Fluorescence Tomography in Small Animal Imaging using an Ultra-fast RTE Solver

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Small animal imaging, which models almost all human diseases, has been widely used in preclinical research. In this context, optical imaging has attracted great attention over past several years, among which fluorescence imaging has the superior sensitivity due to exogenous contrast agent. In particular, fluorescence tomography (FLT) enables the three-dimensional (3D) quantitative recovery of fluorescent source in a non-invasive and non-radiative manner. However, it remains difficult to improve the FLT performance for more accurate and reliable quantification. One crucial fact is the missing of an accurate and fast model for in vivo light propagation. The popular diffusion approximation often fails in small animal imaging, while radiative transfer equation (RTE) is the most accurate of realistic models. It has shown by several groups that RTE-based reconstructions offer significantly better accuracy than DA-based ones. However, the major issue preventing RTE from being popular is its unpractical computational burden (e.g., it usually takes days to complete one reconstruction on 3D mouse model). To distinguish from most groups working on RTE-based reconstructions, we have recently dedicated in the development of numerical solver of RTE to improve its accuracy, efficiency and flexibility for practical use. In this project, we propose an ultra-fast solver of RTE so that RTE-based FLT is feasible (e.g., one reconstruction takes <2 hours), for which we will develop novel scattering-adaptive computation and implement the parallelization via graphics processing unit (GPU). The proposed solver of RTE will be first-of-its-kind to the best of our knowledge. On the other hand, FLT can be further improved by synergetic combination of linear complex-source formulation, simultaneous correction of optical background and various state-of-art reconstruction techniques, such as L1-promoted sparsity, framelet-regularized smoothness, Bregman method and multilevel approach. This proposal is featured by both an ultra-fast solver of RTE and innovative reconstruction techniques. The overall goal of this proposal is to develop fast, accurate and practical RTE-based FLT for small animal imaging. We are motivated by two main independent hypotheses that (1) GPU parallelization and scattering-adaptive computation will allow the ultra-fast solver of RTE, which makes RTE-based FLT feasible; (2) linear complex-source formulation and simultaneous correction of optical background will allow further significant quantitative improvement of FLT when combined with various start-of-art reconstruction techniques. Upon completion of this project, the proposed methods will have been validated in phantom experiments and applied to small animal studies. The software will be made publicly available on web.