Generative Mask Pyramid Network for CT/CBCT Metal Artifact Reduction with Joint Projection-Sinogram Correction

Abstract

A conventional approach to computed tomography (CT) or cone beam CT (CBCT) metal artifact reduction is to replace the X-ray projection data within the metal trace with synthesized data. However, existing projection or sinogram completion methods cannot always produce anatomically consistent information to fill the metal trace, and thus, when the metallic implant is large, significant secondary artifacts are often introduced. In this work, we propose to replace metal artifact affected regions with anatomically consistent content through joint projection-sinogram correction as well as adversarial learning. To handle the metallic implants of diverse shapes and large sizes, we also propose a novel mask pyramid network that enforces the mask information across the network's encoding layers and a mask fusion loss that reduces early saturation of adversarial training. Our experimental results show that the proposed projection-sinogram correction designs are effective and our method recovers information from the metal traces better than the state-of-the-art methods.

Metal Artifact Reduction In Cone-Beam Extremity Images Using Gated Convolutions

“…However, improvements in the corresponding image domain were not shown, which is a crucial end objective of MAR. The benefit of using metal masks has also been shown by [11] for inpainting in projections and by [12] in images.…”

“…We tested the developed model on a clinical dataset against and our results show improvements in image quality for different shape, size and location of metal. The results were compared against LiMAR [3], U-net [9], Mask pyramid [11], and Partial convolution [12] based methods using the same data. Our contribution is to apply deep neural gated convolutional networks based mask guidance for metal inpainting in CBCT images.…”

Face Completion with Semantic Knowledge and Collaborative Adversarial Learning

“…[24] introduces a 3D inpainting task to improve the performance of ultrasound image segmentation. [23,25] convert the metal artifact reduction problem to the X-ray image inpainting problem and the metal artifacts are removed by reconstructing from the inpainted X-ray images.…”

Usefulness of a Metal Artifact Reduction Algorithm in Digital Tomosynthesis Using a Combination of Hybrid Generative Adversarial Networks

“…The network input would weaken the metal trace information, owing to the down-sampling operations of the network. Therefore, we used the mask pyramid U-Net (mask pyramid network [MPN]) [30] to retain the model trace information at each layer to explicitly enable the network to extract more discriminative features to restore the missing information in the metal trace region. As GAN has been reported to be useful for noise reduction [31], including MAR, it can be useful for image quality improvement (artifact reduction) in DT.…”