DuDoNet: Dual Domain Network for CT Metal Artifact Reduction (1907.00273v1)

Abstract: Computed tomography (CT) is an imaging modality widely used for medical diagnosis and treatment. CT images are often corrupted by undesirable artifacts when metallic implants are carried by patients, which creates the problem of metal artifact reduction (MAR). Existing methods for reducing the artifacts due to metallic implants are inadequate for two main reasons. First, metal artifacts are structured and non-local so that simple image domain enhancement approaches would not suffice. Second, the MAR approaches which attempt to reduce metal artifacts in the X-ray projection (sinogram) domain inevitably lead to severe secondary artifact due to sinogram inconsistency. To overcome these difficulties, we propose an end-to-end trainable Dual Domain Network (DuDoNet) to simultaneously restore sinogram consistency and enhance CT images. The linkage between the sigogram and image domains is a novel Radon inversion layer that allows the gradients to back-propagate from the image domain to the sinogram domain during training. Extensive experiments show that our method achieves significant improvements over other single domain MAR approaches. To the best of our knowledge, it is the first end-to-end dual-domain network for MAR.

• The paper proposes DuDoNet, a dual-domain network that simultaneously processes the sinogram and image domains to reduce metal artifacts in CT scans.
• It introduces a novel Radon inversion layer and mask pyramid U-Net architecture to ensure accurate restoration and multi-scale retention of metal mask information.
• Experiments demonstrate significant improvements in PSNR and SSIM over traditional methods, proving its robustness across varying implant sizes.

Insightful Overview of DuDoNet: Dual Domain Network for CT Metal Artifact Reduction

The paper "DuDoNet: Dual Domain Network for CT Metal Artifact Reduction" explores an innovative approach to tackling the persistent issue of metal artifacts in computed tomography (CT) images. CT imaging is a prevalent tool in medical diagnostics and treatments, however, the presence of metallic implants can lead to pronounced artifacts, compromising the quality of the images and potentially leading to erroneous diagnostics. Previous attempts to mitigate these metal artifacts have typically operated in either the sinogram domain or the image domain, each facing its limitations.

Key Contributions and Methodology

The authors introduce DuDoNet, an end-to-end trainable dual-domain network designed to enhance CT imaging quality by addressing metal artifact reduction (MAR) simultaneously in the sinogram and image domains. The core innovation lies in its ability to restore sinogram consistency while concurrently refining CT images, thus effectively reducing metal shadows and secondary artifacts. This dual-domain linkage is facilitated by the novel Radon inversion layer. This layer plays a crucial role as it enables gradients to back-propagate efficiently from the image domain to the sinogram domain during training, which is essential for the end-to-end operation.

Additionally, the segmenting nature of metal artifacts, particularly in scenarios involving smaller implants, requires a solution that preserves image information across multiple scales. To address this challenge, DuDoNet employs a mask pyramid U-Net architecture within the sinogram enhancement network (SE-Net). This architecture ensures that metal mask information is retained without loss across various hierarchical levels of the network.

Numerical Results and Evaluation

The demonstrated efficacy of DuDoNet is supported by extensive experimentation, which reveals substantial improvements over traditional single-domain MAR approaches. Specifically, DuDoNet shows marked advancements in peak signal-to-noise ratio (PSNR) and structured similarity index (SSIM), outperforming existing techniques like linear interpolation (LI), normalized metal artifact reduction (NMAR), and convolutional neural network-based approaches such as CNNMAR. Notably, the paper reports significant results where sizes of metal implants vary, ensuring the robustness of the algorithm across diverse clinical scenarios.

Practical and Theoretical Implications

Practically, DuDoNet offers a promising solution for enhancing the accuracy and reliability of CT scans amidst the increasingly common presence of metallic implants. Its dual-domain approach suggests a paradigm shift in handling complex image artifacts where traditional methods falter. Theoretically, this work contributes to the field of inverse problem-solving in imaging by integrating physical laws through the Radon consistency loss within a deep learning framework, offering a novel perspective on consistency loss implementation that could broaden applications beyond MAR.

Speculation on Future Work

Looking forward, the dual-domain learning framework introduced by DuDoNet holds potential for further developments in signal restoration tasks beyond those limited to metal artifact reduction. Future research might explore applications in image super-resolution, noise reduction, and sparse-view CT reconstruction. Additionally, there may be opportunities to extend this dual-domain network architecture to other medical imaging modalities that suffer from artifact-induced distortions.

In summary, the research presented in this paper contributes meaningfully to the field of medical imaging and artifact reduction, offering an effective and computationally efficient solution in the form of DuDoNet. Its approach of simultaneous refinement in dual domains backed by robust numerical evidence sets a promising direction for future studies and practical applications in AI and medical diagnostics.