Fast Simulation of Dynamic Ultrasound Images Using the GPU

Area of Science:

• Medical Imaging
• Computational Science

Background:

• Simulated ultrasound data is crucial for echocardiography analysis method development.
• Existing simulation methods can be too slow for complex phantoms with many point scatterers.

Purpose of the Study:

• To present efficient parallelized CPU and GPU implementations of the COLE ultrasound simulation algorithm.
• To improve the speed of dynamic simulations with moving point scatterers for echocardiography.

Main Methods:

• Developed parallelized CPU and GPU versions of the COLE algorithm.
• Stored dynamic point scatterer trajectories as spline curves in GPU memory to minimize data transfers.
• Implemented a phase-based subsample delay technique to reduce flickering artifacts.

Main Results:

• Achieved significant speedups on a high-end GPU: two orders of magnitude over parallel CPU, three orders over original COLE, and 27,000 times over Field II.
• Demonstrated efficient simulation of dynamic sequences, including B-mode and tissue Doppler for a full cardiac cycle.
• Successfully eliminated flickering artifacts using the proposed subsample delay technique.

Conclusions:

• Highly efficient parallelized CPU and GPU implementations of the COLE algorithm dramatically accelerate ultrasound simulation.
• The optimized approach enables faster development and validation of quantitative echocardiography analysis methods.
• The open-source release aims to foster wider adoption and further advancements in ultrasound simulation technology.