Efficient Determination of TV Regularization Parameter via Plug-and-Play Local-Global Adaptation for MR Image Reconstruction

In magnetic resonance imaging (MRI) reconstruction, accelerated acquisition often results in undersampled k-space data, introducing aliasing artifacts and noise. Total Variation (TV) regularization is widely employed for its ability to suppress noise while preserving edge details, yet determining its optimal parameter remains challenging. This study proposes an efficient TV regularization parameter determination method via a plug-and-play local-global adaptation framework for compressed sensing MRI (CS-MRI) reconstruction, also referred to as the Local Structure and Global Information Metrics algorithm (LS-GIM). By integrating a dynamic model of local structural information and global information richness, this approach adaptively adjusts the TV parameter during iterations, leveraging the plug-and-play flexibility and robustness. Validated on brain, spine, heart, and knee MRI datasets, the method outperforms traditional techniques in reconstruction fidelity, edge preservation, and noise suppression, demonstrating broad adaptability across varying acceleration factors and anatomical regions. This framework offers an efficient, interpretable solution for TV parameter optimization, enhancing the practicality of CS-MRI in clinical applications.