Advancing Free-Breathing Cardiac Cine MRI: Retrospective Respiratory Motion Correction Via Kspace-and-Image Guided Diffusion Model

Hongming Guo, Ziqing Huang, Qian Yuan, Hanbo Song, Zhiyan Liu, Xianzhao Feng, Anqi Liu, Min Liu, Ke Li, Ruixi Zhou

Abstract: This study introduces the Double-Guidance Diffusion Model (DB-DDPM), a novel conditional Denoising Diffusion Probabilistic Model (DDPM) designed specifically for high-quality correction of respiratory motion, a prevalent challenge in cardiac cine MRI. Respiratory motion, caused by the natural movement of the thorax and diaphragm during breathing, often results in artifacts that can significantly degrade image quality. By leveraging dual-domain conditioning from both the image and frequency domains, the proposed DB-DDPM not only enhances artifact removal efficacy but also significantly accelerates the diffusion model inference process. This improvement in operational speed facilitates the rapid reconstruction of images that are free from the distortions typically introduced by respiratory motion. Our experiments demonstrate that DB-DDPM surpasses existing artifact reduction methodologies in both qualitative and quantitative assessments, establishing a new benchmark for rapid and accurate respiratory motion correction with exceptional robustness in dynamic imaging sequences.

External IDs:dblp:conf/icann/GuoHYSLFLLLZ24