A Diffusion-Based Generative Prior Approach to Sparse-view Computed Tomography

TL;DR

This work tackles sparse-view CT reconstruction by embedding a diffusion-based generative prior within a regularized latent-optimization framework. The RD-DGP method minimizes a data-fidelity term $\tfrac{1}{2}\|\boldsymbol{K}\mathcal{G}(\boldsymbol{z})-\boldsymbol{y}^\delta\|_2^2$ augmented with latent and image-domain priors $\lambda_1\|\boldsymbol{z}\|_2^2$ and TV$(\mathcal{G}(\boldsymbol{z}))$, where $\mathcal{G}$ is a DDIM-based generator. A physics-informed initialization via FBP followed by DDIM inversion, together with a cosine-annealed learning-rate schedule, significantly improves convergence and reconstruction quality. The model is trained on chest CT slices with extensive data augmentation, and a diffusion model pre-trained on Mayo chest CT data is publicly released. Empirical results show RD-DGP generally achieves data-consistent, high-quality reconstructions comparable to or better than existing diffusion-based inverse-solvers, though substantial challenges remain in computational cost and robustness for clinical deployment.

Abstract

The reconstruction of X-rays CT images from sparse or limited-angle geometries is a highly challenging task. The lack of data typically results in artifacts in the reconstructed image and may even lead to object distortions. For this reason, the use of deep generative models in this context has great interest and potential success. In the Deep Generative Prior (DGP) framework, the use of diffusion-based generative models is combined with an iterative optimization algorithm for the reconstruction of CT images from sinograms acquired under sparse geometries, to maintain the explainability of a model-based approach while introducing the generative power of a neural network. There are therefore several aspects that can be further investigated within these frameworks to improve reconstruction quality, such as image generation, the model, and the iterative algorithm used to solve the minimization problem, for which we propose modifications with respect to existing approaches. The results obtained even under highly sparse geometries are very promising, although further research is clearly needed in this direction.

Figures (5)

• Figure 1: Graphical illustration of the Deep Generative Prior approach with Denoising Diffusion Implicit Models. It illustrates how to generate a single iterate $x_0^{K}$ of the optimization method after $K$ reverse diffusion steps starting from $x_0$.
• Figure 2: Visual comparison of different reconstruction methods for Sample C081-35 across three sparse-view configurations ($n_\alpha = 30, 45, 60$). The red box indicates the zoomed region shown in the inset.
• Figure 3: Visual comparison of different reconstruction methods for Sample C081-45 across three sparse-view configurations ($n_\alpha = 30, 45, 60$). The red box indicates the zoomed region shown in the inset.
• Figure 4: Visual comparison of different reconstruction methods for Sample C081-79 across three sparse-view configurations ($n_\alpha = 30, 45, 60$). The red box indicates the zoomed region shown in the inset.
• Figure 5: Ablation study on sample C081-35 under sparse-view CT. Each block reports reconstructions (with PSNR/SSIM overlaid) obtained with different combinations of FBP initialization (Yes/No) and step-size schedule (Yes/No). From left to right, the number of projection angles $n_\alpha$ is $30$, $45$, and $60$.