上海科技大学知识管理系统

Cardiac Cine Magnetic Resonance Imaging (MRI) acquires dynamic images that depict the rhythmic contractions of the heart. Nevertheless, the extended duration of MRI scans may cause discomfort to patients and lead to motion artifacts in the acquired images. There is a compelling clinical need to develop methods that can shorten the scan time and alleviate the impact of motion-induced artifacts. Utilizing the Denoising Diffusion Probabilistic Model, a robust generative deep learning model, we present a supervised learning-based approach for cardiac motion correction. The motion-artifacts reduction begins with sampling from the Gaussian noise and motion-corrupted images, followed by iterative refinement through a U-Net architecture. This architecture is trained to denoise motion-corrupted images across various noise levels. We demonstrate that the framework is effective for cardiac cine MRI motion correction.