ISRSL0: compressed sensing MRI with image smoothness regularized-smoothed [Formula: see text] norm

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Magnetic Resonance Imaging (MRI) acceleration is crucial for clinical applications.
• Compressed Sensing (CS) enables MRI reconstruction from limited k-space data.
• CS-MRI faces challenges like fine structure loss and computational complexity.

Purpose of the Study:

• To enhance Compressed Sensing MRI (CS-MRI) performance, particularly fine structure preservation under high acceleration.
• To introduce a novel framework addressing limitations of existing CS-MRI techniques.

Main Methods:

• A novel framework combining regularized sparse recovery and a sharpening step for CS-MRI.
• Utilizing the Half Quadratic Splitting (HQS) approach to solve the inverse problem.
• Replacing a sub-problem with a denoiser for regularization, aiding Smoothed L0 (SL0) norm optimization.

Main Results:

• The proposed method improves fine structure preservation in CS-MRI at high acceleration factors.
• Analytical illustration of the approach's convergence properties.
• Experimental results demonstrate acceptable performance compared to network-based methods.

Conclusions:

• The novel framework effectively enhances CS-MRI reconstruction quality, particularly for fine details.
• The method offers an improved SL0 algorithm for MRI reconstruction without increased complexity.
• The approach shows promise for accelerating MRI scans while maintaining image fidelity.