A Path Predictive Algorithm for Metal-Artifact Reduction In Cone-Beam CT In Prostate Cases without Thresholding.
Abstract
Purpose
Metal artifacts caused by implanted gold fiducial markers remain a significant challenge in cone-beam CT (CBCT) imaging for prostate radiotherapy. Their high attenuation generates severe streaks and image distortions that decrease overall image quality. This project aims to develop an algorithm that predicts the path of the fiducials in projection space for metal-artifact reduction (MAR).
Methods
10 prostate cancer(PCa) patients treated with volumetric modulated arc therapy were included in this study. The MAR workflow consisted of seed masking, Laplacian-inpainting, and reconstruction. TIGRE, an open source CBCT package in MATLAB was used to manipulate the CBCT projection data. This project derives and applies a closed-form, two-view triangulation solution to estimate fiducial trajectories and models their resulting elliptical paths in CBCT projection space. Fifty gantry angles were selected randomly to evaluate the accuracy of the predictive model against the manually selected position. The predicted trajectories were used to guide projection-space masking and Laplacian-inpainting for metal-artifact reduction without needing to threshold.
Results
The mean ± standard deviation of the distance between the predicted and manually selected fiducial positions was 2.3273 ± 1.288 mm. Artifact removal performed robustly in the reconstructed images, yielding visually and quantitatively improved reconstructions. The artifact-reduced images demonstrated favorable structural fidelity, while high-resolution accuracy was maintained in fiducial tracking across all gantry angles.
Conclusion
This study demonstrates that closed-form two-view triangulation enables accurate, threshold-free prediction of fiducial trajectories in CBCT projection space, providing an effective foundation for metal-artifact reduction. By leveraging analytically derived fiducial paths to guide projection-space masking and Laplacian inpainting, the proposed workflow achieves robust artifact suppression while preserving high-resolution tracking accuracy across gantry angles. This approach reduces algorithmic complexity and computational burden compared with conventional threshold-based methods, and offers a scalable framework for future MAR development in prostate radiotherapy CBCT imaging.