Dynamic MR image reconstruction based on total generalized variation and low-rank decomposition

Area of Science:

• Medical Imaging
• Image Reconstruction
• Signal Processing

Background:

• Compressed sensing (CS) enables faster MRI acquisition but poses reconstruction challenges.
• Dynamic MRI requires efficient methods to capture temporal variations while maintaining image quality.

Purpose of the Study:

• To propose a novel decomposition-based model for compressed sensing-based dynamic MRI reconstruction.
• To leverage Total Generalized Variation (TGV) and nuclear norm for improved reconstruction accuracy.

Main Methods:

• The model decomposes dynamic MRI data into background and dynamic components.
• Nuclear norm is used for the time-coherent background, and spatiotemporal TGV for the sparse dynamic component.
• A first-order primal-dual algorithm is employed for model optimization, with norm estimation for convergence.

Main Results:

• The proposed model achieved superior performance over state-of-the-art methods (kt-SLR, kt-RPCA, L+S, ICTGV).
• Higher Signal-to-Error Ratio (SER) and Structural Similarity Index Measure (SSIM) were obtained on cardiac perfusion and breast DCE-MRI datasets.
• Improved suppression of spatial artifacts and better edge preservation were observed under pseudoradial and Cartesian sampling schemes.

Conclusions:

• The novel decomposition-based model offers significant improvements in dynamic MRI reconstruction.
• High-quality reconstructions are achieved across various sampling schemes and acceleration factors.
• The method demonstrates superior performance and robustness compared to existing techniques.