CIResDiff: A Clinically-Informed Residual Diffusion Model for Predicting Idiopathic Pulmonary Fibrosis Progression

TL;DR

IPF progression prediction is challenged by criteria that rely on changes between two scans; CIResDiff addresses this by generating the follow-up CT from the initial CT using a Clinically-Informed Residual Diffusion model. The forward process shift leverages the lesion difference $e_0=y_0-x_0$ with a schedule $\{\\eta_t\}_{t=1}^T$, i.e., $q(x_t|x_0,y_0)=\mathcal{N}(x_t; x_0+\\eta_t e_0, k^2 \\eta_t I)$, while a learned reverse network recovers $x_0$ from $x_t$ and $y_0$ under a reconstruction loss. The method adds a CLIP-based clinically-informed process to fuse lung-function data into the diffusion reverse process via cross-attention, after aligning the target lung region. Experiments on the OSIC IPF dataset show that CIResDiff outperforms state-of-the-art diffusion and GAN-based methods in PSNR/SSIM and improves diagnostic utility of generated images, supporting earlier, more informed treatment decisions in IPF care.

Abstract

The progression of Idiopathic Pulmonary Fibrosis (IPF) significantly correlates with higher patient mortality rates. Early detection of IPF progression is critical for initiating timely treatment, which can effectively slow down the advancement of the disease. However, the current clinical criteria define disease progression requiring two CT scans with a one-year interval, presenting a dilemma: a disease progression is identified only after the disease has already progressed. To this end, in this paper, we develop a novel diffusion model to accurately predict the progression of IPF by generating patient's follow-up CT scan from the initial CT scan. Specifically, from the clinical prior knowledge, we tailor improvements to the traditional diffusion model and propose a Clinically-Informed Residual Diffusion model, called CIResDiff. The key innovations of CIResDiff include 1) performing the target region pre-registration to align the lung regions of two CT scans at different time points for reducing the generation difficulty, 2) adopting the residual diffusion instead of traditional diffusion to enable the model focus more on differences (i.e., lesions) between the two CT scans rather than the largely identical anatomical content, and 3) designing the clinically-informed process based on CLIP technology to integrate lung function information which is highly relevant to diagnosis into the reverse process for assisting generation. Extensive experiments on clinical data demonstrate that our approach can outperform state-of-the-art methods and effectively predict the progression of IPF.

Figures (4)

• Figure 1: Overview of proposed CIResDiff. (a) and (c) provide the framework of CIResDiff as well as depict its implementation during both the training and testing phases, and (b) illustrates the details of clinically-informed process.
• Figure 1: Quantitative results of ablation analysis, in terms of PSNR and SSIM.
• Figure 2: Diagnostic evaluation.
• Figure 3: Visual comparison of follow-up lung images produced by six different methods. From left to right are the input (initial scan), results of five other comparison methods (2nd-6th columns) and our CIResDiff (7th column), and the ground truth (follow-up scan). The corresponding difference maps between the generated results and GT are shown in the 2nd and 4th rows, where darker colors indicate larger differences. Red boxes show the lesion areas for detailed comparison.