A Study On Efficient Method For Optical Tomography

As an important part of the molecular optical imaging technique,optical molecular imagingtechnique has the advantages of low cost,high sensitivity,high specificity,and non-invasive compared with other imaging techniques.The reconstruction of internal light source distribution of the organism by using optial tomography technique can give the quantitative result of the labeled cells in the internal body.The research of optical tomography focuses on the modeling of light propagation in biological tissues and the accurate reconstruction of the light source.There are still some problems that the calculations of the forward model and inverse problem are inefficient and complex on the research of optical tomography.The fast modeling and calculation of the optical tomography with high efficiency are the hotspots of the optical molecular research.The paper is focused on the research of light propagation and light source reconstruction.We developed molecularoptical simulation environment and proposed many parellel acceleration algorithms for the study of light propagation in complex biological tissues.We also proposed the fast reconstruction method forbioluminescence tomography(BLT)based on the multi-atlas registration and adaptive hexahedral voxeldiscretization to reduce the complex procedure of BLT.Themain content of the paper is introduced as follows.First,we developed a software platform for the simulation of light propagation in turbid media named the“Molecular Optical Simulation Environment(MOSE)”.MOSE is able to simulate light propagation in both turbid media with complicated structures and free space based on the Monte Carlo(MC)method and Lambertian source theory respectively.MOSE supports many simulation models of optical imaging,including BLT,fluorescence molecular tomography(FMT),and diffuse optical tomography(DOT).The graphic user interface of MOSE provides a friendly interface and a 3D visualization of the simulation results,which makes MOSE a very useful tool for studying light propagation.In the accuracy validation experiment,the normalized flux density between the simulation results and the real experiment have good agreement.Second,we proposed two parallel acceleration frameworks for MC simulation of light propagation,which are based on multicore acceleration and GPU acceleration respectively.The photons in MC simulation are independent of one another,which makes it suitable for parrellel acceleration.We implement multicore acceleration of MC simulation with OpenMP technique because OpenMP provides a high-level abstract description of parallel algorithms.Considering the pervasive application of GPU acceleration,we implement the GPU acceleration with compute unified device architecture(CUDA).The storage format,data structure and data transmission are optimized with the feature of GPU hardware.The random number generator which is run on GPU is also optimized.The simulation result shows that both the parallel acceleration frameworkshave large advantages over the traditional MC simulation software.Third,simplified spherical harmonics approximation(SP_N)equations have been widely used in modeling light propagation in biological tissues.However,with the increase of the order N,the computational burden would severely aggravate.In this study,a GPU accelerated framework was proposed for SP_N equations(termed as the GPU-based method),which facilitated its applications in the three dimensional optical imaging.The assembling and inverse calculation of system matrix are both accelerated by GPU-based method which is implemented by CUDA.The SP_N equations were solved with the finite element method whose matrices were stored with a compressed sparse row(CSR)format.The solution of SP_N equations was converted into the problem of a sparse linear system which was solved by the parallel conjugate gradient(CG)algorithm implemented on GPU.The performance of the GPU-based method was evaluated by comparing with the conventional CPU computation(CPU-based method).The results showed that an approving acceleration capability was obtained for the GPU-based method.The acceleration capability of GPU-based method was strongly affected by the size of mesh.The better speed-up ratio was obtained with the increase of the tetrahedral elements count in the mesh,with the best one being up to 25.In addition,the influence of threads distribution on the acceleration capability of GPU-based method was also investigated.The result shows that GPU-based method has a high speed up ratio while maintaining the accuracy.Finally,we present a fast reconstruction method for BLT based on the multi-atlas registration and adaptive hexahedral voxelsdiscretization.The aim of this method is to reduce the complexity of the data processing procedure for BLT by using a hybrid Optical/CT system.We adapt the multi-atlas registration algorithm to estimate the organ positions of the mouse instead of manual segmentation.The registration results indicate that the proposed method based on multi-atlas registration can estimate the distribution of organs with great efficiency and high accuracy.In order to avoid the discretization of the tissues into tetrahedral meshes,the hexahedral voxels are adaptively generated from the segmented CT volume.The adaptive hexahedral voxels for BLT reconstruction can not only save the storage space but are also more efficient in data processing compared with tetrahedral meshes.The reconstructed results demonstrate that the proposed method can be employed to reconstruct the bioluminescent source effectively and accurately.