Combined domain-decomposition and matrix-decomposition scheme for large-scale diffuse optical tomography.

Area of Science:

• Biomedical Optics
• Computational Imaging
• Medical Physics

Background:

• Image reconstruction in diffuse optical tomography (DOT) is computationally intensive.
• Large-scale DOT applications, like breast tumor diagnosis, require efficient computation and storage solutions.
• Existing methods often struggle with the complexity of 3D forward and inverse solvers.

Purpose of the Study:

• To develop a two-level image reconstruction scheme for 3D DOT.
• To combine domain decomposition (DD) for forward calculation and matrix decomposition (MD) for inversion.
• To reduce computation and storage loads for large-scale DOT problems.

Main Methods:

• A two-level scheme combining Schwarz-type DD-based forward calculation and MD-based inversion.
• Forward calculation uses a coarse-grid finite difference method, updated with a parallel DD scheme on a fine grid.
• Inversion employs wavelet-decomposition reconstruction on a coarse grid, followed by Levenberg-Marquardt least-squares with MD on a fine grid.

Main Results:

• The DD-based forward solver and MD-based inversion enable coarse-grain parallel implementation.
• Significant reduction in computation and storage requirements for large-scale DOT.
• Numerical simulations and phantom experiments validated the approach.

Conclusions:

• The proposed two-level reconstruction scheme effectively addresses computational challenges in 3D DOT.
• Matrix decomposition (MD)-based linear inversion demonstrates superiority over traditional methods like algebraic reconstruction technique.
• This approach facilitates practical applications of DOT, particularly in medical diagnostics.