Deep Learning in Ultrasound Localization Microscopy: Applications and Perspectives.

Area of Science:

• Biomedical Imaging
• Super-resolution Microscopy
• Medical Ultrasound

Background:

• Ultrasound localization microscopy (ULM) offers deep in vivo vascular imaging beyond diffraction limits.
• ULM utilizes microbubble tracking for anatomical and hemodynamic microvascular insights.
• Deep learning (DL) is increasingly applied to ULM challenges like denoising and aberration correction.

Purpose of the Study:

• To comprehensively review deep learning applications in ultrasound localization microscopy.
• To focus on DL approaches for sparse microbubble distributions.
• To analyze variations in datasets, tasks, and DL methodologies.

Main Methods:

• Review of existing literature on deep learning in ULM.
• Categorization of DL approaches based on targeted ULM tasks (e.g., denoising, localization, flow estimation).
• Detailed examination of microbubble localization enhancement techniques.

Main Results:

• Deep learning methods often surpass conventional techniques in ULM image quality and processing speed.
• DL's robustness to high microbubble concentrations can shorten ULM acquisition times.
• Significant diversity exists in DL dataset constitution, task focus, and network architectures.

Conclusions:

• Deep learning shows significant promise for advancing ultrasound localization microscopy.
• Further research is needed to address current limitations and fully realize DL's potential in ULM.
• DL applications are crucial for overcoming hurdles in ULM clinical translation.