High-Speed GPU-Based Fully Three-Dimensional Diffuse Optical Tomographic System.

Area of Science:

• Biomedical optics
• Medical imaging
• Computational modeling

Background:

• Diffuse optical tomography (DOT) is a powerful imaging technique but computationally intensive.
• The ill-posed nature of 3D DOT algorithms presents significant computational challenges for real-time applications.

Purpose of the Study:

• To develop a high-speed, fully 3D graphics processor unit (GPU)-based system for diffuse optical tomography.
• To significantly reduce the execution time of 3D DOT algorithms through algorithmic and hardware acceleration.

Main Methods:

• Implemented a GPU-based system utilizing Broyden approach for Jacobian matrix updates and a multinode, multithreaded GPU architecture with CUDA.
• Developed two GPU implementations: a C-based program (C GPU) and a MATLAB-based program (MATLAB GPU).
• Employed the finite element method (FEM) for forward computation, discretizing the domain into tetrahedral elements (up to 66514).

Main Results:

• Achieved a reconstruction time of 0.52 seconds per iteration for a C-based GPU program with 14610 elements.
• The MATLAB-based GPU program achieved a reconstruction time of 0.86 seconds per iteration.
• The system demonstrated a maximum reconstruction rate of 2 frames per second.

Conclusions:

• The developed GPU-based system dramatically accelerates 3D DOT computations.
• This advancement holds potential for enabling real-time or near-real-time diffuse optical tomography applications.
• The study validates the effectiveness of GPU acceleration and algorithmic improvements for complex inverse problems in imaging.