Real-Time Super-Resolution Ultrasound Imaging using GPU Acceleration

• Published in: 2022 IEEE International Ultrasonics Symposium (IUS) pp. 1 - 4
• Main Authors: Praesius, Sebastian Kazmarek, Stuart, Matthias Bo, Schou, Mikkel, Dammann, Bernd, Brandenborg Sorensen, Hans Henrik, Jensen, Jorgen Arendt
• Format: Conference Proceeding
• Language: English
• Published: IEEE 10.10.2022
• Subjects: Array signal processing; Graphics processing units; Imaging; Rats; Streaming media; Superresolution; Ultrasonic imaging
• ISSN: 1948-5727
• DOI: 10.1109/IUS54386.2022.9957589

The method SUper-Resolution ultrasound imaging using the Erythrocytes as targets (SURE) is fully non-invasive, and can reliably visualize vessels with sizes down to 50 μm (1/3 of the wavelength) from a few seconds of data acquisition. Ideally, the acquisition and display should be done in seconds, but this is a challenge since the processing of SURE images is computationally demanding. Graphics Processing Units (GPUs) are specialized for high-throughput parallel processing, and in this paper it was explored whether a NVIDIA GeForce RTX 3090 GPU can enable real-time processing of SURE images, meaning the processing can keep up with the imaging rate of 417 Hz, allowing for a live video feed, similar to conventional ultrasound imaging. In-vivo data was acquired from a Sprague-Dawley rat kidney with a 168 channel GE-L8-18iD 10 MHz linear array probe connected to a Verasonics Vantage 256 scanner at a 62.5 MHz sampling rate. The GPU was used to perform beamforming, motion correction, stationary echo cancellation and peak localization. The resulting processing rate was 475 Hz for the beamforming, and 497 Hz for the proceeding processing steps, resulting in a total rate of 239 Hz. Consequently, SURE images can now be acquired in 2 seconds, and shown approximately 1 second after this, making it possible to visualize super resolution images of the microvasculature at the bedside, for immediate diagnosis of the patient.