Validation and Diagnostic Performance of a CFD-Based Non-invasive Method for the Diagnosis of Aortic Coarctation

Abstract

Purpose: The clinical diagnosis of aorta coarctation (CoA) constitutes a challenge, which is usually tackled by applying the peak systolic pressure gradient (PSPG) method. Recent advances in computational fluid dynamics (CFD) have suggested that multi-detector computed tomography angiography (MDCTA)-based CFD can serve as a non-invasive PSPG measurement. The aim of this study was to validate a new CFD method that does not require any medical examination data other than MDCTA images for the diagnosis of CoA. Materials and methods: Our study included 65 pediatric patients (38 with CoA, and 27 without CoA). All patients underwent cardiac catheterization to confirm if they were suffering from CoA or any other congenital heart disease (CHD). A series of boundary conditions were specified and the simulated results were combined to obtain a stenosis pressure-flow curve. Subsequently, we built a prediction model and evaluated its predictive performance by considering the AUC of the ROC by 5-fold cross-validation. Results: The proposed MDCTA-based CFD method exhibited a good predictive performance in both the training and test sets (average AUC: 0.948 vs. 0.958; average accuracies: 0.881 vs. 0.877). It also had a higher predictive accuracy compared with the non-invasive criteria presented in the European Society of Cardiology (ESC) guidelines (average accuracies: 0.877 vs. 0.539). Conclusion: The new non-invasive CFD-based method presented in this work is a promising approach for the accurate diagnosis of CoA, and will likely benefit clinical decision-making.

Keywords: aortic coarctation; congenital heart disease; hydrodynamics; multidetector computed tomography angiography; non-invasive assessment.

Figure 1

Boundary conditions. Geometry of the aorta with one inlet and four outlet boundaries.

Figure 2

Scheme of the LPM. Outlet boundary condition: a lumped parameter model with only one resistance is coupled to each outlet.

Figure 3

Cross-validation pathway. Five equally-sized groups were stratified so to have approximately the same proportion of genders, ages, and patients with CoA. One of them (20% of the data) was holdout for testing, while the others (80% of the data) were used as the training set. To estimate the CFD performance, we applied a 5-fold cross-validation procedure on all groups: each time, the CFD simulation was performed on a different training set. The parameters f and s, obtained from the simulation results, were used to build a prediction equation by using logistic regression, before testing the prediction model on the unseen test set.

Figure 4

Rank correlation matrix among PSPG, f, and s. PSPG, f, and s were all positively correlated (p < 0.05). Lighter shades of blue correspond to lower correlation coefficients that the darker shades. PSPG, peak systolic pressure gradient; f, viscous friction in Equation 1; s, expansion loss in Equation 1.

Figure 5

CFD performance for the diagnosis of CoA. (A) ROC for CoA diagnosis in the training set. (B) ROC for CoA diagnosis in the test set.

Figure 6

Comparison between the diagnostic capacity of the CFD method and of the ESC guidelines criteria. Sensitivities (A), specificities (B), and accuracies (C) of CoA diagnoses conducted by the CFD (gray-blue bars) and the narrowing rate (pink bars) methods. The latter method was applied according to current guidelines.