Author: Alan Rui Li, Qihui Lyu, Dan Ruan, Ke Sheng 👨🔬
Affiliation: Department of Radiation Oncology, University of California, Los Angeles, University of California San Francisco, Department of Radiation Oncology, University of California, San Francisco 🌍
Purpose:
The Sparse Primary Sampling (SPS) grid was shown in a previous computational study to improve image quality by correcting scatter-induced effects and artifacts in Cone-beam Computed Tomography (CBCT) reconstruction via a head phantom. In this study, we demonstrate the speed and performance of the SPS grid CBCT scatter correction technique with a previous computational-based CBCT scatter correction technique utilizing a low-count Monte Carlo simulation smoothed with 3-D Richardson-Lucy Denoising.
Methods:
All projections are obtained using MC-GPU with a custom CT phantom and simulated DICOM converted Head and Neck phantom. The two tested scatter correction techniques utilized the same geometry. The SPS grid technique was run with 5e1011 particles per projection while the low-count MC technique was run at 5e107 particles projections. For the SPS grid technique, scatter rejecting smart pixels are placed at a pixel density of 0.0756%. The SPS reconstruction problem is formulated as a constrained optimization problem with Total Variation regularization and scatter distribution using the L2 difference norm regularization. The low-count MC projections are denoised utilizing 3-D RL fitting and reconstructed via a standard image reconstruction problem utilizing Total Variation regularization. Both optimization problems were solved using FISTA. HU accuracy, CNR, in various ROI were compared among scatter-free, uncorrected total signal, SPS-reconstructed, and 3D-RL images.
Results:
The SPS grid technique achieves comparable image quality to the 3-DRL technique, reducing ROI RMSE from scatter by 99.3% compared to 98.5% for 3-DRL. The SPS grid technique shows faster computation speed with a nearly ~3x reduction in computation time over the 3-DRL method (SPSCT,HN =[ 222.59 s, 175.42 s], 3-DRLCT,HN = [671 s, 597 s]).
Conclusion:
This study demonstrates the novel SPS grid scatter correction technique has comparable image quality improvement and superior computational speed than the previous low-count MC-based scatter reduction via 3-D RL denoising.