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

TL;DR

This paper introduces an uncertainty-aware loss function for deep learning-based metal artifact reduction in CT images, improving artifact removal by leveraging uncertainty maps from initial training weights.

Contribution

It proposes a novel uncertainty constraint loss that enhances metal artifact removal by focusing on high-frequency regions, compatible with any MAR framework.

Findings

Significant improvement in artifact removal performance.

Effective highlighting of high-frequency artifact regions.

Compatibility with various 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.