Physics-Driven Zero-Shot MRI Reconstruction with Non-local Image Priors
Researchers have developed a new zero-shot self-supervised learning (ZS-SSL) framework for accelerated Magnetic Resonance Imaging (MRI) reconstruction. This framework aims to overcome the limitations of existing ZS-SSL methods, such as supervision scarcity and optimization instability, which can lead to artifacts. The proposed approach integrates physical consistency with non-local image priors, featuring a CSM-Guided Dynamic Repository, SPIRiT-based regularization, and a Non-Local Self-Similarity (NSS) Pixel Bank. Experiments on the FastMRI dataset show that this method achieves state-of-the-art performance, especially at high acceleration factors. AI