Premium
TH‐EF‐BRB‐08: An Efficient and Continuous Surface Reconstruction Method On Point Cloud Captured by a 3D Surface Imaging System
Author(s) -
Liu W,
Cheung Y,
Sawant A,
Ruan D
Publication year - 2015
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4926306
Subject(s) - surface reconstruction , point cloud , robustness (evolution) , computer vision , curvature , standard deviation , artificial intelligence , computer science , iterative reconstruction , surface (topology) , photogrammetry , algorithm , mathematics , geometry , biochemistry , chemistry , statistics , gene
Purpose: Surface photogrammetry‐based systems (e.g., VisionRT, C‐Rad) are gaining increasing acceptance for patient localization and motion monitoring in radiotherapy. The purpose of this work is to accurately and efficiently reconstruct a continuous surface from noisy point clouds captured by a VisionRT system with high‐speed‐image‐capture capability. In contrast to the conventional approach of explicit/discrete meshing‐type scheme that is often error‐prone, our method yields a continuous representation of the underlying patient surface, which is particularly advantageous for subsequent surface registration and motion tracking. Methods: We developed a level‐set based surface reconstruction method by optimizing a regularized fitting energy from the noisy point cloud from VisionRT. This approach offers additional robustness to noise and missing measurements caused by view occlusion. We applied an efficient narrowband evolving scheme and significantly reduced the time‐complexity by one order of magnitude. Validation was performed on surface data from a lung cancer patient during CT‐simulation by comparing the local geometry of our reconstructed surface against that of the skin surface from CT, within comparable ROIs in the chest area based on mean curvature distributions. Results: Our method successfully reconstructed a continuous and smooth surface, with the missing region reasonably filled. We obtained two similar mean curvature distributions from both reconstructed surface and CT surface, with empirical mean and standard deviation of (µ = −2.7E‐3mm −1 , σ = 7.0E‐3mm −1 ) and (µ = −2.5E‐3mm −1 , σ = 5.3E‐3 mm −1 ), respectively. The similarity of those two distributions demonstrated the ability of our method in preserving local geometry of the reconstructed surface. Conclusion: We have proposed an efficient and accurate surface reconstruction method on point cloud from VisionRT. The proposed method has generated a continuous representation of the underlying patient surface with good robustness against noise and missing measurement. It serves as an important first step for further development of motion tracking methods during radiotherapy. NIH 5R01CA169102‐02