用于提高超分辨定位显微成像空间分辨率的数据处理方法研究

发布时间：2018-03-27 05:36

本文选题：超分辨定位显微成像　切入点：数据处理　出处：《华中科技大学》2016年博士论文

【摘要】：超分辨定位显微成像将荧光显微成像的空间分辨率提高了超过一个数量级,使生命科学领域的研究者可以从分子水平观测生物精细结构与功能。为了提升该方法的测量精确度,拓宽该技术的应用范围,需要研究进一步提高超分辨定位显微成像空间分辨率的方法。其中,提升超分辨定位成像数据处理方法的精度是提高该技术空间分辨率的直接、有效手段。本文通过分析超分辨定位成像中的关键数据处理步骤,并针对提升数据处理精度所遇到的难点,从高密度分子定位、漂移校正、分子筛选以及图像可视化方面进行了研究：(1)高精度快速高密度分子定位方法。本文分析了限制当前高密度定位方法计算性能的瓶颈,通过优化初始模型预估与多分子拟合的非凸优化问题求解,开发了高精度、少计算量的高密度分子定位方法PALMER (PArallel Localization of Multiple Emitters via Bayesian information criterion Recommendation),并基于ImageJ图像处理软件开发了此方法的实用性插件,解决了当前高密度分子定位算法无法兼容快速与高精度性能的问题。通过仿真与实验验证,证实PALMER方法具有高的定位精度与分子检测率,且定位速度比著名的高密度定位算法DAOSTORM快一个数量级。比起常规的稀疏分子数据采集与定位,通过高密度分子采集与PALMER方法的结合,可以将奈奎斯特空间分辨率提升近四倍。(2)高精度漂移校正方法。本文深入分析了互相关漂移校正方法的数学模型,并利用定位数据集中不同时间的子数据集均描述同一目标结构的冗余性特征,开发了基于冗余互相关计算的漂移校正方法(Redundant Cross Correlation, RCC)。仿真与实验验证的结果显示,RCC算法具有高精度、高鲁棒性的特点。RCC方法一方面可以在处理此前互相关漂移校正方法不适用的低信号数据集时,获得纳米量级的高精度校正结果;另一方面还可以提升超分辨重建图像的有效空间分辨率：相比其他的互相关漂移校正方法,RCC算法将空间分辨率提升了约10%。(3)高性能分子筛选以及图像可视化方法。本文基于拓展定位数据集有效数据维度的基本思想,利用对目标结构的先验知识从定位数据集中提取结构各向异性特征,开发了基于各向异性系数的分子筛选方法。仿真与实验数据分析结果显示,该分子筛选方法可以有效滤除定位数据集中的背景噪声,包括其他方法所不能滤除的非特异性簇状聚合杂点,从而显著提高定位数据集的信噪比。同时,本文讨论了各向异性特征可以与高斯渲染图像可视化方法相结合,用于非线性地增强超分辨重建图像中的空间结构信息。综上,本文利用超分辨定位显微成像的固有特征,发展了一系列高精度的数据处理方法,进一步提高了超分辨定位显微成像的空间分辨率,有望促进超分辨定位显微成像在精细结构与功能解析中的应用。
[Abstract]:Super-resolution localization microscopic imaging improves the spatial resolution of fluorescence microscopic imaging by more than one order of magnitude, enabling researchers in the field of life sciences to observe biological fine structures and functions at the molecular level. In order to widen the scope of application of this technique, we need to study the method of further improving the spatial resolution of micro-imaging of super-resolution positioning, among which, the improvement of the precision of super-resolution positioning imaging data processing method is the direct way to improve the spatial resolution of the technology. In this paper, we analyze the key data processing steps in super-resolution positioning imaging, and aim at the difficulties in improving the accuracy of data processing, from high-density molecular positioning, drift correction, Molecular screening and image visualization are studied. (1) High precision, fast and high density molecular localization methods. This paper analyzes the bottleneck that limits the computational performance of current high density localization methods. By optimizing the initial model prediction and solving the non-convex optimization problem with multi-molecular fitting, a high precision is developed. PALMER (parallel Localization of Multiple Emitters via Bayesian information criterion criterion recommendation), a high-density molecular location method with less computation, is developed based on ImageJ image processing software. The problem that the current high density molecular localization algorithm can not be compatible with fast and high precision is solved. The simulation and experimental results show that the PALMER method has high localization accuracy and molecular detection rate. The localization speed is one order of magnitude faster than the famous high-density localization algorithm DAOSTORM. Compared with the conventional sparse molecular data acquisition and localization, the combination of high-density molecular acquisition and PALMER method, Nyquist spatial resolution can be improved by nearly four times. 2) High precision drift correction method. The mathematical model of cross correlation drift correction method is deeply analyzed in this paper. The redundant features of the same target structure are described by using the sub-datasets of the location dataset at different times. A drift correction method based on redundant cross-correlation calculation is developed. The simulation and experimental results show that the proposed algorithm has high accuracy. The characteristic of high robustness. On the one hand, RCC method can obtain high precision correction results in nanometer order when dealing with low signal data sets which were not applicable to the previous cross-correlation drift correction method. On the other hand, it can improve the effective spatial resolution of super-resolution reconstructed image. Compared with other cross-correlation drift correction methods, RCC algorithm can improve the spatial resolution by about 10%. This paper is based on the basic idea of extending the effective data dimension of location data set. Using the prior knowledge of target structure to extract the anisotropic feature of structure from the location data set, a molecular screening method based on anisotropy coefficient is developed. The results of simulation and experimental data analysis show that, The molecular screening method can effectively filter the background noise in the location dataset, including the non-specific cluster-shaped aggregated clutter that cannot be filtered by other methods, thus significantly improving the signal-to-noise ratio of the location dataset. At the same time, In this paper, we discuss that the anisotropic feature can be combined with the visualization method of Gao Si rendering image to enhance the spatial structure information in the super-resolution reconstruction image. A series of high-precision data processing methods have been developed to further improve the spatial resolution of super-resolution positioning microscopic imaging, which is expected to promote the application of super-resolution positioning microscopic imaging in fine structure and functional analysis.
【学位授予单位】：华中科技大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：Q-336

【相似文献】