【作者】 谌雅琴；

【导师】 李刚；

【作者基本信息】 天津大学 ， 生物医学工程， 2005， 博士

【摘要】 生物组织化学成分无创分光检测具有极其重要的临床和现实意义。但目前该技术的基本传感理论研究还相对薄弱,存在许多问题有待解决。本文从组织光学的观点出发,并结合空间分辨稳态漫反射测量技术,深入研究了生物组织化学成分无创分光检测的基本传感理论。首先对光在生物组织中的传输理论——蒙特卡罗模拟(MCS)和漫射近似理论进行了深入、系统地研究。其中,利用考虑多层组织结构的蒙特卡罗模型MCML分析了不同组织光学特性参数对漫反射波形的影响,给出了MCS的统计特性和时间特性的具体数学描述;修正并统一了可取得最佳漫反射率预测性能的漫射模型的各重要参变量定义,并给出了漫射模型的适用范围的量化指标。其次,利用基于统计学原理的数据压缩技术——主成分分析(PCA)详细分析了MCS产生的仿真漫反射率数据与组织光学特性参数之间的一一对应关系,有效地证明了PCA确实适用于提取漫反射率数据的主要特征,给出了应用PCA得到漫反射率数据的主分量表示时应注意的问题;并针对单层和双层组织模型,分别给出了在不同测量条件下从漫反射率波形反推组织光学特性参数的可能性。然后,针对单层组织模型的光学特性参数反演问题,详细评估了目前两种主要的非线性反演算法:非线性最小二乘拟合和传统神经网络方法的预测性能和特点,其中,非线性最小二乘拟合以漫射模型作为正向光传输模型,传统神经网络则采用MCS;在此基础上,提出将基于MCS的主成分分析-神经网络(PCA-NN)方法应用于从漫反射率波形反推组织吸收系数和约化散射系数,并利用Intralipid-10%溶液实验对PCA-NN方法的有效性进行了初步实验验证。最后,将基于MCS的PCA-NN方法推广用于解决考虑多层组织结构的组织光学特性参数反演问题。具体针对上层组织厚度已知的双层组织模型的双层组织光学特性参数反演问题,提出多种基于双层组织模型所对应的漫反射率数据与上层和/或下层组织光学特性参数对应关系建立的PCA-NN及其改进算法,给出了这些算法用于从漫反射率波形反推双层组织光学特性参数的可行性及必要条件,并分析了应用这些算法时,上层组织厚度的测量不确定性对双层组织光学特性参数的预测精度的影响,为实际进行组织化学成分无创分光检测提供理论依据。更多还原

【Abstract】 Noninvasive measurement of tissue components using spectroscopy is of the utmostimportance in the clinic and practice. However, the study on the basic transductiontheory for this technology is in a relatively fundamental stage, many problemsremained unresolved. From the point of view of tissue optics, and employing thespatially resolved steady-state diffuse reflectance measurement technique, we deeplystudied the basic transduction theory for the noninvasive measurement of tissuecomponents using spectroscopy in this thesis.Firstly, we deeply and systematically studied the theories of light transport in thetissues —— Monte Carlo simulations (MCS) and diffusion approximation theory.By using the Monte Carlo modeling for multi-layered tissues (MCML), the influencesof different tissue optical properties on the diffuse reflectance profiles were analyzedand the mathematical descriptions of MCS statistical and temporal characteristicswere tabulated. In addition, we modified and unified the definitions for those keyparameters in the diffusion model to obtain the optimum predictive ability of diffusereflectance, and provided the quantitative indexes for the extent of the diffusionmodel.Then, we exploited principal component analysis (PCA), a data compressiontechnique based on statistics, to analyze the unique corresponding relationshipbetween the MCS simulated diffuse reflectance data and the tissue optical propertiesin detail, and effectively proved that PCA is feasible for extracting the maincharacteristics of the diffuse reflectance profile. We also referred to the noticeableproblems when PCA is applied to obtain the principal components of the diffusereflectance data. Furthermore, the possibilities of estimating the tissue opticalproperties from the diffuse reflectance profiles in the case of single-and two-layertissue models were demonstrated.Next, as for the inverse problem of the tissue optical properties in the single-layertissue model, we validated the predictive abilities and characteristics of the presentmost popular nonlinear inverse algorithms, nonlinear least-square fitting andtraditional neural network methods. The diffusion model was used in the nonlinearleast-square method and MCS in the neural network method. Based on the aboveresearches, a MCS-based PCA-NN method was proposed to retrieve the tissueabsorption and reduced scattering coefficients from the diffuse reflectance profile.The effectiveness of this method was preliminarily evaluated by experiments with theIntralipid-10% solution.Finally, the MCS-based PCA-NN method was extended to resolve the inverseproblem of the tissue optical properties in the multi-layered turbid media. Particularlyfor extracting the optical properties of the two layers in the two-layer tissue modelwith the top layer thickness known a priori, we proposed several PCA-NN and itsmodified algorithms which were established on the relationships between the diffusereflectance data and the optical properties of the top layer and/or the bottom layer. Thepossibility and essential conditions were also referred to for these algorithms toestimate the optical properties of the two layers from the diffuse reflectance profile.What’s more, we analyzed the influence of the measurement uncertainty of the toplayer thickness on the predictive accuracies of the optical properties of the two layers,which will provide the theoretic basis for the noninvasive measurement of tissuecomponents using spectroscopy in practice.更多还原