Paper Arxiv: https://arxiv.org/abs/2501.04140
🔴 Our method is purely data-driven, which does not need any prior motion model or 4DCT.
Abstract:
In image-guided radiotherapy (IGRT), four-dimensional cone-beam computed tomography (4D-CBCT) is critical for assessing tumor motion during a patient’s breathing cycle prior to beam delivery. However, generating 4D-CBCT images with sufficient quality requires significantly more projection images than a standard 3D-CBCT scan, leading to extended scanning times and increased imaging dose to the patient. To address these limitations, there is a strong demand for methods capable of reconstructing high-quality 4D-CBCT images from a 1-minute 3D-CBCT acquisition. The challenge lies in the sparse sampling of projections, which introduces severe streaking artifacts and compromises image quality. This paper introduces a novel framework leveraging spatiotemporal Gaussian representation for 4D-CBCT reconstruction from sparse projections, achieving a balance between streak artifact reduction, dynamic motion preservation, and fine detail restoration. Each Gaussian is characterized by its 3D position, covariance, rotation, and density. Two-dimensional X-ray projection images can be rendered from the Gaussian point cloud representation via X-ray rasterization. The properties of each Gaussian were optimized by minimizing the discrepancy between the measured projections and the rendered X-ray projections. A Gaussian deformation network is jointly trained to deform these Gaussian properties to obtain a 4D Gaussian representation for dynamic CBCT scene modeling. The final 4D-CBCT images are reconstructed by voxelizing the 4D Gaussians, achieving a high-quality representation that preserves both motion dynamics and spatial detail.
Key Results:
1. Clinical Data
The proposed method can recon high temporal (50 phases) CBCT from a 1-min scan from Varian Truebeam. Top one is 10 phases, bottom one is 50 phases.
2. SPARE Data
Only showing one case. Please see assets folder for detailed list of Gifs for 29 reconstructions.
To run the code Follow R2_Gaussians (https://github.com/ruyi-zha/r2_gaussian) to install dependencies.
For your own data
- run init_pcd.py to generate the initial points from CGLS reconstruction (example: https://drive.google.com/file/d/10NPiS79sBjeubEN_UqlX-VD-4nmq81yB/view?usp=sharing).
- As an example, put the downloaded data in ./data/spare_dataset/MC_V_P2_NS_01. Then, run (python.exe train.py -s data/spare_dataset/MC_V_P2_NS_01).
For datasets used, please refer to: https://image-x.sydney.edu.au/spare-challenge/ for details.
Comparison methods
🔴 Not fair comparison, since our method did not rely on any prior information.:
MC-FDK: the motion-compensated FDK [1] implemented by Dr Simon Rit from the CREATIS laboratory. A prior DVF is built from the pre-treatment 4D-C. Using this DVF, a FDK reconstruction is performed but with the backprojected traces deformed to correct for respiratory motion.
MA-ROOSTER: the motion-aware spatial and temporal regularization reconstruction [2] implemented by Dr Cyril Mory from the CREATIS laboratory. The reconstruction is solved iteratively by enforcing spatial smoothness as well as temporal smoothness along a warped trajectory according to the prior DVF built from the pre-treatment 4D-CT.
MoCo: the data-driven motion-compensated method [3] implemented by Dr Matthew Riblett from the Virginia Commonwealth University and Prof Geoffrey Hugo from the Washington University. The motion-compensation DVF is built using groupwise deformable image registration of a preliminary 4D-CBCT reconstruction computed by the PICCS method.
MC-PICCS: the motion-compensated (MC) prior image constrained compressed sensing (PICCS) reconstruction [4] implemented by Dr Chun-Chien Shieh from the University of Sydney. The reconstruction is solved using a modified PICCS algorithm, where the prior image is selected to be the MC-FDK reconstruction.
Prior deforming: the reconstruction is solved by deforming the pre-treatment 4D-CT to match with the CBCT projections [5]. This method was implemented by Dr Yawei Zhang and A/Prof Lei Ren at the Duke University.
[1] S. Rit, J. W. H. Wolthaus, M. van Herk, and J.-J. Sonke, On-the-fly motion-compensated cone-beam CT using an a priori model of the respiratory motion, Med Phys 36, 2283 (2009), ISSN 2473-4209, URL http://dx.doi.org/10.1118/1.3115691.
[2] C. Mory, G. Janssens, and S. Rit, Motion-aware temporal regularization for improved 4D cone-beam computed tomography, Phys Med Biol 61, 6856 (2016), ISSN 0031-9155.
[3] M. J. Riblett, G. E. Christensen, E. Weiss, and G. D. Hugo, Data-driven respiratory motion compensation for four-dimensional cone-beam computed tomography (4D-CBCT) using groupwise deformable registration, Med Phys 45, 4471(2018), https://aapm.onlinelibrary.wiley.com/doi/pdf/10.1002/mp.13133.
[4] C.-C. Shieh, V. Caillet, J. Booth, N. Hardcastle, C. Haddad, T. Eade, and P. Keall, 4D-CBCT reconstruction from a one minute scan, in Engineering and Physical Sciences in Medicine Conference (Hobart, Australia., 2017).
[5] L. Ren, J. Zhang, D. Thongphiew, D. J. Godfrey, Q. J. Wu, S.-M. Zhou, and F.-F. Yin, A novel digital tomosynthesis (DTS) reconstruction method using a deformation field map, Med Phys 35, 3110 (2008), ISSN 0094-2405, URL http://www.ncbi.nlm.nih.gov/pmc/articles/ PMC2809715/.
This work is inspired by: