Improved robust tensor principal component analysis for accelerating dynamic MR imaging reconstruction.

Area of Science:

• Medical Imaging
• Image Reconstruction
• Tensor Decomposition

Background:

• Dynamic magnetic resonance imaging (dMRI) requires a balance between speed and accuracy.
• Reconstructing dMRI from highly undersampled K-space data is a significant challenge.
• Existing methods may not fully exploit the low-rank structures inherent in multi-way MRI data.

Purpose of the Study:

• To propose an improved robust tensor principal component analysis (RTPCA) method for dMRI reconstruction.
• To enhance the accuracy and computational efficiency of dMRI reconstruction from undersampled K-space data.
• To effectively leverage low-rank structures in multi-way MRI datasets for improved reconstruction.

Main Methods:

• Formulated the MR reconstruction problem as a high-order low-rank tensor plus sparse tensor recovery problem.
• Employed an improved robust tensor principal component analysis (RTPCA) with a novel tensor nuclear norm (TNN).
• Integrated the core matrix nuclear norm from tensor singular value decomposition (t-SVD) into TNN to enforce low-rank structures.

Main Results:

• The proposed RTPCA method demonstrated superior performance compared to state-of-the-art methods.
• Achieved higher MR image reconstruction accuracy, especially for 4D datasets.
• Showcased improved computational efficiency in both 3D and 4D dMRI reconstruction.

Conclusions:

• The novel RTPCA method effectively reconstructs dynamic MRI from highly undersampled K-space data.
• The integration of core matrix nuclear norm significantly enhances the exploitation of low-rank properties.
• The proposed approach offers a promising solution for faster and more accurate dMRI acquisition and reconstruction.