A continuous surface reconstruction method on point cloud captured from a 3D surface photogrammetry system.

Area of Science:

• Medical Imaging
• Computer Vision
• Computational Geometry

Background:

• Surface photogrammetry systems like VisionRT capture 3D patient data but often yield noisy point clouds with missing measurements.
• Accurate surface reconstruction is crucial for applications such as medical registration and motion tracking in radiotherapy.

Purpose of the Study:

• To develop and evaluate a robust level-set based method for continuous surface reconstruction from noisy point clouds acquired by the VisionRT system.
• To improve the accuracy and efficiency of surface reconstruction for patient data in medical applications.

Main Methods:

• A level-set based approach was developed to reconstruct an implicit, continuous surface representation by optimizing a regularized fitting energy.
• The method incorporates robustness to noise and missing data, addressing limitations of discrete meshing techniques.
• An efficient narrowband evolving scheme was employed for solving the reconstruction problem.

Main Results:

• The method achieved submillimeter root mean squared error (RMSE) on phantom data with varying noise and missing data levels.
• Reconstructed surfaces from clinical point clouds accurately represented patient anatomy, successfully filling in missing data regions.
• Local geometry analysis showed strong agreement between reconstructed surfaces and CT-derived surfaces, validating the method's fidelity.

Conclusions:

• An accurate and robust level-set based continuous surface reconstruction method was successfully developed for 3D photogrammetry point clouds.
• The method provides a continuous surface representation that is resilient to noise and missing measurements.
• This work represents a significant advancement for subsequent developments in radiotherapy motion tracking.