Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Unlocking the Potential of Early Epochs: Uncertainty-aware CT Metal Artifact Reduction

Xinquan Yang, Guanqun Zhou, Wei Sun, Youjian Zhang, Zhongya Wang, Jiahui He, Zhicheng Zhang

TL;DR

The paper addresses metal artifact reduction in CT by introducing an uncertainty constraint loss (UC loss) that leverages an uncertainty image derived from restoration results using initial training weights. This uncertainty-guided weighting focuses the MAR network on high-frequency, artifact-prone regions and is designed to be plug-and-play across MAR architectures. Across synthetic (DeepLesion-like) and clinical (CLINIC-metal) datasets, UC loss yields notable PSNR and SSIM gains, especially for image-domain networks, and achieves clearer artifact suppression and texture restoration in clinical images. The approach has practical significance for improving diagnostic quality in CT scans with metallic implants, without requiring changes to existing MAR models.

Abstract

In computed tomography (CT), the presence of metallic implants in patients often leads to disruptive artifacts in the reconstructed images, hindering accurate diagnosis. Recently, a large amount of supervised deep learning-based approaches have been proposed for metal artifact reduction (MAR). However, these methods neglect the influence of initial training weights. In this paper, we have discovered that the uncertainty image computed from the restoration result of initial training weights can effectively highlight high-frequency regions, including metal artifacts. This observation can be leveraged to assist the MAR network in removing metal artifacts. Therefore, we propose an uncertainty constraint (UC) loss that utilizes the uncertainty image as an adaptive weight to guide the MAR network to focus on the metal artifact region, leading to improved restoration. The proposed UC loss is designed to be a plug-and-play method, compatible with any MAR framework, and easily adoptable. To validate the effectiveness of the UC loss, we conduct extensive experiments on the public available Deeplesion and CLINIC-metal dataset. Experimental results demonstrate that the UC loss further optimizes the network training process and significantly improves the removal of metal artifacts.

Unlocking the Potential of Early Epochs: Uncertainty-aware CT Metal Artifact Reduction

TL;DR

The paper addresses metal artifact reduction in CT by introducing an uncertainty constraint loss (UC loss) that leverages an uncertainty image derived from restoration results using initial training weights. This uncertainty-guided weighting focuses the MAR network on high-frequency, artifact-prone regions and is designed to be plug-and-play across MAR architectures. Across synthetic (DeepLesion-like) and clinical (CLINIC-metal) datasets, UC loss yields notable PSNR and SSIM gains, especially for image-domain networks, and achieves clearer artifact suppression and texture restoration in clinical images. The approach has practical significance for improving diagnostic quality in CT scans with metallic implants, without requiring changes to existing MAR models.

Abstract

In computed tomography (CT), the presence of metallic implants in patients often leads to disruptive artifacts in the reconstructed images, hindering accurate diagnosis. Recently, a large amount of supervised deep learning-based approaches have been proposed for metal artifact reduction (MAR). However, these methods neglect the influence of initial training weights. In this paper, we have discovered that the uncertainty image computed from the restoration result of initial training weights can effectively highlight high-frequency regions, including metal artifacts. This observation can be leveraged to assist the MAR network in removing metal artifacts. Therefore, we propose an uncertainty constraint (UC) loss that utilizes the uncertainty image as an adaptive weight to guide the MAR network to focus on the metal artifact region, leading to improved restoration. The proposed UC loss is designed to be a plug-and-play method, compatible with any MAR framework, and easily adoptable. To validate the effectiveness of the UC loss, we conduct extensive experiments on the public available Deeplesion and CLINIC-metal dataset. Experimental results demonstrate that the UC loss further optimizes the network training process and significantly improves the removal of metal artifacts.
Paper Structure (13 sections, 4 equations, 4 figures, 1 table)

This paper contains 13 sections, 4 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Method
  3. Metal Artifact in Initial Training
  4. Uncertainty Constraint Loss
  5. Experiments and Results
  6. Dataset
  7. Implementation Details
  8. Evaluation Metrics
  9. Experiment Results on Synthetic Dataset
  10. Quantitative Comparison on Different MAR Network.
  11. Visual Comparison of Different MAR Network.
  12. Experiment on Clinical Dataset
  13. Conclusions

Figures (4)

  • Figure 1: Left: example image of CT with metal artifact. Middle: visualization of the restoration result of different training weight, in which the training epoch increases from left to right. Right: the uncertainty image calculated by the restoration results.
  • Figure 2: The overview of the proposed uncertainty constraint loss.
  • Figure 3: Visual comparison of different MAR network with or without the UC loss for supervision on synthetic dataset. The red rectangle denotes the enlarged region.
  • Figure 4: Visual comparison of MAR network with or without the UC loss for supervision on clinic dataset. The red rectangle denotes the enlarged region.