ADN: Artifact Disentanglement Network for Unsupervised Metal Artifact Reduction

Abstract

Current deep neural network based approaches to computed tomography (CT) metal artifact reduction (MAR) are supervised methods that rely on synthesized metal artifacts for training. However, as synthesized data may not accurately simulate the underlying physical mechanisms of CT imaging, the supervised methods often generalize poorly to clinical applications. To address this problem, we propose, to the best of our knowledge, the first unsupervised learning approach to MAR. Specifically, we introduce a novel artifact disentanglement network that disentangles the metal artifacts from CT images in the latent space. It supports different forms of generations (artifact reduction, artifact transfer, and self-reconstruction, etc.) with specialized loss functions to obviate the need for supervision with synthesized data. Extensive experiments show that when applied to a synthesized dataset, our method addresses metal artifacts significantly better than the existing unsupervised models designed for natural image-to-image translation problems, and achieves comparable performance to existing supervised models for MAR. When applied to clinical datasets, our method demonstrates better generalization ability over the supervised models. The source code of this paper is publicly available at https://github.com/liaohaofu/adn. Index Terms-Image enhancement/restoration (noise and artifact reduction), neural network, X-ray imaging and computed tomography.

Harmonized Segmentation of Neonatal Brain MRI

“…While training, we used the Adam optimizer (Kingma and Ba, 2014) with its default parameters and a decaying cyclical learning rate scheduler (Smith, 2017) with a base learning rate of 2 • 10 −6 and a maximum learning rate of 2 • 10 −3 . The choice of optimizer was based on knowledge of previous image translation literature (Isola et al, 2017;Zhu et al, 2017;Liao et al, 2019;Ranzini et al, 2020) where it yielded good results. At the same time, a varying learning rate during training was shown to improve results in fewer iterations when compared to using a fixed value (Smith, 2017).…”

Photon‐counting normalized metal artifact reduction (NMAR) in diagnostic CT

“…The biggest problem for deep learning MAR is that training data must be generated artificially, leading to disparities between artifacts in the training and real artifacts. Methods to treat this problem have been proposed as well 7,23–25 …”

Projection-domain iteration to estimate unreliable measurements

“…[21], a review was provided for the state-of-art technologies in metal artifact reduction, and the limitations of these technologies were also pointed out. Most recently, machine leaning methods are explored to battle the metal artifacts in CT [22][23][24][25]. In ref.…”

“…In ref. [22], an unsupervised deep neural network artifact disentanglement network was proposed to decouple the metal artifacts and the CT images for clinical applications. Reference [23] suggested a conditional generative adversarial network CGAN for data domain sinogram [24] reported a convolutional neural network based metal artifact reduction (CNN-MAR) framework.…”

“…In the second category [21][22][23][24][25], the beam hardening effects are "learned" from a large set of measurements with and without metals. The learned model is automatically achieved after the training phase.…”