Author: Weixing Cai, Laura I. Cervino, Yabo Fu, Licheng Kuo, Tianfang Li, Xiang Li, Jean M. Moran, Huiqiao Xie, Hao Zhang 👨🔬
Affiliation: Department of Medical Physics, Memorial Sloan Kettering Cancer Center 🌍
Purpose: This study introduces a novel spatiotemporal Gaussian neural representation framework to reconstruct high-temporal dynamic CBCT images from 1-minute acquisition, preserving motion dynamics and fine spatial details without relying on prior motion models.
Methods: Dynamic CBCT reconstruction was performed on 29 Monte-Carlo (MC) simulated cases and one clinical case. For the clinical scan, projections from a 1-minute Varian TrueBeam acquisition were sorted into 50 phases, with approximately 18 projections per phase. Our framework employs a differentiable 4D Gaussian representation initialized from average CBCT images. Gaussian points are characterized by position, covariance, rotation, and density, offering a compact and dynamic model for CBCT scenes. A Gaussian deformation network, incorporating a HexPlane encoder and multi-head decoder, predicts Gaussian deformations, optimized to minimize L1 and structural similarity index measure (SSIM) losses between rendered and measured projections. Adaptive Gaussian control refines the representation by pruning underutilized Gaussians and densifying points in high-gradient regions. The method was benchmarked on the AAPM SPARE challenge datasets, and further validated with clinical CBCT scans from a Varian TrueBeam system to ensure clinical relevance. The primary objectives were to evaluate target alignment and target motion. Therefore, each phase of the reconstructed CBCT was registered to the ground truth using the Elastix package, focusing on pixels within the planning target volume (PTV).
Results: On the MC simulation datasets, our method achieved translational and rotational errors of 0.54 mm (LR), 0.76 mm (SI), 1.36 mm (AP), 0.55° (rAP), and 0.93° (rSI), and 1.31° (rLR) respectively for PTV alignment. From a 1-min Varian TrueBeam scan, the 50-phase dynamic CBCT was successfully reconstructed, demonstrating effective streak artifact suppression, respiratory motion preservation, and fine detail restoration.
Conclusion: The spatiotemporal Gaussian framework overcomes the limitations of sparse projection sampling, offering high-temporal dynamic CBCT reconstructions without dependence on prior motion models.