a distributed stenosis model.
• Application of a polynomial chaos approach to quantify uncertainty in iFR predictions based on input parameter variations.
• Comparison of simulated iFR with invasive measurements from 52 patients with stable coronary artery disease.

Main Results:

• Satisfactory agreement was found between 1D-0D/3D model-derived iFR and invasive measurements (bias 0.0-0.005).
• Sensitivity analysis revealed that iFR predictions are most influenced by uncertainties in vascular geometry and coronary flow.
• The 1D-0D method showed a slight overestimation of invasive iFR (bias -0.036), highlighting the need for improved flow estimates.

Conclusions:

• The validated 1D-0D solver provides a reliable tool for patient-specific haemodynamic index estimation.
• Vascular geometry and coronary flow are critical parameters for accurate iFR prediction, while pressure waveform uncertainties have minimal impact.
• Steady-state simulations demonstrate strong correlation with unsteady simulations for iFR prediction, suggesting potential for computational efficiency.