Semi-Automated Generation and Hemodynamic Assessment of Surgical Baffle Geometry for Biventricular Repair

1. Problem Statement

Clinical Context:
Double Outlet Right Ventricle (DORV) is a complex congenital heart defect where both the aorta and pulmonary artery arise from the right ventricle (RV). Successful repair often requires constructing an intraventricular "baffle" (a tunnel) to redirect oxygenated blood from the left ventricle (LV) through a Ventricular Septal Defect (VSD) to the aorta, while preserving the native RV geometry.

Current Limitations:

• Manual Intraoperative Design: Currently, baffles are designed and shaped manually by surgeons during the operation on an unpressurized, arrested heart.
• Lack of Preoperative Planning: There is no standardized method to generate patient-specific, watertight 3D models preoperatively to quantitatively assess hemodynamic performance (e.g., pressure gradients) before surgery.
• Anatomical Heterogeneity: The variability in DORV anatomy makes it difficult to create generic solutions, leading to high rates of reoperation and suboptimal flow pathways.
• Modeling Gaps: Existing computational methods often rely on idealized geometries or manual patch construction that fails to integrate seamlessly with native anatomy for Computational Fluid Dynamics (CFD) analysis.

2. Methodology

The authors propose a semi-automated computational pipeline that transforms preoperative CT imaging into a simulation-ready model with an optimized baffle geometry. The workflow consists of four main stages:

A. Medical Imaging and Segmentation

• Input: Preoperative systolic-phase Cardiac Computed Tomography Angiography (CTA).
• Segmentation: A deep learning approach using nnU-Net (trained on the ImageCHD dataset) automatically segments seven anatomical structures: LV, RV, LA, RA, myocardium, aorta, and pulmonary artery.
• Refinement: Results are manually refined in 3D Slicer by a pediatric cardiologist to ensure accuracy at critical interfaces (VSD and aortic valve).

B. Computational Domain Preparation

• Unified Mesh: The segmented structures are merged into a single manifold mesh representing the continuous blood pool ($M_{AoBiV}$).
• Boundary Definition: The mesh is truncated at the VSD and aortic valve to create a "capped" domain ($M_{cappedRV}$) containing the native RV wall, a VSD inlet cap, and an aortic outlet cap.
• Surgeon-Guided Boundary: A congenital cardiac surgeon interactively defines the baffle attachment suture line on the ventricular surface using 12–15 discrete points in 3D Slicer. This sparse set is converted into a dense, smooth 3D polyline ($B_{baffle}$) using principal component analysis and spline interpolation.

C. Patient-Specific Baffle Surface Construction

This is the core algorithmic contribution, designed to create a tunnel that meets specific area constraints while remaining smooth:

1. Target Area Calculation ($A_{target}$): The target cross-sectional area is calculated based on the patient's Body Surface Area (BSA) and normative pediatric aortic annulus z-scores to ensure physiologic sizing.
2. Custom Centerline Construction: Instead of standard medial-axis algorithms (like VMTK), which fail in wide, non-tubular ventricular cavities, the authors use a Piecewise Cubic Hermite Interpolating Polynomial (PCHIP). This enforces a smooth path with tangents aligned to the inlet and outlet normals.
3. Cross-Sectional Optimization:
   • The domain is sliced perpendicular to the centerline.
   • The "baffle segment" of each cross-section is reshaped using a parabolic displacement profile to meet the $A_{target}$ constraint, while the native RV wall segment remains fixed.
   • Convexity Enforcement: To prevent geometric irregularities (sharp inward curves) that would be mechanically impossible for surgical patches (pericardium), the authors apply an upper convex envelope smoothing algorithm to the deformation amplitudes. This ensures a globally smooth, convex tunnel profile.
4. Surface Generation: A Thin-Plate Spline (TPS) interpolation is used to generate the final 3D baffle surface ($S_{baffle}$) that connects the surgeon-defined boundary to the smoothed cross-sectional loops. This ensures $C^2$ continuity and minimal bending energy.

D. Hemodynamic Simulation

• CFD Setup: Steady-state Navier-Stokes simulations are performed using SimVascular.
• Boundary Conditions: Patient-specific inflow rates are derived from echocardiograms. A parabolic velocity profile is applied at the VSD inlet, and a resistance boundary condition is applied at the aortic outlet.
• Analysis: The primary metric is the pressure drop ($\Delta P$) across the reconstructed outflow tract.

3. Key Contributions

1. Semi-Automated Framework: A novel pipeline that bridges the gap between manual surgical planning and quantitative CFD, reducing the time to generate patient-specific models.
2. Physiologically Constrained Geometry: The method uniquely enforces cross-sectional area targets along the tunnel while preserving the native RV geometry, ensuring the baffle is not just a shape filler but a functional flow conduit.
3. Robust Surface Generation: The combination of PCHIP centerlines, convexity-enforced amplitude smoothing, and TPS interpolation produces watertight, non-self-intersecting meshes suitable for high-fidelity simulation.
4. Validation against Clinical Data: The framework was retrospectively validated against four real-world DORV cases, comparing virtual predictions with post-operative echocardiography.

4. Results

The framework was applied to four DORV patients who had previously undergone biventricular repair.

• Geometric Success: The pipeline successfully generated smooth, watertight, and topologically consistent baffle geometries for all four cases.
• Hemodynamic Performance:
  • Simulated pressure drops ($\Delta P_{sim}$) across the virtual baffles ranged from 1.89 to 5.42 mmHg.
  • These values were significantly lower than the actual post-operative pressure gradients measured via echocardiography ($\Delta P_{echo}$), which ranged from 4.93 to 12.67 mmHg.
  • The virtual designs consistently predicted lower resistance than the surgically constructed baffles, suggesting the algorithm can identify more efficient flow pathways than manual intraoperative shaping.
• Flow Patterns: Streamline visualizations showed coherent, unidirectional flow from the VSD to the aorta with no large-scale flow reversal or separation.

5. Significance and Future Work

• Clinical Impact: This tool represents a step toward computational decision support for complex congenital heart surgery. It allows surgeons to virtually test and optimize baffle geometries preoperatively, potentially reducing reoperation rates and improving long-term hemodynamic outcomes.
• Scalability: While tested on DORV, the methodology is applicable to other complex intraventricular flow rerouting scenarios, including "borderline" single-ventricle patients being considered for biventricular conversion.
• Future Directions:
  • Automation: Moving from surgeon-defined boundaries to fully automated boundary selection based on hemodynamic optimization.
  • Material Modeling: Incorporating structural mechanics to account for the stiffness of pericardial patches and suture-line constraints, which currently affect the final implanted shape.
  • Dynamic Simulation: Transitioning from steady-state rigid-wall models to Fluid-Structure Interaction (FSI) or coupled electromechanical models to capture the effects of ventricular contraction on flow.

In conclusion, this paper demonstrates that a semi-automated, physics-informed computational approach can generate superior, patient-specific baffle geometries for biventricular repair, offering a quantitative basis for preoperative surgical planning that outperforms current manual methods.