Purpose: To facilitate fast and accurate 4D image guidance for lung and upper abdomen radiotherapy, we propose a novel 4D cone beam CT (4D‐CBCT) reconstruction algorithm that offers high‐quality images from highly under‐sampled projections under the conventional 1‐min CBCT scan. Methods: As opposed to conventional reconstruction approaches via image intensity domain, our method integrates patient‐specific CT information and reconstructs 4D‐CBCT image via motion vector domain. It solves an optimization problem, where a deformation vector field is determined, such that, when being used to deform the patient prior CT image, the projections match the measurements. We develop a forward‐backward splitting (FBS) method to solve this problem. Specifically, an intermediate variable is introduced to split the original problem into two iteratively preformed sub‐problems, i.e., reconstruction of the intermediate variable based on projections and the currently reconstructed 4DCBCT image, and deformable registration between the prior image and the intermediate variable. This hybrid algorithm achieves a solution with both right geometrical information defined by measured CBCT projections and correct intensity values from the prior CT, yielding high quality 4D‐CBCT images. Both experiments on a moving ball phantom and patient studies have been performed to validate the proposed method. Results: Even with very limited number of projections from a 1‐min CBCT scan, artifacts‐free 4D‐CBCT images with correct HU values can be obtained by the proposed algorithm in both phantom and patient studies. To validate the anatomical geometry accuracy, the ground‐truth ball center location along the superior‐inferior direction is compared to that measured in the reconstructed 4D‐CBCT. Satisfactory agreements (0.364mm average and 0.504mm maximum error) are observed. Conclusion: A hybrid 4D‐CBCT reconstruction algorithm is developed for highly under‐sampled projections from the conventional 1‐min CBCT scan. It is capable of reconstructing high‐quality images free of under‐sampled streaking with accurate HU values. This work is supported in part by NIH (1R01CA154747‐01), Varian Medical Systems through a Master Research Agreement, the Early Career Award from Thrasher Research Fund, and the University of California Lab Fees Research Program.
ASJC Scopus subject areas
- Radiology Nuclear Medicine and imaging