Noise Controlled CT Super-Resolution with Conditional Diffusion Model
Yuang Wang, Siyeop Yoon, Rui Hu, Baihui Yu, Duhgoon Lee, Rajiv Gupta, Li Zhang, Zhiqiang Chen, Dufan Wu
TL;DR
The paper tackles noise amplification in CT super-resolution by developing a noise-controlled framework based on a Conditional Denoising Diffusion Probabilistic Model (DDPM). It trains a conditional diffusion process using LR conditions $y$ to generate HR images $x_0$, with a forward diffusion $q(x_t|x_{t-1})=N\left(x_t|\sqrt{1-\beta_t}x_{t-1}, \beta_t I\right)$ and a learnable reverse $p_\theta(x_{t-1}|x_t,y)$, optimized via $\theta^* = \arg\min_\theta E_{x_0,y,t,\epsilon} \|\epsilon_\theta(x_t,y,t)-\epsilon\|^2$. The framework uses hybrid training data: noise-matched simulations for HR/LR pairs and segmented real bone details to preserve structure without introducing noise amplification, implemented with a U-Net-based conditional predictor and 1000 diffusion steps. Real CT experiments show improved spatial resolution and texture preservation, particularly in bone structures, compared to simulation-only and non-noise-matched baselines. Overall, the method offers a practical, diffusion-based route to high-resolution CT imaging with controlled noise, holding promise for clinical use, while highlighting dependencies on accurate segmentation and integration of real details.
Abstract
Improving the spatial resolution of CT images is a meaningful yet challenging task, often accompanied by the issue of noise amplification. This article introduces an innovative framework for noise-controlled CT super-resolution utilizing the conditional diffusion model. The model is trained on hybrid datasets, combining noise-matched simulation data with segmented details from real data. Experimental results with real CT images validate the effectiveness of our proposed framework, showing its potential for practical applications in CT imaging.