Fast Compressed Sensing of 3D Radial T₁ Mapping with Different Sparse and Low-Rank Models.

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging
• Image Reconstruction

Background:

• Compressed sensing models are crucial for accelerating MRI acquisition.
• Limited knowledge exists on the comparative performance of sparse and low-rank models in 3D radial quantitative MRI.
• T1 relaxation time mapping is essential for various diagnostic applications.

Purpose of the Study:

• To compare the performance of different compressed sensing regularization models for 3D radial T1 mapping.
• To identify the optimal compressed sensing model for quantitative MRI reconstruction.
• To evaluate reconstruction quality using normalized root mean squared error and structural similarity index.

Main Methods:

• Utilized 3D radial T1 relaxation time mapping data.
• Compared total variation, low-rank, and Huber penalty function regularization approaches.
• Employed simulation and ex vivo specimen data for model evaluation.
• Solved large-scale compressed sensing models using a GPU-accelerated primal-dual proximal splitting algorithm.

Main Results:

• The model combining spatial total variation and locally low-rank regularization demonstrated superior performance.
• A model incorporating spatial and contrast dimension total variation also showed strong results.
• Reconstruction times varied significantly, with low-rank methods being more computationally intensive (2-113 min).

Conclusions:

• The combination of spatial total variation and locally low-rank regularization is recommended for 3D radial T1 mapping.
• Image reconstruction performance is significantly influenced by the characteristics of the imaged object.
• Efficient GPU implementation enables feasible computation times for high-quality quantitative MRI.