Clinical Evaluation of A Novel 4D-Cbct Reconstruction Scheme Based on Simultaneous Motion Estimation and Image Reconstruction
J. Wang,J. Dang,L. Ouyang,X. Gu,T. Pan
DOI: https://doi.org/10.1016/j.ijrobp.2014.05.747
2014-01-01
Abstract:In the present 4D cone-beam CT (4D-CBCT) process, image reconstruction, and motion estimation are performed as two sequential steps. In such a process, motion estimation accuracy is limited by the image quality of reconstructed 4D-CBCT, which is often degraded due to the limited number of projections at each phase. We have recently proposed a novel 4D-CBCT image reconstruction scheme that is able to perform simultaneous motion estimation and image reconstruction (SMEIR). The purpose of this work is to optimize and characterize the performance of the SMEIR through patient evaluation studies. The SMEIR algorithm consists of two steps: (1) motion-compensated image reconstruction; and (2) motion model estimation from projections directly. In the motion-compensated image reconstruction, we utilized the projections from all of the other phases to reconstruct a reference phase 4D-CBCT by explicitly considering the motion model between different phases. In the motion model estimation, we obtained the updated inverse consistent motion model from the projections directly, overcoming the limitation of estimating the motion model from reconstructed images. A lung cancer patient was scanned for 4.5 minutes and total 1679 CBCT projections were acquired. The projections were grouped into 10 phases according to the respiration signal. 4D-CBCT at each phase was reconstructed by total variation (TV) minimization. To evaluate the performance of the SMEIR algorithm on the conventional 1-minute CBCT scan, the projections at each phase were downsampled by different factors ranging from 2 to 10. Four-dimensional-CBCT images were reconstructed from the downsampled projections by using the standard FDK, TV, and SMEIR. Using the 4D-CBCT reconstructed from fully sampled projections, relative error of image and error of tumor motion trajectory were analyzed to quantify the performance of different reconstruction algorithms. The SMEIR algorithm outperforms FDK and TV in both image reconstruction accuracy and tumor tracking trajectory accuracy. When the average number of projections at each phase decreases to 19, relative image reconstruction errors for FDK, TV, and SMEIR are 38.80%, 14.85%, and 10.05%, respectively; the maximum tumor tracking errors for FDK, TV, and SMEIR are 2.77 mm, 1.94 mm, and 0.54 mm, respectively. The patient study results show that the SMEIR algorithm can achieve 1-mm tumor tracking accuracy even when the average number of projections at each phase reduces to 19. These results suggest that the SMEIR algorithm enable the use of conventional 1-minute CBCT for accurate of motion modeling and 4D-CBCT image reconstruction.