MGMAR: Metal-Guided Metal Artifact Reduction for X-ray Computed Tomography

Hyoung Suk Park; Kiwan Jeon

eess.IVMar 13, 2026

MGMAR: Metal-Guided Metal Artifact Reduction for X-ray Computed Tomography

Authors: Hyoung Suk Park, Kiwan Jeon

Organizations: National Institute for Mathematical Sciences, Daejeon, 34047, Republic of Korea

Abstract

An X-ray computed tomography (CT), metal artifact reduction (MAR) remains a major challenge because metallic implants violate standard CT forward-model assumptions, producing severe streaking and shadowing artifacts that degrade diagnostic quality. We propose MGMAR, a metal-guided MAR method that explicitly leverages metal-related information throughout the reconstruction pipeline. MGMAR first generates a high-quality prior image by training a conditioned implicit neural representation (INR) using metal-unaffected projections, and then incorporates this prior into a normalized MAR (NMAR) framework for projection completion. To improve robustness under severe metal corruption, we pretrain the encoder-conditioned INR on paired metal-corrupted and artifact-free CT images, thereby embedding data-driven prior knowledge into the INR parameter space. This prior-embedded initialization reduces sensitivity to random initialization and accelerates convergence during measurement-specific refinement. The encoder takes a metal-corrupted reconstruction together with a recursively constructed metal artifact image, enabling the latent field to capture metal-dependent global artifact patterns. After projection completion using the INR prior, we further suppress residual artifacts using a metal-conditioned correction network, where the metal mask modulates intermediate features via adaptive instance normalization to target metal-dependent secondary artifacts while preserving anatomical structures. Experiments on the public AAPM-MAR benchmark demonstrate that MGMAR achieves state-of-the-art performance, attaining an average final score of 0.89 on 29 clinical test cases.

Figures & tables

Figure 1: Schematic diagram of the MGMAR framework, which exploits metal-related information throughout the pipeline via a metal artifact-aware latent-conditioned INR for prior image generation and a metal-embedded residual learning network.

Figure 2: Prior images obtained by an INR with different random initializations. Given the projection data $P$ from a validation case (B02 #916), the INR is optimized by minimizing the naive loss $\mathcal{L}_{\text{\tiny naive}}$ for $10{,}000$ iterations (WW/WL $=1500/-250$ HU for CT images). Although the final loss values are similar, the resulting reconstructions differ substantially.

Figure 3: Constructed metal-artifact images $\mu_{\text{\tiny MA}}^{(k)}$ across recursion steps ( $k=0,1,2$ ). The RMSE with respect to the target $\mu_{\text{\tiny MA}}^{*}$ is reported at the top of each panel (WW/WL $=2000/-500$ ).

Figure 4: Comparison of prior images obtained by INRs initialized with random, meta, and the proposed data-driven weights across the number of iterations $N_{\text{\tiny iter}}$ (WW/WL = 1500/-250 HU for CT images).

Figure 5: Visual comparison of prior, NMAR-corrected, and residual-corrected images on a validation case (B04 #948) (WW/WL = 300/0 HU for CT images, WW/WL = 50/0 HU for difference images).

	RMSE $\downarrow$	PSNR $\uparrow$	SSIM $~\uparrow$	Runtime (s)
Uncorrected	0.0143 $\pm$ 0.0059	37.44 $\pm$ 3.01	0.9307 $\pm$ 0.0295	-
Prior image
$N_{\text{\tiny iter}}=0$	0.0045 $\pm$ 0.0029	47.82 $\pm$ 3.56	0.9857 $\pm$ 0.0161	6
$N_{\text{\tiny iter}}=1000$	0.0042 $\pm$ 0.0024	48.36 $\pm$ 3.52	0.9860 $\pm$ 0.0162	41
NMAR-corrected	0.0037 $\pm$ 0.0017	49.24 $\pm$ 3.43	0.9869 $\pm$ 0.0113	2
Residual-corrected	0.0032 $\pm$ 0.0016	50.74 $\pm$ 3.60	0.9904 $\pm$ 0.0097	1

Table 1: Quantitative results of each stage on the validation set, along with the average runtime per case. The quantitative results are reported as mean

\pm

standard deviation.

Figure 6: Effect of the metal-artifact image $\mu_{\text{\tiny MA}}$ for prior image generation (WW/WL = 1500/-250 HU for CT images, WW/WL = 200/0 HU for difference images).

Figure 7: Effect of metal conditioning in residual learning (WW/WL = 1000/0 HU for CT images, WW/WL = 200/0 HU for difference images).

	RMSE $\downarrow$	PSNR $\uparrow$	SSIM $~\uparrow$
Data-driven initialization
without $\mu_{\text{\tiny MA}}$	0.0053 $\pm$ 0.0031	46.46 $\pm$ 3.52	0.9835 $\pm$ 0.0179
with $\mu_{\text{\tiny MA}}$	0.0045 $\pm$ 0.0029	47.82 $\pm$ 3.56	0.9857 $\pm$ 0.0161
Residual correction
without $\chi_{D}$	0.0033 $\pm$ 0.0017	50.67 $\pm$ 3.64	0.9903 $\pm$ 0.0102
with $\chi_{D}$	0.0032 $\pm$ 0.0016	50.74 $\pm$ 3.60	0.9904 $\pm$ 0.0097

Table 2: Ablation study on the effects of the metal-artifact image

\mu_{\text{\tiny MA}}

in prior image generation and metal conditioning in the residual learning stage. The quantitative results are reported on the validation set.

Ranking	1	2	3	NMAR	Proposed
Final score	0.96	0.98	0.99	1.82	0.89
CTN	0.74	0.81	0.82	2.03	0.72
Noise	0.01	0.19	0.01	1.06	0.01
Sharpness	0.72	0.52	0.54	0.58	0.73
Streak	1.29	1.65	1.48	1.99	1.33
SSIM	0.75	0.95	0.98	2.03	0.72

Table 3: Final benchmark score and average values of each evaluation metric over all 29 test cases for the selected methods ranked first, second, and third in the AAPM-MAR challenge, traditional NMAR, and MGMAR.

Figure 8: Visual comparison of corrected images obtained by MGMAR and an earlier version of our method (ranked second in the AAPM-MAR challenge) for head (first row), thorax (second row), and pelvis (third row) test cases (WW/WL $=200/0$ HU).

Explore similar work

GraphMAR: Geometry-Aware Graph Learning Framework for Spatially Adaptive CT Metal Artifact Reduction
May 17, 2026Zilong Li, Chenglong Ma, Yiming Lei +7Graph Neural Networks Image Restoration
Shanghai Key Lab of Intelligent Information Processing, College of Computer Science and Artificial Intelligence, Fudan University, Shanghai 200433, China · Institute of Science and Technology for Brain-inspired Intelligence, Fudan University, Shanghai 200433, China · College of Computer Science and Technology, Qingdao University, Qingdao 266071, China +2
SCMA: Structure-Conditioned and Metal-Aware Flow Matching for CT Metal Artifact Reduction
Jul 30, 2026Heran Wang, Jianing Sun, Xu Jiang +3Flow Matching Image Reconstruction
School of Mathematical Sciences, Capital Normal University, Beijing 100048, China · College of Mathematics Science, Inner Mongolia Normal University, Hohhot 010022, China · National Center for Applied Mathematics Beijing, Capital Normal University, and the Academy for Multidisciplinary Studies, Capital Normal University, Beijing 100048, China
Balancing Efficiency and Restoration: Lightweight Mamba-Based Model for CT Metal Artifact Reduction
Apr 8, 2026Weikai Qu, Sijun Liang, Xianfeng Li +7Image Restoration Computed Tomography
New York University, America · BaiyunPort, China · University of Technology Sydney, Australia +6

May 17, 2026J/K move · Enter open · S save

GraphMAR: Geometry-Aware Graph Learning Framework for Spatially Adaptive CT Metal Artifact Reduction

Zilong Li, Chenglong Ma, Yiming Lei, Yuanlin Li+6

Shanghai Key Lab of Intelligent Information Processing, College of Computer Science and Artificial Intelligence, Fudan University, Shanghai 200433, China · Institute of Science and Technology for Brain-inspired Intelligence, Fudan University, Shanghai 200433, China · College of Computer Science and Technology, Qingdao University, Qingdao 266071, China+2

Graph Neural Networks Image Restoration Computed Tomography

Open paper AskPDF arXiv

Computed tomography (CT) metal artifact reduction (MAR) aims to reduce the severe streaking artifacts induced by metallic implants and other high-density objects. Effective MAR generally requires both accurate artifact localization and artifact removal. Sinogram-domain methods can exploit explicit geometric cues, such as metal traces, to identify metal-corrupted measurements, while requiring raw projection data, which is often unavailable in clinical and practical scenarios. Image-domain methods are more flexible and widely applicable, yet they usually lack comparable geometric guidance, limiting their ability to localize artifacts and leading to suboptimal results. To address this limitation, we propose GraphMAR, a geometry-aware learning framework for explicit artifact identification and spatially adaptive MAR in the image domain. The key idea is to introduce graph-based geometric modeling as an image-domain analogue of sinogram metal traces. Specifically, we first construct a geometric graph from the metal mask and derive a geometric density graph that coarsely localizes artifact-prone regions according to inter-implant geometry. We then design GraphMoE, a graph-routed mixture-of-experts module that builds a polar-coordinate artifact graph in feature space and adaptively routes different experts to different spatial regions for MAR. By aligning the learned routing maps with the geometric density graph, GraphMAR provides explicit and interpretable artifact localization while enabling region-adaptive artifact reduction. Experiments on both simulated and real-world datasets demonstrate that GraphMAR achieves superior MAR performance compared with existing methods. To the best of our knowledge, this is the first work to introduce graph-based modeling for CT MAR and to enable explicit artifact identification in the image domain, improving both restoration quality and interpretability.

Abstract

Explore similar work

GraphMAR: Geometry-Aware Graph Learning Framework for Spatially Adaptive CT Metal Artifact Reduction

SCMA: Structure-Conditioned and Metal-Aware Flow Matching for CT Metal Artifact Reduction

Balancing Efficiency and Restoration: Lightweight Mamba-Based Model for CT Metal Artifact Reduction