Computational Modeling of Cardiovascular Hemodynamics

Abstract

Patient-specific computational fluid dynamics (CFD) modelling of cardiovascular haemodynamics enables non-invasive assessment of wall shear stress (WSS), oscillatory shear index (OSI), and fractional flow reserve (FFR) -- parameters inaccessible by imaging alone that are critical for atherosclerosis risk stratification and interventional planning. This study developed and validated a semi-automated CFD pipeline integrating CT angiography (CTA)-derived patient-specific vascular geometries, pulsatile boundary conditions from phase-contrast MRI, and three turbulence modelling approaches-- laminar Navier-Stokes, k-omega SST RANS, and large eddy simulation (LES) -- for coronary artery and carotid bifurcation haemodynamics. The pipeline was validated against 4D flow MRI velocity measurements in 40 patients (20 coronary, 20 carotid) and invasive FFR wire measurements in the coronary cohort. CFD-derived FFR (FFR-CT) demonstrated excellent diagnostic accuracy against invasive FFR (AUC = 0.924; sensitivity 92.4%; specificity 88.6% at FFR threshold 0.80), outperforming CTA stenosis assessment alone (AUC = 0.764). For carotid bifurcation, LES captured transitional flow patterns (vortex shedding frequency 4.2 +- 0.8 Hz) missed by RANS models, with time-averaged WSS within 8.4% of 4D flow MRI measurements. A physics-informed neural network (PINN) surrogate model trained on 500 CFD simulations achieved 94.2% FFR-CT prediction accuracy with 1,200x speedup (12 s vs 4 h per case), enabling real-time clinical deployment. These findings establish the validated CFD pipeline and PINN surrogate as clinically actionable tools for non-invasive haemodynamic assessment.