Intrinsic Geometry Reconstruction for the Study of Geometric Variations in Linac-Based CBCT System
G Zou*, I Chetty, J Jin, Henry Ford Health System, Detroit, MISU-E-J-5 Sunday 3:00PM - 6:00PM Room: Exhibit Hall
Purpose: To develop a software to derive the intrinsic geometric parameters of a cone-beam-CT (CBCT) system from its projection images, for studying geometric variations in linac-based CBCT.
Methods: The projection images were taken with an IsoCal phantom, which has 16 steel ball-bearings (BBs) with known geometry, using a Varian on-board imaging system. The computational pipeline includes 4 components: 1) extraction of each BB position in the projection image with sub-pixel precision using a progressive thresholding method followed by ellipse contour fitting; 2) computation of a 3-by-4 matrix which transforms each 3D-phantom coordinates of BB into its corresponding 2D-image coordinates; 3) derivation of a complete set of CBCT geometric parameters from the matrix in a phantom coordinate system; 4) distinguishing the effect of phantom setup by defining a new coordinate system intrinsic to the tractories of the moving gantry and minimizing the discrepancy between the transformed and nominal coordinates. The software was used to study the repeatability of CBCT geometric variations in a 2-month period.
Results: The source and imager showed repeatable 0.3-0.5 mm sine-like variation patterns with random noise in both U and V directions during a 360° gantry rotation, consistent with our knowledge that the gantry-head sagging may induce gantry-angle dependent variation. The noise levels for the source-U, source-V, imager-U, imager-V were about 0.3, 0.1, 0.2 and 0.15 mm, respectively, suggesting the detection limits of the method. The source-to-imager axial direction variation for the source and imager showed ~2 mm periodic noise patterns, possibly due to the fact that projection positions of BBs in the imager are not sensitive to the axial variation.
Conclusion: We have developed a software module to derive geometric parameters of the CBCT system. The method has accuracies of 0.1-0.3 mm in the U- and V-directions, but is not as accurate in the axial direction.
Funding Support, Disclosures, and Conflict of Interest: The study is supported by NIH Grant Number 5 R01 CA166948-02.