Enhancing Ultrasound Molecular Imaging: Toward Real-Time RPCA-Based Filtering to Differentiate Bound and Free...

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Biology

Background:

• Ultrasound molecular imaging (UMI) utilizes microbubbles as contrast agents to detect cancer biomarkers.
• A key challenge in UMI is differentiating bound microbubbles (molecular signal) from free-floating ones (background noise).

Purpose of the Study:

• To develop and validate a fast GPU-based method using robust principal component analysis (RPCA) for distinguishing bound from free-floating microbubbles in UMI.
• To assess the accuracy and computational efficiency of the RPCA method for real-time applications.

Main Methods:

• Simulations were used to evaluate RPCA accuracy (Dice coefficient, RMS error) based on the number of frames.
• Experiments with stationary and flowing microbubbles in phantoms validated the method.
• The RPCA method was applied to in vivo data from mouse models of breast cancer using targeted and non-targeted microbubbles.

Main Results:

• RPCA with 20 frames achieved a Dice score of 0.95 and a computation time of 0.2 seconds in phantom studies, suggesting real-time potential.
• In vivo, RPCA using 20 frames showed good agreement with Destruction-Templated Emulation (DTE), yielding a Dice score of 0.82.
• The method successfully distinguished targeted microbubbles from background noise in preclinical cancer models.

Conclusions:

• The proposed GPU-based RPCA method offers a fast and accurate solution for separating bound microbubbles in ultrasound molecular imaging.
• This technique has significant potential for improving early cancer detection and molecular characterization using UMI.
• The findings support the feasibility of real-time implementation for enhanced clinical translation of UMI.