Improved dynamic MRI reconstruction by exploiting sparsity and rank-deficiency

Area of Science:

• Medical Imaging
• Signal Processing
• Applied Mathematics

Background:

• Dynamic MRI reconstruction from undersampled K-space data is challenging.
• Existing methods use l(1)-norm and Schatten-q norm for sparsity and low-rank penalties, respectively.
• These penalties are surrogates for NP-hard l(0)-norm and matrix rank.

Purpose of the Study:

• To improve dynamic MRI reconstruction by employing a non-convex lp-norm as a better sparsity penalty.
• To develop efficient algorithms for solving the proposed reconstruction problem.
• To evaluate the proposed method on Dynamic Contrast Enhanced (DCE) MRI datasets.

Main Methods:

• Reconstruction formulated as a least squares minimization problem.
• Regularization using lp-norm (0

• Derivation of efficient algorithms to solve the minimization problem.

Main Results:

• The proposed method with lp-norm penalty demonstrates superior performance compared to existing techniques.
• Both quantitative and qualitative analyses confirm the effectiveness of the new approach.
• Successful application on Dynamic Contrast Enhanced (DCE) MRI data.

Conclusions:

• The use of lp-norm as a sparsity penalty offers significant improvements in dynamic MRI reconstruction.
• The developed efficient algorithms enable practical application of this advanced technique.
• This work advances the field of dynamic MRI by providing a more accurate and efficient reconstruction method.