Purpose: Apply skeletonization scheme to extract gold marker locations from 2D planar images and to construct a 3D marker model for automatic 6D registration of prostate SBRT. Further, to evaluate the angle dependency of intrabeam 3D motion monitoring using 2D monoscopic X-ray images.
Methods: 2D paired-planar images of markers were masked and skeletonized to represent the markers as 2D lines and to construct a 3D motion model of the marker position in space. An iterative closest point (ICP) algorithm was utilized to calculate the 6D transformation. The skeletonization scheme was compared at different kV/MV energies and techniques and the 6D registration was validated using a prostate phantom. 36 projected images were acquired over a full gantry rotation and were used to evaluate the range of angle directions and separations that preserves the approximation that two adjacent images can accurately localize the marker in 3D space.
Results: Applying the skeletonization technique, kV and 2.5MV/6MV/16MV images showed similar Skeleton coordinates (differences <0.45mm). Using CBCT registration as ground truth for the 6D iso-center movement, the 2D to 3D registration accuracy of 2D-paired images from 11 observations had medians of 0.1mm, -0.2mm, -0.2mm, 0.1ᵒ, 0.0ᵒ, 0.0ᵒ in vertical, longitudinal, lateral, yaw, pitch, roll directions, respectively. The mean accuracy of 3D motion estimated from two consecutive kV images were <0.3mm for angle separation of 10ᵒ to 160ᵒ. The standard deviations (SD) were less than 0.3mm for angle separations of 20o to 120o. SD could be greater than 0.5mm if angle separation were 10ᴼ, 160ᴼ or 170ᵒ.
Conclusion: The skeletonization scheme is feasible and has an accuracy in 6D comparable to 3D match using CBCT. The 2D line projection from the skeletonized 3D model can be used to estimate the 3D motion during treatment from 2D projected images with an angle separation as small as 10ᵒ.