Nonlinear image reconstruction algorithm for diffuse optical tomography using iterative block solver and automatic mesh generation from tomosynthesis images

In this paper, we report a nonlinear 3D image reconstruction algorithm featuring 3D finite element forward modeling with an iterative multi-right-hand-side solver and an automatic mesh generation technique for efficient geometry modeling. The forward mesh was generated based on 3D tomosynthesis images acquired simultaneously with optical measurements. An efficient iterative solver based on a QMR algorithm with the capacity of solving multi-RHS was used to enhance the computational efficiency. The mesh generation algorithm was developed based on a moving mesh process and is able to generate high-quality mesh with low computational complexity. In addition, an approximated adjoint method was used to form the Jacobian matrix for the inverse problem. The performance of the was satisfactory performances in numerical simulations. For a typical reconstruction, the run-time was under 5 minutes on a Pentium-based PC. It is worth mentioning that the mesh generation module not only works for binary 2D or 3D image stacks, but also for any other binary description of the object, which makes it generalizable to many other potential applications.