Reference-guided sparsifying transform design for compressive sensing MRI.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Signal Processing

Background:

• Compressive sensing (CS) MRI reconstructs images from undersampled k-space data.
• Traditional CS-MRI methods use analytical sparsifying transforms (e.g., total-variation, wavelets).
• Nonparametric dictionary-based methods offer improved synthesis-based reconstruction but focus on learning representation bases.

Purpose of the Study:

• To present a new framework for analysis-based CS-MRI reconstruction.
• To utilize a learned sparsifying transform from a reference image to guide reconstruction.
• To improve the accurate reconstruction of anatomical structures in undersampled MRI data.

Main Methods:

• Developed a novel framework for analysis-based CS-MRI reconstruction.
• Learned a sparsifying transform from a reference image.
• Applied the learned transform to guide the reconstruction of undersampled k-space data.

Main Results:

• The proposed analysis-based approach demonstrated high performance.
• Experimental data showed superior results compared to traditional CS-MRI methods.
• The learned transform effectively captured and utilized anatomical structure information.

Conclusions:

• The new framework offers an effective analysis-based approach for CS-MRI reconstruction.
• Learning sparsifying transforms from reference images enhances anatomical accuracy.
• This method shows significant potential for improving undersampled MRI image quality.