MRI reconstruction of multi-image acquisitions using a rank regularizer with data reordering

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging (MRI)
• Image Reconstruction

Background:

• Undersampled MRI acquisitions are crucial for faster scanning but often result in lower image quality.
• Low-rank matrix completion theory offers potential for reconstructing high-quality images from limited data.
• Dynamic MRI applications, like cardiac perfusion imaging, benefit from rapid acquisition techniques.

Purpose of the Study:

• To enhance rank-constrained reconstructions for undersampled multi-image MRI acquisitions.
• To introduce a novel reordering-based approach for improved MRI reconstruction.
• To leverage prior image information for more accurate undersampled data reconstruction.

Main Methods:

• A reordering-based rank-constrained reconstruction method was developed, utilizing prior image estimates.
• The method minimizes the nuclear norm of reordered matrix values, differing from standard nuclear norm minimization.
• The technique was validated on brain diffusion imaging and dynamic contrast-enhanced myocardial perfusion datasets.

Main Results:

• High-quality images were successfully reconstructed from data undersampled by factors of 3 (diffusion) and 3.5 (perfusion).
• The reordering approach yielded visually superior image quality compared to standard nuclear norm minimization.
• Root mean squared errors were reduced by approximately 18% for diffusion and 16% for perfusion imaging.

Conclusions:

• The proposed reordered low-rank constraint effectively incorporates prior image information.
• This method demonstrates significant improvements over standard low-rank constraints in undersampled multi-image MRI.
• The technique shows promise for enhancing the quality of dynamic and diffusion MRI scans.