Compressed sensing reconstruction for whole-heart imaging with 3D radial trajectories: a graphics processing unit...

Area of Science:

• Magnetic Resonance Imaging
• Medical Imaging
• Image Reconstruction

Background:

• Whole-heart coronary MRI acquisition is limited by long scan times.
• Isotropic 3D radial trajectories enable undersampling for faster volumetric data acquisition.
• Compressed sensing (CS) reconstruction accelerates imaging and improves quality but has high computational costs.

Purpose of the Study:

• To present a parallelized, graphics processing unit (GPU)-implemented iterative CS reconstruction for 3D radial MRI.
• To evaluate the speed and efficacy of the GPU-accelerated CS reconstruction compared to traditional methods.
• To assess CS's ability to reduce artifacts and enhance image quality in 3D whole-heart coronary MRI.

Main Methods:

• Developed a parallelized iterative CS reconstruction algorithm utilizing a commercial GPU.
• Compared GPU implementation execution times against a C++ version.
• Assessed image quality and artifact suppression in phantom and whole-heart coronary MRI datasets using undersampled 3D radial acquisition and CS reconstruction.

Main Results:

• The GPU implementation of CS reconstruction achieved a 34-54 times speed-up compared to the C++ implementation.
• CS reconstruction improved image quality, enhancing vessel sharpness and reducing noise-like artifacts compared to the 3D gridding algorithm.
• Demonstrated the efficacy of CS in suppressing streaking artifacts in 3D whole-heart coronary MRI.

Conclusions:

• GPU-accelerated CS reconstruction significantly reduces computational overhead for 3D radial MRI.
• This approach enables faster, higher-quality whole-heart coronary MRI acquisition.
• The developed method offers a practical solution for accelerating MRI scans while maintaining diagnostic image quality.