Mesh-Based Ultrasound Bladder Volume Estimation for Radiotherapy Enabled By Point-Cloud Completion and Crust Triangulation
Abstract
Purpose
Accurate, frequent bladder volume monitoring is essential in pelvic radiotherapy because inter-fraction filling variability affects target coverage and organ-at-risk sparing. Yet current options are suboptimal: CBCT offers 3D anatomy but increases radiation and workload, while ultrasound bladder scanners use simplified geometric assumptions and have limited accuracy and reproducibility, especially for irregular geometries.
Methods
We propose a radiotherapy-oriented framework for bladder volume assessment. In total, 550 retrospective pelvic CT scans were used to reconstruct 3D bladder models (425/125 train/test). A deterministic multi-view sampling strategy emulating multi-angle ultrasound generated 2,975 meridian-like sections from the training set, which were converted into sparse partial point clouds. EPtrNet was then used to complete the global bladder shape, followed by Crust triangulation to obtain closed surfaces for mesh-based volume estimation. Performance was evaluated by (i) volumetric accuracy versus ground truth using MAE/MAPE, Pearson correlation, and Bland–Altman analysis with a commercial bladder scanner as comparator, and (ii) morphological consistency using the surface area of watertight reconstructions.
Results
The workflow converged stably with consistent geometric accuracy. On the independent test set, completion achieved a Chamfer Distance (L2²) of 393.85 ± 222.27 and a Hausdorff95 of 24.17 ± 7.71, enabling reliable closed-surface reconstruction and mesh-based volume computation. In the clinical validation cohort, estimated volumes closely matched ground truth (MAE 16.06 mL; MAPE 5.35%; r = 0.9907) and outperformed a commercial bladder scanner. Surface area deviation from CT was 5–12%, indicating robust preservation of non-convex bladder geometry relevant to radiotherapy planning.
Conclusion
This point cloud–based, CT-referenced framework enables accurate, geometry-aware bladder volume monitoring without additional ionizing radiation, offering a practical solution for high-frequency assessment and adaptive decision-making in pelvic radiotherapy.