Parallel finite element reconstruction for spectrally-solved bioluminescence tomography

In Bioluminescence Tomography (BLT), a priori information is necessary to obtain unique BLT reconstruction. When using the explored a priori information, such as complex heterogeneous model and multispectral measurements, the reconstruction speed especially in whole-body small animal molecular imaging becomes slow. Because of the flexibility of the finite element method (FEM), reconstruction algorithms based on it are extensively developed. Fully 3D BLT reconstruction though is challenging. Although the computer hardware technology is rapidly developed, BLT reconstruction time becomes long, even impossible in a single PC when fine discretization is used to obtain the desirable spatial resolution. In this paper, a fully parallel 3D finite element reconstruction algorithm is proposed for spectrally-resolved BLT reconstruction. In this algorithm, anatomical information and multispectral measurements can be easily used to improve BLT reconstruction. Specifically, for parallelization, the reconstruction domain is partitioned into several subdomains corresponding to the number of nodes of the cluster. Establishment of a parallel linear relationship between the measured data and the unknown source distribution and optimization is performed to improve the reconstruction speed. In a numerical reconstruction, the effectiveness and potential is verified with a large-volume homogeneous phantom.