Trajectory optimized NUFFT: Faster non-Cartesian MRI reconstruction through prior knowledge and parallel...

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Image Reconstruction
• Computational Imaging

Background:

• Non-uniform fast Fourier transform (NUFFT) is crucial for reconstructing images from non-Cartesian MRI data.
• Current NUFFT methods face computational bottlenecks due to complex data access patterns, limiting the use of advanced acquisition techniques.

Purpose of the Study:

• To develop a faster and more accurate NUFFT algorithm tailored for specific MRI trajectories and GPU architectures.
• To address the computational limitations of existing NUFFT methods for non-Cartesian MRI.

Main Methods:

• Developed TRON (TRajectory Optimized NUFFT), a customized NUFFT algorithm optimized for radial trajectories and GPU hardware.
• Benchmarked TRON's speed and accuracy using the Shepp-Logan phantom and whole-body continuous golden-angle radial MRI datasets.

Main Results:

• TRON demonstrated a 6-30x speed improvement over competing methods, varying by dataset.
• TRON achieved superior accuracy compared to other tested NUFFT implementations.

Conclusions:

• Specializing NUFFT for specific trajectories, like radial, yields substantial performance gains.
• TRON offers a highly efficient and accurate solution for non-Cartesian MRI reconstruction and can be extended to other trajectories.
• The TRON code is publicly available for download and further development.