Real-time reconstruction of sensitivity encoded radial magnetic resonance imaging using a graphics processing unit

Area of Science:

• Medical Imaging
• Computational Imaging
• Magnetic Resonance Imaging

Background:

• Real-time applications of non-Cartesian parallel MRI are limited by lengthy image reconstruction times.
• Current reconstruction times often exceed acquisition times, hindering real-time image display.

Purpose of the Study:

• To present a novel reconstruction algorithm for real-time sensitivity encoded radial imaging (radial SENSE) using commodity graphics hardware (GPUs).
• To enable dynamic adjustment of temporal resolution and interactive improvement of reconstruction quality.

Main Methods:

• Developed a GPU-accelerated reconstruction algorithm for radial SENSE.
• Utilized a golden ratio radial profile order for flexible reconstruction from varying numbers of profiles.
• Incorporated a user-adaptable regularization term for interactive quality enhancement, especially with undersampled data.
• Implemented a self-contained reconstruction process computing all necessary parameters in real-time.

Main Results:

• Demonstrated real-time reconstruction of radial SENSE images on commodity GPUs.
• Showcased the ability to adjust temporal resolution dynamically.
• Validated the algorithm's effectiveness for real-time SENSE and k-t SENSE reconstructions.
• Successfully performed offline reconstructions with interactive parameter adjustments.

Conclusions:

• The developed GPU algorithm significantly reduces reconstruction times, overcoming a key barrier to real-time non-Cartesian parallel MRI.
• The approach allows for flexible temporal resolution and interactive quality control, enhancing the utility of radial SENSE imaging.
• This method facilitates real-time imaging applications and interactive offline analysis, advancing the field of accelerated MRI.