Deep Learning-Based Scatter-Free Virtual Monochromatic CBCT Imaging for Metal Artifact Reduction: A Feasibility Study
Abstract
Purpose
Metal artifacts are one of the common imaging artifacts that degrade the image quality of cone-beam CT (CBCT). Thus, several studies have investigated metal artifact reduction (MAR) techniques to restore anatomical information around metallic implants, including sinogram inpainting, multi-energy-driven virtual monochromatic imaging, and deep learning. This study aims to synthesize scatter-free virtual monochromatic (VM) CBCT projection data from conventional CBCT projection data using a deep neural network (DNN) to achieve MAR without increasing patient radiation exposure. Materials &
Methods
We collected diagnostic CT images of 10 H&N cancer patients from The Cancer Imaging Archive. Elekta’s XVI CBCT system was modeled using MC-GPU, a GPU-accelerated Monte Carlo code for CBCT simulation, to obtain CBCT projection data under a clinical protocol. We also modeled an 80 keV monochromatic X-ray source and acquired scatter-free CBCT projection data that do not contain 2 major sources of metal artifact: beam-hardening and scattered X-rays. We utilized 6 patients for DNN development and reserved 4 patients for external testing. Subsequently, we employed Swin UNET TRansformers (Swin UNETR) to synthesize artifact-free CBCT projection data from the conventional CBCT projection data. The synthesized results were compared to those obtained via MC simulation in the projection and image domains to validate the MAR performance of the proposed approach.
Results
The mean and standard deviation of the mean absolute error (MAE) and peak signal-to-noise ratio (PSNR) of the synthesized CBCT projection data were 2.19x10-2 ± 7.83x10-3 and 43.93 dB ± 2.62, respectively. The mean and standard deviation of the MAE, PSNR, and structural similarity index measure (SSIM) of the reconstructed images were 10.3 HU ± 2.1, 41.85 dB ± 1.32, and 0.977 ± 0.007, respectively.
Conclusion
We demonstrated the feasibility of an AI to synthesize scatter-free VM CBCT projection data from CBCT projection data acquired under clinical scanning conditions.