Room: Karl Dean Ballroom C
Purpose: Recently, we developed a novel two-dimensional antiscatter grid (2DASG) to reduce x-ray scatter in CBCT, and consequently improve the reconstructed image quality. However, the 2DASGâ€™s septal shadows may produce grid-line-artifacts (GLA) in CBCT projections, and lead to ring artifacts in the reconstructed images. In this work, we develop and present a novel ring suppression method based on an adaptive total variation minimization (a-TVM) principle.
Methods: We have developed and fabricated 2DASG prototypes with grid ratios of 8, 12, and 16, and pitch of 3mm and 2mm using an additive manufacturing method known as Powder Bed Laser Melting. We acquire full revolution CBCT scans of various phantoms, and use the FDK algorithm to reconstruct CBCT images. We apply our method in the projection domain at the GLA locations, and do not apply any other form of correction to any of the images. Our method uses a nonlinear a-TVM formulation with a gradient-based regularization term. The developed method is fully adaptive, and is applied iteratively to each projection until GLA are sufficiently suppressed. Performance of the method is assessed by visual inspection of the reconstructed phantom images, and by evaluating standard deviation of pixels in the uniformity material section of the Catphan phantom. Contrast-to-noise ratio is calculated in the material inserts section, and qualitative evaluation of impact on the spatial resolution is performed in the line pair high-resolution section in the Catphan phantom.
Results: When using our method, the standard deviation was reduced by 52% - 73% in the uniform material section of the phantom, and contrast-to-noise ratio in material inserts was increased by 17.4% - 105.2%. Impact on spatial resolution was minimal, and high-resolution features appear to be well preserved.
Conclusion: We have developed an automated method for suppressing ring artifacts, while preserving high spatial resolution features, in CBCT images.
Funding Support, Disclosures, and Conflict of Interest: This project is supported in part by NIH/NCI R21CA198462. Tesla K40 GPU was provided by NVIDIA Corporation. CBCT electron density phantom was provided by Gammex Inc.
Signal Processing, Pattern Recognition, Numerical Analysis