Stage-by-stage Wavelet Optimization Refinement Diffusion Model for Sparse-View CT Reconstruction

TL;DR

This paper introduces SWORD, a wavelet-based diffusion model with a three-stage optimization process that improves sparse-view CT reconstruction by enhancing stability and capturing multi-scale features.

Contribution

The paper proposes a novel wavelet-guided diffusion model with a three-stage optimization, operating on wavelet components for improved stability and reconstruction quality in sparse-view CT.

Findings

Outperforms existing methods quantitatively.

Achieves superior qualitative reconstruction results.

Ensures stable training through wavelet domain processing.

Abstract

Diffusion models have emerged as potential tools to tackle the challenge of sparse-view CT reconstruction, displaying superior performance compared to conventional methods. Nevertheless, these prevailing diffusion models predominantly focus on the sinogram or image domains, which can lead to instability during model training, potentially culminating in convergence towards local minimal solutions. The wavelet trans-form serves to disentangle image contents and features into distinct frequency-component bands at varying scales, adeptly capturing diverse directional structures. Employing the Wavelet transform as a guiding sparsity prior significantly enhances the robustness of diffusion models. In this study, we present an innovative approach named the Stage-by-stage Wavelet Optimization Refinement Diffusion (SWORD) model for sparse-view CT reconstruction. Specifically, we establish a unified mathematical model integrating low-frequency and high-frequency generative models, achieving the solution with optimization procedure. Furthermore, we perform the low-frequency and high-frequency generative models on wavelet's decomposed components rather than sinogram or image domains, ensuring the stability of model training. Our method rooted in established optimization theory, comprising three distinct stages, including low-frequency generation, high-frequency refinement and domain transform. Our experimental results demonstrate that the proposed method outperforms existing state-of-the-art methods both quantitatively and qualitatively.