Towards Patient-Specific Surgical Planning for Bicuspid Aortic Valve Repair: Fully Automated Segmentation of the Aortic Valve in 4D CT

TL;DR

The paper tackles the challenge of planning BAV repair by enabling patient-specific modeling through fully automated segmentation of 4D CT data. It employs nnU-Net to produce multi-label segmentations of the aortic cusps and root, validated against manual ground truth, and derives surgical-relevant metrics such as geometric cusp height, commissural angle, and annulus diameter. The results show clinically usable accuracy with Dice scores near 0.7 and high ICCs for key measurements, though temporal consistency across the cardiac cycle needs improvement. This work demonstrates the translational potential of automated 4D CT segmentation for risk stratification and planning in BAV repair, with future work focused on temporal modeling and dataset expansion.

Abstract

The bicuspid aortic valve (BAV) is the most prevalent congenital heart defect and may require surgery for complications such as stenosis, regurgitation, and aortopathy. BAV repair surgery is effective but challenging due to the heterogeneity of BAV morphology. Multiple imaging modalities can be employed to assist the quantitative assessment of BAVs for surgical planning. Contrast-enhanced 4D computed tomography (CT) produces volumetric temporal sequences with excellent contrast and spatial resolution. Segmentation of the aortic cusps and root in these images is an essential step in creating patient specific models for visualization and quantification. While deep learning-based methods are capable of fully automated segmentation, no BAV-specific model exists. Among valve segmentation studies, there has been limited quantitative assessment of the clinical usability of the segmentation results. In this work, we developed a fully automated multi-label BAV segmentation pipeline based on nnU-Net. The predicted segmentations were used to carry out surgically relevant morphological measurements including geometric cusp height, commissural angle and annulus diameter, and the results were compared against manual segmentation. Automated segmentation achieved average Dice scores of over 0.7 and symmetric mean distance below 0.7 mm for all three aortic cusps and the root wall. Clinically relevant benchmarks showed good consistency between manual and predicted segmentations. Overall, fully automated BAV segmentation of 3D frames in 4D CT can produce clinically usable measurements for surgical risk stratification, but the temporal consistency of segmentations needs to be improved.

Figures (5)

• Figure 1: Anatomy of a BAV with left-right coronary cusp fusion. 3D side view (left), aortic view (center) and 2D views (right) are shown. The structures segmented are the left coronary cusp (LCusp, red), non-coronary cusp (NCusp, green), right coronary cusp (RCusp, blue), root wall (yellow), left ventricular outlet (LVO, orange) and sinotubular junction (STJ, purple)
• Figure 2: Illustration of the semi-automatic ground truth segmentation workflow. For each 4D scan, one 3D diastolic frame and one systolic frame are manually annotated. Within each cardiac phase, deformable transformations are calculated from the manually labeled frame to each of the remaining frames, and the transformations are applied to the manual segmentation to obtain segmentations of all remaining frames.
• Figure 3: Illustration of measurement protocols using the Markup module in 3D Slicer. The left two panels show the valve from an aortic perspective, and the right panel from a ventricular perspective.
• Figure 4: Plots of Dice scores of LCusp (red), NCusp (green), RCusp (blue), root wall (yellow) as a function of frame number. Segmentation results with strong (left) and weaker (right) temporal consistency are shown. The significant drop in Dice scores on the left subplot demonstrates that the deep learning model struggles to accurately segment the transitional frame during a cardiac cycle when the valves are in the process of opening.
• Figure 5: Segmentation accuracy metrics: Dice, mean symmetric mesh distance, and 95th percentile distance for all three cusp components and the aortic root wall. The metrics compare the predicted and ground truth segmentations across all test sets.