Sparse-view CT image reconstruction using conditional embedding fusion diffusion model

doi:10.1016/j.neucom.2025.131748

Sparse-view CT image reconstruction using conditional embedding fusion diffusion model

2025-10-06

0

PRE

AI

DOI：10.1016/j.neucom.2025.131748

Original

Original request for help

Save

Abstract

En 中文

The main goal of sparse-view CT reconstruction is to reconstruct high-quality images from fewer projection data, thereby reducing the radiation dose. However, the reduction in the amount of projection data also makes the reconstruction process ill-posed, leading to an increase in artifacts in the reconstructed image. Recently, diffusion models have shown strong generative capabilities, providing new ideas for sparse-view CT reconstruction. But due to the stochastic nature of the diffusion process, how to guide the generation process to achieve high-quality reconstruction remains a major challenge. To address this issue, we propose a Conditional Embedding Fusion Diffusion Model (CEF-DM) to improve reconstruction quality. Specifically, we design a FourierNet to generate an initial reconstruction, which serves as a condition to guide the CEF-DM in generating the remaining detail residuals. CEF-DM employs a conditional attention embedding module (CAEM) to comprehensively incorporate conditional input and time-step information throughout the generation process. The initial reconstruction is then summed with the residual details to obtain the final reconstruction. The experimental results show that our method outperforms existing methods in terms of reconstruction accuracy and image quality, providing an efficient and robust solution for sparse view CT reconstruction with potential clinical application value.

Sparse-view CT image reconstruction using conditional embedding fusion diffusion model

Abstract

Journal

Organization