IPv2: An Improved Image Purification Strategy for Real-World Ultra-Low-Dose Lung CT Denoising

TL;DR

This work systematically redesigns the original image purification strategy and proposes an improved version termed IPv2, which consistently improves background suppression and lung parenchyma restoration across multiple mainstream denoising models.

Abstract

The image purification strategy constructs an intermediate distribution with aligned anatomical structures, which effectively corrects the spatial misalignment between real-world ultra-low-dose CT and normal-dose CT images and significantly enhances the structural preservation ability of denoising models. However, this strategy exhibits two inherent limitations. First, it suppresses noise only in the chest wall and bone regions while leaving the image background untreated. Second, it lacks a dedicated mechanism for denoising the lung parenchyma. To address these issues, we systematically redesign the original image purification strategy and propose an improved version termed IPv2. The proposed strategy introduces three core modules, namely Remove Background, Add noise, and Remove noise. These modules endow the model with denoising capability in both background and lung tissue regions during training data construction and provide a more reasonable evaluation protocol through refined label construction at the testing stage. Extensive experiments on our previously established real-world patient lung CT dataset acquired at 2% radiation dose demonstrate that IPv2 consistently improves background suppression and lung parenchyma restoration across multiple mainstream denoising models. The code is publicly available at https://github.com/MonkeyDadLufy/Image-Purification-Strategy-v2.

Figures (5)

• Figure 1: Motivation of the improved image purification strategy (IPv2). (a) Ultra-low-dose CT (uLDCT) image with severe noise that degrades structural clarity. (b) Corresponding normal-dose CT (NDCT) image. (c) NDCT processed by the IPv1 strategy, which serves as the test label. (d) NDCT processed by the IPv2 strategy. (e) Denoising result of (a) obtained by training a flow matchingFM (FM) model in an end-to-end manner under the IPv1 strategy. (f) Denoising result of (a) obtained by training a FMFM model in an end-to-end manner under the IPv2 strategy. The green enlarged boxes in (a) to (f) indicate background regions, and the red enlarged boxes indicate lung tissue regions. The red arrows denote the presence of noise, while the green arrows denote the absence of noise. Please zoom in for details.
• Figure 2: Overview of the proposed improved image purification strategy (IPv2).
• Figure 3: Comparison of visual quality for various denoising methods under the previous purification strategy (IPv1) and the proposed strategy (IPv2). (c) Label constructed with IPv1, (d) Label constructed with IPv2. Red arrows indicate the presence of noise, green arrows indicate the absence of noise, and yellow arrows indicate slightly inferior denoising performance.
• Figure 4: Visualization of processed NDCT results generated by different modules within the purification strategy, which serve as labels during testing. (c) Result from the previous purification strategy (IPv1). (d) to (f) Results from adding Module 1, adding Module 3, and adding both Module 1 and Module 3 to IPv1, respectively. Red arrows indicate the presence of noise, green arrows indicate the absence of noise, and yellow arrows indicate the removal of useful information.
• Figure 5: Denoising results on uLDCT using the FMFM model with different modules in the image purification strategy. (c) Label constructed with IPv2. (d) to (g) Denoising results of (a) using FMFM trained on IPv1 without adding any module, adding Module 1, adding Module 2, and adding both Module 1 and Module 2, respectively. Red arrows indicate the presence of noise, green arrows indicate the absence of noise, and yellow arrows indicate the removal of useful information.