Determination of Individual Lung Pulmonary Vascular Resistance in Children with Glenn Physiology: A Two-Center Study (RF_FR_417)

Background: Single ventricle hypoplastic left heart disease in children is managed through three procedures at different ages: (1) Stage 1 palliation, (2) Glenn, and (3) Fontan. The goal of these procedures is to separate systemic and pulmonary blood flow to improve circulation. After the Fontan procedure, the superior and inferior vena cava (SVC and IVC) are connected to the pulmonary arteries (PA), creating a total cavopulmonary connection (TCPC) that allows deoxygenated blood to flow passively to the pulmonary circuit. Despite advancements in improving the Fontan conduit, patients continue to experience long-term complications related to the hemodynamics of the Fontan circulation.

As a result, there is a growing interest in using a digital twin to predict the hemodynamics of a given TCPC design, aiming to minimize power loss and balance hepatic flow distribution. To develop these models, the pulmonary vascular resistance (PVR) of each lung should be computed for each patient, where flow data from cardiovascular magnetic resonance (CMR) and pressure data from catheterization (CATH) is readily available as standard of care. In this study, we rigorously evaluated the performance of our in-silico PVR estimation algorithm using datasets from two independent centers.

Methods: A patient cohort (n = 23) with Glenn physiology, who underwent both CMR and CATH exams at two centers, was retrospectively identified for this study (Fig. 1a). For each patient, a 3D model of the Glenn pathway was generated (Fig. 1b). This model was then used in an optimization pipeline that utilizes computational fluid dynamics (CFD) simulations to compute PVR in the left and right lungs by optimizing to CMR flow and CATH pressures. Also, a 0D lumped parameter (LP) model was used to assess whether a simplified model could estimate PVR values like those obtained from the CFD model.

Results: The CFD optimization pipeline resulted in mean flow differences of -0.005 ± 0.03 L/min for the left pulmonary artery (LPA) and 0.01 ± 0.03 L/min for the right pulmonary artery (RPA), relative to CMR measured flow. Mean pressure differences were 0.42 ± 0.93 mmHg and 0.37 ± 0.76 mmHg for the LPA and RPA, respectively. The indexed mean optimized PVR estimates across both centers were 5.80 ± 2.61 WU·m² for the LPA and 5.32 ± 2.67 WU·m² for the RPA. PVR estimates were comparable across the two centers and between LPA/RPA (Fig. 2) and compared well with the clinical PVR computed from CMR and CATH measurements, with a mean difference of 0.28 ± 0.45 WU·m² for the LPA and -0.04 ± 0.38 WU·m² for the RPA (Fig. 3). The simplified 0D LP model produced very similar PVRs to those estimated via the 3D CFD model, with a mean difference of -0.04 ± 0.16 WU·m² and 0.05 ± 0.27WU·m² for the LPA and RPA.

Conclusion: In a two-center study, our optimized pipeline for both CFD and LP models generated PVR estimates that closely aligned with PVR values derived from CMR and CATH data.

Figure 1: a) Patient demographic and flow/pressure data measured at the superior vena cava (SVC), left pulmonary artery (LPA), right pulmonary artery (RPA), and left atrium (LA); b) Computational model representation of the Glenn pathway. The inlet boundary condition is prescribed to be the CMR measured inlet flow (QSVC), while the outlet uses simple resistance boundary conditions. The optimization cost function minimizes the mean squared error between CFD and CATH inlet pressure (PSVC), and between CFD and CMR outlet flows (QLPA and QRPA). The target parameters for optimization are the outlet resistances RLPA and RRPA. The optimization uses the simplex method with a tolerance of 1e-4 to minimize the mean squared error between CMR and CFD outlet flows, and between CATH and CFD inlet pressure.

Figure 2: Comparison between PVRs for clinical, CFD, and LP data, and between the two centers (Center 1: Orange; Center 2: Blue).

Figure 3: Linear regression and Bland-Altman analysis of clinical PVR and estimated PVR using CFD in both lungs, indexed by body surface area (BSA) (Center 1: Orange; Center 2: Blue).