Unsupervised Low-dose CT Reconstruction with One-way Conditional Normalizing Flows

TL;DR

Low-dose CT reconstruction remains challenging due to noise and the lack of labeled data. The authors propose OW-CNFs, an unsupervised conditional normalizing flow framework that uses a strict one-way transformation and an unsupervised conditionalization strategy to enable high-resolution training and fast, detail-preserving reconstruction. The approach combines an OS-SART–based fidelity step with a CNF prior in a four-block Glow-like backbone conditioned at all layers, delivering competitive performance against both unsupervised and supervised methods while offering faster reconstruction than diffusion-based models. Experimental results on two datasets demonstrate strong quantitative gains (PSNR/SSIM) and favorable speed, highlighting the method's potential for practical clinical LDCT reconstruction without paired data.

Abstract

Deep-learning methods have shown promising performance for low-dose computed tomography (LDCT) reconstruction. However, supervised methods face the problem of lacking labeled data in clinical scenarios, and the CNN-based unsupervised denoising methods would cause excessive smoothing in the reconstructed image. Recently, the normalizing flows (NFs) based methods have shown advantages in producing detail-rich images and avoiding over-smoothing, however, there are still issues: (1) Although the alternating optimization in the data and latent space can well utilize the regularization and generation capabilities of NFs, the current two-way transformation strategy of noisy images and latent variables would cause detail loss and secondary artifacts; and (2) Training NFs on high-resolution CT images is hard due to huge computation. Though using conditional normalizing flows (CNFs) to learn conditional probability can reduce the computational burden, current methods require labeled data for conditionalization, and the unsupervised CNFs-based LDCT reconstruction remains a problem. To tackle these problems, we propose a novel CNFs-based unsupervised LDCT iterative reconstruction algorithm. It employs strict one-way transformation when performing alternating optimization in the dual spaces, thus effectively avoiding the problems of detail loss and secondary artifacts. By proposing a novel unsupervised conditionalization strategy, we train CNFs on high-resolution CT images, thus achieving fast and high-quality unsupervised reconstruction. Experiments on different datasets suggest that the performance of the proposed algorithm could surpass some state-of-the-art unsupervised and even supervised methods.

Figures (9)

• Figure 1: The reconstructed images of the two-way NFs. Compared to the normal-dose images, structure distortion and noise residuals can be easily observed.
• Figure 2: Condition examples generated with the BM3D denoiser.
• Figure 3: The network structure of our conditional normalizing flows.
• Figure 4: Some examples of the "RRM" dataset ((a)-(c)) and "LIDC-IDRI" dataset ((d)-(f)).
• Figure 5: Reconstruction results of each method on the "RRM" dataset at dose $I_{0}=1\times{10^3}$. The display window of the gray value range is set to $[0,1.0]$.