Lung 4D-CBCT Reconstruction Using Prior Information and Limited-Angle Projections: Phantom and Patient Studies

Abstract

Purpose/Objective(s): Current application of 4D-CBCT in clinics is limited due to its long scan time and high imaging dose. This study is to develop and clinically evaluate a 4D-CBCT reconstruction method using limited-angle projections to reduce scan time and dose. Materials/Methods: A motion-modeling and free-form deformation (MMFD) method is proposed to reconstruct 4D-CBCT images using only orthogonal limited-angle projections. In this method, each phase of the onboard 4D-CBCT is viewed as a deformation of a prior CT image. To solve the deformation field, the MM-FD method applies a principal component analysis based motion-modeling to generate a coarse estimation of the deformation field, followed by a constrained free-form deformation algorithm to further fine-tune the field. Data acquired from a dynamic thorax physical phantom and three lung cancer patients were used to evaluate the MM-FD method. 4D-CBCT reconstructions from a full sample of projections were used as the ground truth. The accuracy of the MM-FD method was evaluated by calculating the volume percentage difference (VPD) and center-of-mass shift (COMS) of the reconstructed tumor volume. The accuracy of the FDK, MM-only and FD-only methods was also evaluated for comparison. Results: Results showed that MM-FD was substantially more accurate than FDK, MM-only and FD-only methods. For the phantom study using orthogonal 30 projections, the average ( standard deviation) VPD values for MM-only, FD-only and MM-FD methods were 223.3% ( 2.5%), 29.2% ( 24.9%) and 6.6% ( 2.6%), respectively. The corresponding COMS values were 2.0mm ( 1.6mm), 3.5mm ( 2.7mm) and 1.1mm ( 0.4mm), respectively. For the patient study, the average ( standard deviation) VPD values for MM-only, FD-only and MM-FD methods were 36.0% ( 32.2%), 21.4% ( 13.9%) and 9.6% ( 6.1%). The corresponding COMS values were 2.2mm ( 1.5mm), 1.6mm ( 0.6mm) and 1.1mm ( 0.5mm). The MM-FD method yielded similar reconstruction accuracy using different scan directions. The average ( standard deviation) VPD and COMS were 6.6% ( 2.6%) and 1.1mm ( 0.5mm) using orthogonal 30 projections from anterior-posterior and left-lateral directions, and 8.3% ( 2.5%) and 1.3mm ( 0.4mm) using projections from left-anterior-oblique and left-posterior-oblique directions. The MM-FD method was less accurate when using projections of lower sampling frequency. The average ( standard deviation) VPD and COMS were 6.6% ( 2.6%) and 1.1mm ( 0.4mm) using 1/projection, and 23.1% ( 5.3%) and 1.2mm ( 0.4mm) using 4/projection. Conclusions: The MM-FD method can accurately reconstruct 4D-CBCT images using only limited-angle projections and therefore shows great potential to reduce the scan time and dose for 4D-CBCT imaging. Author Disclosure: Y. Zhang: A. Employee; Duke University Medical Center. F. Yin: A. Employee; Duke University Medical Center. E. Research Grant; Varian Medical Systems, National Institute of Health. Q. Patent/License Fee/Copyright; Patent. S. Leadership; AAPM. T. Pan: A. Employee; The University of Texas, MD Anderson Cancer Center. E. Research Grant; Cancer Prevention Research Institute of Texas. I. Vergalasova: A. Employee; Duke University Medical Center. L. Ren: A. Employee; Duke University Medical Center. E. Research Grant; Varian Medical Systems, National Institute of Health.