Monte Carlo-based fluorescence molecular tomography reconstruction method accelerated by a cluster of graphic...

Area of Science:

• Biomedical Imaging
• Optical Physics
• Computational Science

Background:

• High-speed fluorescence molecular tomography (FMT) reconstruction in 3-D heterogeneous media remains a significant challenge in diffusive optical imaging.
• Current methods are often computationally intensive, limiting their clinical applicability.

Purpose of the Study:

• To develop a fast and accurate FMT reconstruction method for 3-D heterogeneous media.
• To leverage Monte Carlo (MC) simulations and graphics processing units (GPUs) for accelerated image reconstruction.

Main Methods:

• Developed a GPU-accelerated MC simulation using the Message Passing Interface (MPI) standard for parallel computation of Green's functions.
• Implemented a load-balancing strategy to optimize computational efficiency.
• Utilized the Fréchet derivative to form a Jacobian matrix for fluorochrome distribution reconstruction.

Main Results:

• Achieved FMT reconstruction in 10 minutes using a 6-GPU cluster, a substantial improvement over the 6-hour reconstruction time on multi-CPU nodes.
• Demonstrated high accuracy and suitability for heterogeneous media with refractive-index-unmatched boundaries.

Conclusions:

• The GPU cluster-accelerated MC-based FMT method offers a reliable and efficient approach for high-speed 3-D imaging.
• This advancement has the potential to significantly impact preclinical and clinical applications of FMT.