MRI reconstruction with joint global regularization and transform learning.

Area of Science:

• Medical Imaging
• Image Reconstruction
• Signal Processing

Background:

• Sparsity-based regularization is crucial for addressing data scarcity in image reconstruction.
• Learned sparsifying transforms from image patches are effective regularizers in Magnetic Resonance Imaging (MRI).

Purpose of the Study:

• To enhance MRI reconstruction by integrating global regularization with patch-based transform learning.
• To develop and evaluate a novel cost function incorporating both patchwise and global regularization terms.

Main Methods:

• Developed a new algorithm to solve a cost function combining patchwise and global regularization terms.
• Implemented and tested the mixed regularization approach on MRI data.

Main Results:

• The proposed mixed regularization approach demonstrated improved MRI reconstruction performance.
• Performance gains were observed in comparison to methods using only patchwise or global regularization.

Conclusions:

• Integrating global regularization with learned patch-based transforms offers superior MRI reconstruction.
• The novel mixed approach effectively addresses limitations of individual regularization strategies.