Research Of The Light Propagation In Optical Imaging Based On Numerical And Random Statistical Methods

Molecular imaging is a promising and rapidly developing biomedical researchfield, which enables the visualization of the cellular function and the follow-up of themolecular process in living organisms noninvasively, and can be applied to the earlydisease diagnosis, tumor cell detection and drug improvement. Among molecularimaging modalities, optical molecular has become a research focus over the past yearsbecause of its significant advantages in sensitivity, cost-effectiveness and non-ionizingradiation.The study of photon transport is one of the most important research contents inoptical imaging and our work is focused on this issue. According to the analysis on theactual demand of optical imaging and the characters of the different light transportmodels, we employed the numerical and random statistic methods to research thephoton transport issue of optical imaging. The main work of this dissertation can besummarized as follows:1. Since the physical experiment is usually complicated and expensive for theoptical imaging, research methods based on simulation platforms have obtainedextensive attention. We developed a simulation method to research optical imagingthrough a soft platform named Molecular Optical Simulation Environment (MOSE). Inthis method, Monte Carlo (MC) method was used to simulate photon transport inbiological tissues. A central processing unit (CPU) parallelization strategy wasemployed to improve the computational efficiency of MC method. In this strategy, thephoton packages which need to be simulated in MC method were distributed to severalPCs which are connected by LAN with proper proportion, so the cost of the parallelcomputation system could be reduced. For the character of high resolution, thenon-contact imaging system has become quite popular recently, so the research for thephoton progress in free space is necessary. With the photon flux distribution on thesurface of the tissues which produced by MC method, we employed the hybridradiosity-radiance theorem to simulate the non-contact measurement mode of opticalimaging. A serial of simulated and physicial experiments were used to varify theefficiency of the proposed method.2. A point weighted least-squares (PWLS) meshless method is proposed to obtainthe numerical solution of the light transport in tissues. For its acceptable accuracy inthe case of high scattering and small computation cost, the diffusion equation (DE) wasthe most popular photon transport model in optical imaging. The numerical solution of DE is usually obtained by the finite element method (FEM). However, theeffectiveness of FEM is heavily affected by the finite element mesh which is used todiscretize the domain of interested and construct the approximation function.Unfortunately, the generation of high quality mesh for the three-dimensional irregulardomain with complex geometrical structure is still a challenging and time-consumingtask. The proposed method does not need any mesh to construct the approximationfunction, so that complicated and time-consuming mesh generation can be avoid.Moreover, the linear relationship between source distribution and photon flux densitywas established by minimizing the weighted residuals quadratic sum of all collocationpoints for governing equation and boundary condtion, so no numerical integral isneeded. This proposed method is suitable to be applied to the domain with complexstructure because of the pre mentioned characters. The performance of the proposedmethod was dementrate by the experiments with phantom and numerical mouse.3. A graphic processing unit (GPU) parallel accelerated algorithm was proposed toobtain the numerical solution for the radiative equation (RTE). RTE is considered asthe most accurate model for light transport in tissues. The discrete is a widely usedmethod to obtain the numerical solution of RTE. The discrete ordinates reduce the RTEto a serial of differential equations which can be solved by source iteration (SI).However, the tremendous time consumption of SI, which is partly caused by theexpensive computation of each SI step, limits its applications. Utilizing the calculationindependence on the levels of the discrete ordinate equations and spatial elements, theproposed method reduces the time cost of each SI step by parallel calculation. Thephoton reflection at the boundary was calculated based on the results of the last SI stepto ensure the calculation independence on the level of the discrete ordinate equation.An element sweeping strategy was proposed to detect the calculation independence onthe level of the spatial element. A GPU parallel frame called the compute unifieddevice architecture (CUDA) was employed to carry out the parallel computation. Thesimulation experiments, which were carried out with a PC of GTX260graphic cardand Xoen CPU, indicated that the time cost of each SI step can be reduced by morethan two orders of magnitudes with the proposed method.