Optimization of k-space trajectories for compressed sensing by Bayesian experimental design.

Area of Science:

• Medical Imaging
• Computational Imaging
• Magnetic Resonance Imaging (MRI)

Background:

• Nonlinear sparse Magnetic Resonance Imaging (MRI) reconstruction requires efficient k-space sampling strategies.
• Current sampling methods may not be optimal for advanced reconstruction algorithms.
• Bayesian experimental design offers a framework for optimizing data acquisition.

Purpose of the Study:

• To formulate k-space sampling optimization for nonlinear sparse MRI reconstruction as a Bayesian experimental design problem.
• To develop an efficient algorithm for optimizing MRI acquisition trajectories.
• To evaluate the performance of optimized trajectories against standard designs using clinical data.

Main Methods:

• Approximated Bayesian inference using standard signal processing primitives.
• Developed an efficient optimization algorithm applicable to Cartesian and spiral trajectories.
• Tested the algorithm on clinical resolution brain image data acquired from a Siemens 3T scanner.

Main Results:

• The developed optimization algorithm efficiently determines k-space sampling trajectories.
• Automatically optimized trajectories significantly improved image quality compared to low-pass, equispaced, and randomized designs.
• The approach provides insights into nonlinear design optimization for MRI.

Conclusions:

• A novel Bayesian experimental design approach effectively optimizes k-space sampling for nonlinear sparse MRI reconstruction.
• Optimized trajectories lead to superior image quality in clinical brain imaging.
• This method offers a more efficient and effective strategy for MRI data acquisition.