Answers are grounded in this paper and cite their sources.
The prolonged duration of magnetic resonance imaging (MRI) presents a formidable challenge, necessitating the emergence of undersampling implementation as the primary strategy for expediting the imaging process. The optimization of the undersampling mask holds the potential to enhance imaging quality under equivalent acceleration. Diffusion model has showcased exceptional performance in image generation, offering heightened flexibility and an unsupervised nature. Consequently, it serves as a robust deep generation method for effectively addressing the inverse problem in MR reconstruction. Denoising diffusion probabilistic model (DDPM), distinguished by its enhanced flexibility in controlling the noise distribution, demonstrates superior adaptability to various undersampling modes, establishing itself as a promising deep learning method. In this study, we employ a novel approach to directly learn undersampling masks from data points, applying it to a reconstruction method for DDPM defined in K-space. Experimental evaluations conducted on publicly available fast MRI datasets reveal the method's commendable performance, surpassing conventional random bar mask-based and U-Net-based reconstruction methods and achieving superior reconstruction quality.
No takes yet. Share an insight, caveat, or question.
Li et al. (2025) studied this question.
Synapse has enriched 7 closely related papers on similar clinical questions. Consider them for comparative context: