[论文解读] Unsupervised Polychromatic Neural Representation for CT Metal Artifact Reduction

Qing Wu, Lixuan Chen+5 more|arXiv (Cornell University)|Jun 27, 2023

Advanced X-ray and CT Imaging被引用 8

一句话总结

Polyner 是一种无监督的多色隐式神经表示，通过建模非线性金属效应并使用可微分的多色前向模型来重建金属影响的 CT 图像，在域内外表现出竞争力或优于其他方法的 MAR 性能，且无需外部训练数据。

ABSTRACT

Emerging neural reconstruction techniques based on tomography (e.g., NeRF, NeAT, and NeRP) have started showing unique capabilities in medical imaging. In this work, we present a novel Polychromatic neural representation (Polyner) to tackle the challenging problem of CT imaging when metallic implants exist within the human body. CT metal artifacts arise from the drastic variation of metal's attenuation coefficients at various energy levels of the X-ray spectrum, leading to a nonlinear metal effect in CT measurements. Recovering CT images from metal-affected measurements hence poses a complicated nonlinear inverse problem where empirical models adopted in previous metal artifact reduction (MAR) approaches lead to signal loss and strongly aliased reconstructions. Polyner instead models the MAR problem from a nonlinear inverse problem perspective. Specifically, we first derive a polychromatic forward model to accurately simulate the nonlinear CT acquisition process. Then, we incorporate our forward model into the implicit neural representation to accomplish reconstruction. Lastly, we adopt a regularizer to preserve the physical properties of the CT images across different energy levels while effectively constraining the solution space. Our Polyner is an unsupervised method and does not require any external training data. Experimenting with multiple datasets shows that our Polyner achieves comparable or better performance than supervised methods on in-domain datasets while demonstrating significant performance improvements on out-of-domain datasets. To the best of our knowledge, our Polyner is the first unsupervised MAR method that outperforms its supervised counterparts. The code for this work is available at: https://github.com/iwuqing/Polyner.

研究动机与目标

• Motivate a nonlinear, polychromatic view of CT MAR to preserve all measurement information including metal traces.
• Propose Polyner, an unsupervised INR-based method that incorporates a polychromatic forward model.
• Introduce a regularization that preserves physical energy-level properties across spectra.
• Demonstrate robustness to out-of-domain data and compare with supervised MAR methods.

提出的方法

• Model the CT MAR problem as polychromatic image reconstruction across N energy levels using an INR to map coordinates to N LAC maps.
• Derive a differentiable polychromatic CT forward model that aggregates energy-resolved projections into measurements.
• Train the INR by minimizing a data-consistency loss between real and predicted measurements plus an energy-dependent smoothness loss over adjacent energy levels.
• Use a forward-model-driven optimization to recover polychromatic LACs and output the final monochromatic image at an energy representative of the spectrum.
• Implement hash encoding with a two-layer MLP for the INR and optimize with Adam on sampled X-ray views.

实验结果

研究问题

• RQ1 Can unsupervised, polychromatic INR-based reconstruction compete with supervised MAR methods on in-domain data?
• RQ2 Does incorporating a polychromatic forward model improve MAR performance over linear/monochromatic approaches?
• RQ3 What is the impact of energy-level discretization and energy-smooth regularization on MAR results?
• RQ4 How does the method perform on out-of-domain or real-world datasets compared to supervised baselines?

主要发现

• Polyner achieves PSNR/SSIM comparable to or better than supervised MAR methods on in-domain data and outperforms them on out-of-domain data.
• Ablation shows the polychromatic forward model substantially improves MAR performance (PSNR gain of about 3.92 dB over linear forward models).
• Energy-dependent smooth loss enhances consistency across energy levels and improves reconstruction quality.
• Increasing the number of energy levels N improves MAR performance and image details.
• Polyner remains unsupervised and does not require external training data, while delivering robust performance on diverse datasets.