Knowledge-based self-calibration method of calibration phantom by and for accurate robot-based CT imaging systems

Abstract Modern X-ray imaging system with feature of low-dose and high spatial resolution imaging has become a prevalent noninvasive imaging technique in clinic. High-resolution imaging of cone-beam computed tomography (CBCT) system crucially relies on accurate geometric calibration. Accurate coordinates of bearing balls (BBs) on calibration phantom are essential prerequisites to precisely realize geometric calibration of CBCT systems. However, due to phantom-making error, BB position deviations may occur between the actual value and theoretical value, further resulting in estimation error of realistic parameters in geometric calibration. Current algorithms that can estimate BB coordinates are insufficient and not robust. In this study, we proposed a knowledge-based self-calibration method by means of a robot-based CT system to fill the vacancy. The moving knowledge from one single manipulator of dual manipulators is utilized for automatic self-calibration of the calibration phantom, which is then for accurate geometry calibration for the dual manipulator-based imaging system. Specifically, different projections of BBs can be obtained by moving the robotic manipulator, on which X-ray tube is installed several times, while the detector and the calibration phantom remain stationary. Hence, the actual BB coordinates can be estimated by minimizing two separate cost functions determined by reprojection errors and relocation errors of BBs. Simulation study shows that the mean of absolute distance values of the estimated coordinates are 0, 0.0164, and 0.0190 mm, and the standard deviation values of Gaussian noise are 0%, 10%, and 15% of a single pixel size on BB projection, respectively. Experiments demonstrate that the image quality of the reconstructed results has been improved to a large extent after utilizing the proposed self-calibration method. The performance is even superior to that of images acquired by performing geometric calibration on a high-precise calibration phantom with more BBs. The proposed method provides a flexible utility value to acquire high precision geometry and generate high-quality images for CBCT systems.