Abstract
Fontan surgical planning involves designing grafts to perform optimized hemodynamic performance for the patient’s long-term health benefit. The uncertainty of post-operative boundary conditions (BC) and graft anastomosis displacements may significantly affect the optimized graft designs and lead to undesired outcomes, especially for hepatic flow distribution (HFD). This study develops a computational framework to automatically optimize patient-specific Fontan grafts with the maximized possibility of keeping post-operative results within clinically acceptable thresholds. Using a two-phased reliability-based robust optimization (RBRO) strategy, we demonstrate up to a 79.2% improvement in HFD reliability compared to traditional deterministic methods.
Introduction
Fontan surgery is a life-saving procedure for children born with single-ventricle heart defects. It involves a complex rerouting of systemic venous blood directly to the lungs. While surgeons use computational fluid dynamics (CFD) to plan these grafts, current models often assume "nominal" conditions: that post-operative blood flow will match pre-operative measurements and that the surgeon will place the graft with sub-millimeter precision.
In reality, post-operative boundary conditions (BC) are uncertain, and "anastomosis displacement"—the slight misalignment of the graft during surgery—is unavoidable. These discrepancies can lead to poor hepatic flow distribution (HFD), where the "hepatic factor" from the liver doesn't reach both lungs equally, potentially causing serious complications like pulmonary arteriovenous malformations (PAVM). This research introduces a planning tool that produces "robust" designs—grafts that perform well even when conditions aren't exactly as predicted.
Methods
The researchers developed a two-phased Reliability-Based Robust Optimization (RBRO) strategy to address these challenges.
Uncertainty and Surrogate Modeling
Uncertainties in blood flow and graft placement were modeled using Gaussian distributions based on clinical literature (e.g., assuming a maximum translational displacement of 5mm and rotational error of 10°). Because full CFD simulations are computationally expensive, the team utilized Gaussian Process Regression to build surrogate models. These "proxies" predict hemodynamic outcomes—HFD, power loss (iPL), and wall shear stress (%WSS)—in seconds rather than hours.
Two-Step Optimization
The optimization follows a two-step process:
- Phase 1 (Deterministic Optimization - DO): Finds the "ideal" design under nominal conditions.
- Phase 2 (Optimization Under Uncertainty - OUU): Uses the DO results as "warm starts" to search for designs that maximize the probability of staying within clinical thresholds despite uncertainty.
Stochastic collocation was employed to efficiently quantify how uncertainties propagate through the model, making the OUU process roughly 37 times more efficient than traditional sampling methods.
Results
The framework was evaluated using retrospective data from four Fontan patients (F1-F4) and two synthetic extreme cases (T1 and T2) representing unbalanced pulmonary artery flow splits.
- Reliability Boost: The RBRO framework improved the reliability of HFD by up to 79.2% compared to standard deterministic optimization ($p < 0.0001$).
- Hemodynamic Safety: Mean values for indexed power loss (iPL < 0.03) and wall shear stress (%WSS < 10%) consistently satisfied clinically acceptable thresholds across all optimized cases.
- Performance in Complex Cases: In cases with highly unbalanced flow (T2), the OUU method significantly outperformed DO, proving its necessity for complex patient anatomies where standard planning often fails.
Discussion and Conclusion
The "best" design on a computer screen is often the most fragile. A design that is mathematically perfect at a specific coordinate might fail if the surgeon is off by a few millimeters. The RBRO approach favors "flatter" optima where performance remains stable across a range of errors.
This study demonstrates the effectiveness of the RBRO framework in addressing the uncertainties of boundary conditions and anastomosis displacements. This planning technique is immediately applicable as a building block to enable technology for optimal long-term outcomes for pediatric Fontan patients and can also be applied to other pediatric and adult cardiac surgeries. Future work will focus on larger cohorts and integrating growth models to account for long-term vascular remodeling as patients transition into adulthood.
Summary updated