Physical Parameter-Guided Deep Learning Ultrasound Localization Microscopy Framework Based on Diffusion Model.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Ultrasound Technology

Background:

• Deep learning shows promise for Ultrasound Localization Microscopy (ULM), a key microvascular imaging technique.
• ULM's application is limited by a lack of diverse, high-quality ground truth data for training deep learning models.
• Quantitative microbubble imaging data is scarce, hindering advanced deep learning approaches in ULM.

Purpose of the Study:

• To introduce a novel framework for synthesizing microbubble ultrasound images.
• To address the critical data gap in training deep learning models for ULM.
• To enable accurate deep learning model development for ULM without experimental ground truth.

Main Methods:

• Developed a Physical parameter-guided Diffusion (PgD) framework.
• Synthesized microbubble images using transducer specifications and acoustic waveform parameters.
• Utilized synthetic images as a training dataset for deep learning-based ULM.

Main Results:

• Generated synthetic data demonstrated high fidelity to experimental ground truth (SSIM 0.97).
• Deep learning-based ULM trained with synthetic data outperformed conventional methods.
• Achieved 5-10 μm improvement in spatial resolution with fewer frames.

Conclusions:

• The PgD framework offers a robust and generalizable solution for ULM data synthesis.
• Validated the framework's effectiveness in enhancing deep learning-based ULM.
• Highlights the framework's significant value for advancing ULM applications.