Ultrafast Ultrasound Localization Microscopy by Conditional Generative Adversarial Network.

Area of Science:

• Biomedical imaging
• Medical ultrasound technology
• Microscopy

Background:

• Ultrasound localization microscopy (ULM) visualizes microvasculature at subwavelength resolution.
• Ultrafast ULM faces challenges in balancing acquisition time, data processing, and imaging resolution.
• Deep learning (DL) shows promise for accelerating ULM, but still requires significant data frames.

Purpose of the Study:

• To develop a novel deep learning-based method (ULM-GAN) for ultrafast ultrasound localization microscopy.
• To enable super-resolution image reconstruction from minimal ultrasound data, reducing acquisition time.

Main Methods:

• A modified conditional generative adversarial network (cGAN) framework, termed ULM-GAN, was developed.
• ULM-GAN reconstructs super-resolution images from averaged low-resolution ultrasound data (l frames).
• Performance was evaluated using numerical simulations and phantom experiments.

Main Results:

• ULM-GAN achieved approximately 40-fold reduction in data acquisition time and 61-fold reduction in computational time compared to Gaussian fitting.
• Spatial resolution was maintained, as indicated by the resolution scaled error (RSE).
• Phantom experiments demonstrated ultrafast acquisition (~0.33 s) and processing (~0.60 s).

Conclusions:

• ULM-GAN offers a significant speed improvement for ultrasound localization microscopy.
• The method enables ultrafast imaging, making it suitable for observing rapid in vivo biological activities.
• ULM-GAN addresses key challenges in implementing ultrafast ULM.