【作者】 罗斌；

【导师】 何赛灵；

【作者基本信息】 浙江大学 ， 光学工程， 2007， 博士

【摘要】 以近红外光为探测手段的漫射光层析成像(DOT,Diffuse Optical Tomography)技术由于安全、轻便、不受电磁干扰、成本低等优点而在生物医学领域有大好的应用前景。DOT探测系统简易,成像理论也不复杂,但在许多医学应用领域(如早期乳腺癌的光学探测)至今仍未推广到临床应用。最主要的困难在图像重构上。简而言之,就是在DOT的实际应用中,要么重构图像质量太差而不能用于医学诊断,要么重构计算时间太长而无实用价值。本论文研究在DOT图像重构中使用的正问题理论模型的性能(精度和计算时间效率)以及重构算法的性能。目的是获得一套有实用价值的正问题模型和重构算法,从而在可接受的时间内得到质量合格的、能用于医学诊断的重构图像。本论文由七章组成。第1章简要介绍漫射成像的理论基础、现有的各种重构方法和研究概况。第2章提出并描述一种提高混浊介质中光子传输Monte Carlo仿真(MC仿真)的时间效率的方法。第3章研究两种版本的传统漫射模型的偏差(与MC仿真对比)。了解这些偏差是为了改进漫射模型(作为DOT重构计算中的正问题模型)以提高其精度(在不降低其时间效率的前提下)。这对于改善DOT重构图像的质量是至关重要的(因为DOT逆问题是高度病态的)。第3章还介绍求解漫射方程的有限元方法(FEM,Finite Element Method),并研究漫射边值问题中的零值边界的有效性和效率。漫射模型在介质表面附近比在介质深处有更大的偏差。第4章研究这种表面偏差并提出一种修正漫射近似法。这种修正既可用于纯粹的漫射模型,又可用于漫射模型与MC仿真或光子传输方程混合的混合模型,并且在不降低模型的计算效率的前提下显著地减小了表面偏差。因而,修正的漫射模型(纯粹的或混合的)比传统模型更适宜作DOT重构中的正问题模型。这是本论文的主要贡献(它有助于改善DOT重构图像的质量)。第5章首先提供一种在MC仿真中直接计算漫射光时间自相关函数的方法,该方法快速而精确。然后研究关于漫射光时间自相关函数的漫射模型的精度(通过与MC仿真对比)。数值实验显示,漫射模型用于计算归一化光场时间自相关函数时的精度极高(比用于计算光强时的精度高得多),几乎可看作精确模型。因而,在重构介质的动态参数(如布朗漫射系数)时,漫射模型是一个十分理想的正问题模型。第6章内容是关于DOT重构算法和重构图像质量的。首先,提出并描述一种基于有限元法和非线性共轭梯度法(NCG,Nonlinear Conjugated Gradient,一种非线性优化算法)的二步重构算法。该算法能够改善DOT重构图像的质量。然后,研究因为正问题模型的模型误差而造成的DOT重构图像质量的内在限制。此限制就是:DOT重构中逆问题的病态特性以及正问题模型的模型误差使得被重构介质的吸收和散射系数分布的细节几乎总不能被正确重构出来,即,DOT重构图像中的细节信息是不可靠的,只有粗略信息才是可靠的、可用的。第6章的研究揭示了DOT重构计算所用正问题模型的精度对于重构图像质量的极端重要性。最后一章(第7章)先简要叙述本论文相关的实验工作(包括一个荧光漫射光乳腺成像系统、一个简单实验和我们正在搭建的一个用于研究的DOT/FDOT实验系统)。最后,列出本论文后续研究的主要内容。更多还原

【Abstract】 Diffuse optical tomography(DOT)using near-infrared light is promising in many biomedical applications such as breast cancer detection and functional brain imaging since the detection using near-infrared light is harmless to biological tissues.Also the detection of DOT is convenient and anti-jamming to electromagnetic environments,and the price of the DOT system is low.Neither the detection nor the theory of DOT is complicated.However,many biomedical applications of DOT(including the optical detection of early breast cancer)are not applied to clinic yet.The primary difficult resides in the image reconstruction in practical applications,i.e.,either the quality of the reconstructed image is too bad to be used in medical diagnosis,or the computational time for the reconstruction is too long to be acceptable.The present study focuses on the performance(accuracy and computational efficiency)of the models and algorithms used in the image reconstruction of DOT.The goal is to develop efficient forward models and reconstruction algorithms in order to obtain the images whose quality are high enough to be used in medical diagnosis.The present thesis consists of seven chapters.A brief overview of the fundamental of DOT and various reconstruction methods at present is given in Chapter 1.The content of Chapter 2 is included to present a method to improve the computational efficiency of the Monte Carlo(MC)simulation of photon transport in turbid media.In Chapter 3,the deviations of two versions of the traditional diffusion model(compared with the MC simulation)are studied.Knowing the details of these deviations helps us to improve the accuracy of the diffusion models(as the forward models used in the image reconstruction of DOT)without debasing the computational efficiency of the models.And this is very important to the improvement of the quality of the reconstructed images(since the inverse problem of DOT is badly ill-posed).An introduction of the finite element method (FEM)to solve the diffusion equation is also included in Chapter 3.And the validity and efficiency of the zero boundary condition are discussed.The deviation of the diffusion model is relatively larger in the region near the surface of the medium than in the deep region.This surface deviation is studied in Chapter 4.A modification of the diffusion approximation is presented.This modification can be applied in both the pure diffusion model and the hybrid models(MC diffusion hybrid model and transport diffusion hybrid model)and significantly reduces the surface deviation without debasing the computational efficiency of the models.Therefore,the modified models are more suitable than the traditional modelsto be the forward models in DOT image reconstructions.This is the primary contribution of the present thesis.(It can improve the quality of the reconstructed images of DOT).The temporal autocorrelation of the diffuse light is also studied in this thesis.A straightforward method to calculate the correlation function in the MC simulation is present in Chapter 5.This method is fast and accurate.Also in Chapter 5 the accuracy of the correlation diffusion model is studied by comparing it with the MC simulation. Numerical experiments show that the diffusion model is really accurate to calculate the normalized temporal autocorrelation of the diffuse field(though the diffusion model is not so accurate to calculate the diffuse intensity),i.e.,the correlation diffusion model is almost exact.Therefore,the correlation diffusion model is an ideal forward model in the reconstruction of the dynamic properties(such as the Brownian coefficient)of the medium.The content of Chapter 6 is about the reconstruction algorithms and the quality of the reconstructed images in DOT.A two-step reconstruction algorithm based on the FEM and NCG(Nonlinear Conjugated Gradient,a nonlinear optimization method)is presented to improve the quality of the reconstructed images.The limitation of the reconstructed images by the forward model in the reconstructions is also studied.The limitation is that almost all the time the details of the absorption and scattering distribution of the medium are not correctly reconstructed due to the-low accuracy of the forward model and the ill-posedness of the inverse problem of the DOT imaging.Only the strong contrast across a large region and the significant characters of the reconstructed image are reliable and useful.The study in Chapter 6 reveals that the accuracyof the forward model used in the reconstruction of DOT is extremely important to the quality of the reconstructed image.In the last chapter,Chapter 7,we first give a brief description of the experiment work for this thesis,including an optical mammography system,a simple experiment,and a DOT/FDOT experiment system which we are building for research.And at last,we list the research tasks to be taken after the research of this thesis.更多还原