Rapid gridding reconstruction with a minimal oversampling ratio

Area of Science:

• Medical Imaging
• Computational Science
• Signal Processing

Background:

• Magnetic resonance image reconstruction from non-Cartesian data is typically solved using convolution interpolation (gridding).
• Gridding parameters, like oversampling ratio and kernel width, allow trade-offs between accuracy and computational resources.
• Existing methods often use an oversampling ratio of two, which can be computationally intensive.

Purpose of the Study:

• To investigate significant reductions in computation memory and time for magnetic resonance image reconstruction.
• To maintain high accuracy while optimizing gridding parameters.
• To develop an improved method for designing convolution kernels and presampling techniques.

Main Methods:

• Derived a simple equation for optimal Kaiser-Bessel convolution kernel design based on oversampling ratio and kernel width.
• Evaluated the impact of kernel presampling using linear interpolation.
• Developed a new method for selecting optimal presampled kernels.
• Implemented a minimal oversampling ratio (1.125 to 1.375) instead of the typical ratio of two.

Main Results:

• Achieved significant reductions in computation memory (one-third) and time (one-eighth) for 3-D reconstruction.
• Maintained the same level of accuracy compared to using an oversampling ratio of two.
• Demonstrated that linear interpolation for presampling adds negligible error.
• Showcased the effectiveness of the optimized Kaiser-Bessel kernel and presampling method.

Conclusions:

• A minimal oversampling ratio combined with an optimally designed, presampled convolution kernel offers substantial computational savings for magnetic resonance image reconstruction.
• The developed methods provide a more efficient approach to gridding without compromising image quality.
• This optimization is crucial for accelerating MRI workflows and reducing hardware demands.