4D MR phase and magnitude segmentations with GPU parallel computing.

Area of Science:

• Medical Imaging
• Computational Science

Background:

• Large four-dimensional cardiac magnetic resonance (MR) image datasets require efficient segmentation.
• Accurate segmentation of the aorta is vital for analyzing cardiac flow information.

Purpose of the Study:

• To develop and compare efficient segmentation algorithms for cardiac MR images.
• To evaluate the accuracy and speed of CPU- vs. GPU-based segmentation methods.

Main Methods:

• Proposed phase-contrast segmentation algorithms using mean-based calculations and least mean squared curve fitting on a CPU.
• Developed a graphics processing unit (GPU)-based algorithm fitting flow data to Gaussian waveforms.
• Applied level sets to magnitude images using CPU and GPU segmentations as initial conditions.

Main Results:

• CPU-based algorithms provided initial segmentations in under 10 seconds but with reduced accuracy.
• The GPU-based algorithm generated initial segmentations in 0.5 seconds.
• The GPU algorithm, combined with level sets, demonstrated superior segmentation accuracy.

Conclusions:

• GPU-accelerated algorithms offer a significant improvement in speed and accuracy for cardiac MR image segmentation.
• The proposed GPU method is a promising approach for efficient and precise aortic segmentation.