Improving sub-pixel accuracy in ultrasound localization microscopy using supervised and self-supervised deep...

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Imaging

Background:

• Ultrasound localization microscopy (ULM) reconstructs microvascular structures and measures blood flow using microbubbles in contrast-enhanced ultrasound (CEUS) images.
• Accurate microbubble localization is crucial for ULM fidelity due to the large size of CEUS bubble traces compared to microbubbles.
• Existing methods struggle with noisy data and accurately pinpointing microbubble centers for high-resolution imaging.

Purpose of the Study:

• To develop advanced deep learning methods for improved microbubble detection and localization in CEUS data.
• To enhance the spatial resolution of ULM for better visualization of microvascular networks.
• To increase the accuracy of velocity profile measurements in both micro- and macrovessels.

Main Methods:

• Introduction of a supervised super-resolution blind deconvolution network (SupBD-net) based on residual learning.
• Development of a new loss function for a self-supervised blind deconvolution network (SelfBD-net) to handle unknown bubble trace morphologies.
• Comparison of SupBD-net and SelfBD-net against existing deep learning and blind deconvolution techniques using synthetic data.

Main Results:

• SupBD-net achieved the lowest error in bubble center location (<0.1 λ), outperforming other methods.
• SelfBD-net maintained accuracy (<0.15 λ) for unknown bubble trace morphologies where supervised methods failed.
• Both SupBD-net and SelfBD-net demonstrated superior performance in separating closely located bubbles and microvessels, and in accurate velocity profile measurements.

Conclusions:

• The proposed residual learning-based methods significantly improve the spatial resolution and accuracy of ULM.
• SupBD-net and SelfBD-net enable distinguishing neighboring microvessels separated by 0.15 λ, surpassing previous techniques.
• These advancements hold promise for more detailed and accurate microvascular imaging in clinical and research applications.