Related Experiment Video
Updated: Dec 31, 2025
06:56
Human Fetal Blood Flow Quantification with Magnetic Resonance Imaging and Motion Compensation
Published on: January 7, 2021
2.7K
Fast GPU Implementation of a Scan-Specific Deep Learning Reconstruction for Accelerated Magnetic Resonance Imaging.
Chi Zhang1,2, Sebastian Weingärtner1,2,3, Steen Moeller2
1Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, MN.
IEEE International Conference on Electro Information Technology : [Proceedings]. IEEE International Conference on Electro Information Technology
|January 2, 2020
Summary
This study accelerates the RAKI MRI reconstruction algorithm using GPU parallel programming and CPU multiprocessing. These strategies achieve over 60x speedup, making fast MRI reconstruction practical.
Area of Science:
- Medical Imaging
- Computer Science
- Artificial Intelligence
Background:
- Rapid Magnetic Resonance Imaging (MRI) is crucial for clinical applications.
- RAKI (Reconstruction Algorithm with Knowledge-based Intelligence) offers satisfying results for highly accelerated MRI.
- RAKI's implementation is computationally intensive due to multiple convolutional neural networks.
Purpose of the Study:
- To develop acceleration strategies for RAKI implementation.
- To leverage GPU parallel programming and CPU multiprocessing for RAKI optimization.
- To enable practical, time-efficient RAKI reconstruction for clinical use.
Main Methods:
- Implemented acceleration strategies for RAKI using GPU parallel programming.
- Limited iteration numbers in the network training stage for optimization problems.
- Parallelized RAKI training tasks using CPU multiprocessing to utilize GPU resources fully.
Main Results:
- Achieved a speedup of over 60x compared to conventional, sequential RAKI implementation.
- Maintained visually satisfying reconstruction results despite optimization efforts.
- Reduced RAKI reconstruction time to minutes, facilitating practical application.
Conclusions:
- Accelerated RAKI implementation is feasible and highly effective.
- GPU parallel programming and CPU multiprocessing significantly enhance RAKI performance.
- The optimized RAKI algorithm can be readily integrated into practical clinical workflows.
