Fast 3D Ultrasound Localization Microscopy via Projection-based Processing Framework

Area of Science:

• Medical Imaging
• Ultrasound Technology
• Computational Imaging

Background:

• Three-dimensional ultrasound localization microscopy (3D ULM) offers detailed vascular visualization for improved diagnostics.
• Clinical adoption of 3D ULM is hindered by high computational demands and lengthy processing times for full 3D reconstruction.

Purpose of the Study:

• To develop a computationally efficient 3D ULM framework for in vivo imaging.
• To reduce processing time and enable potential real-time feedback in 3D ULM.

Main Methods:

• Reformulated the 3D ULM pipeline into efficient 2D operations (beamforming, clutter filtering, motion estimation, microbubble separation/localization).
• Utilized a row-column array (RCA) based system for in vivo pig kidney imaging.
• Implemented the framework on a single NVIDIA RTX A6000 Ada GPU.

Main Results:

• Achieved reconstruction of a 25*27.4*27.4 mm3 volume in 0.52 seconds (70% of acquisition time) at 400 Hz frame rate.
• Maintained ULM image quality comparable to conventional 3D processing.
• Demonstrated high quantitative agreement with conventional 3D methods (SSIM=0.93, velocity slope=0.93, R2=0.88).

Conclusions:

• The proposed 2D-operation-based framework significantly accelerates 3D ULM processing.
• This advancement shows potential for real-time feedback during scanning, enhancing robustness and workflow efficiency.
• Enables faster and more accessible high-resolution 3D vascular imaging for clinical applications.