Basic study of random sampling for compressed sensing using MRI simulator.

Area of Science:

• Medical Imaging
• Computational Science

Background:

• Magnetic Resonance Imaging (MRI) is a key tomography technology for anatomical depiction.
• Compressed Sensing (CS) enables MRI reconstruction from sparse data.
• Developing efficient random sampling methods is crucial for CS-based MRI.

Purpose of the Study:

• To evaluate the accuracy of numerical simulations for compressed sensing MRI.
• To compare a basic simulation with an advanced method using actual MRI scanner data.
• To assess the performance of different random sampling strategies, including those considering Hermitian symmetry.

Main Methods:

• A basic simulation used a 2D Shepp-Logan phantom and k-space data.
• An advanced simulation employed an MRI simulator (Bloch Solver) with k-space data from an actual MRI scanner (Vantage Titan 3T) and a phantom.
• Two random sampling methods were tested: Gaussian (GA) and Gaussian with Hermitian symmetry (GH). Image quality was assessed using Normalized Root Mean Squared Error (NRMSE).

Main Results:

• In basic simulations, Gaussian with Hermitian symmetry (GH) outperformed Gaussian (GA) sampling due to efficient data acquisition.
• However, in the advanced simulation incorporating phase errors from actual scanner data, GH resulted in worse NRMSE than GA.
• The presence of phase errors in the advanced method invalidated the assumptions of Hermitian symmetry, leading to reconstruction inaccuracies.

Conclusions:

• Numerical simulation using an MRI simulator with phase maps closely mimics actual MRI conditions and is valuable for validating sampling methods.
• Gaussian sampling with Hermitian symmetry is effective under ideal conditions but requires accurate phase correction for real-world MRI data.
• The study underscores the importance of accounting for phase errors in realistic MRI simulations for compressed sensing.