超分辨定位成像中的快速数据处理方法研究

发布时间：2018-04-19 05:07

本文选题：超分辨定位成像 + 荧光分子定位　；参考：《华中科技大学》2014年博士论文

【摘要】：超分辨定位成像技术可实现高达20nm的空间分辨率,可以从分子水平研究细胞内的复杂工作机制,已经逐渐成为生命科学研究领域里必不可少的研究工具,但其缺点是成像速度较慢(通常是分钟量级)。因此,近年来人们致力于研究快速成像探测技术以及高密度分子定位技术,希望在不牺牲成像视场和空间分辨率的条件下,提高超分辨定位成像的时间分辨率。但是,快速成像探测技术会产生巨大的数据流和海量的数据,对数据传输、数据存储以及数据分析都形成了很大的挑战。同时,高密度分子定位技术需要使用复杂的分析模型,极大地加重了数据分析的负担。本文将针对以上问题,从高效的数据处理算法和先进的计算架构两条途径着手解决超分辨定位成像中的快速数据处理问题。主要工作包括： (1)快速稀疏分子定位方法。利用光学成像系统点扩散函数的辐射对称性,开发了快速、高精度的稀疏分子定位算法MrSE。仿真和实验结果表明,MrSE算法的定位精度可以接近理论极限,而定位速度比传统高精度算法快1000多倍。通过优化基于MrSE算法的数据分析流程,并结合GPU并行处理架构,实现了整个数据分析过程的GPU并行化处理,使数据处理速度提高到650Mpix/s,满足了快速探测对数据实时处理的需求。 (2)高效数据削减方法。利用超分辨定位成像中分子的稀疏性,开发了一种高效的数据削减算法。仿真和实验结果表明,在常规的稀疏分子定位成像中,此算法可以在不影响超分辨重建质量的同时,将数据量削减约20倍。进一步将此算法与探测器中的FPGA嵌入式计算平台相结合,实现了高达532Mpix/s的嵌入式数据削减,显著降低了快速探测带来的数据传输和数据存储的压力。 (3)快速高密度分子定位方法。通过对高密度分子图像进行去重叠,开发了可以对重叠分子进行精确定位的快速高密度定位算法。仿真和实验结果表明,该算法可以显著提高超分辨定位成像技术的空间分辨率与时间分辨率,其效果与传统高密度分子定位算法DAOSTORM相近,但算法复杂度却降低了两个数量级。基于此算法,进一步开发了专用的硬件处理器POWERs。POWERs可以独立完成数据处理任务,缓解了计算机系统的压力。其数据处理速度可达200Mpix/s,满足了常规高密度分子定位成像对数据在线处理的需求。本文利用超分辨定位成像技术所具有的特征,发展了一系列高效的数据处理方法,通过进一步结合先进的计算架构,解决了超分辨定位成像中的快速数据分析、数据传输以及数据存储的问题,对快速成像探测以及高密度分子定位技术的应用与推广具有积极意义。
[Abstract]:Super-resolution localization imaging technology can achieve spatial resolution of up to 20nm, and can study the complex working mechanism of cells from the molecular level. It has gradually become an indispensable research tool in the field of life science research.But the drawback is that the imaging speed is slow (usually the order of minutes).Therefore, in recent years, people have been devoted to the research of rapid imaging detection technology and high-density molecular positioning technology, hoping to improve the time resolution of super-resolution localization imaging without sacrificing the imaging field of view and spatial resolution.However, rapid imaging detection technology will produce huge data flow and huge amount of data, which poses a great challenge to data transmission, data storage and data analysis.At the same time, the high density molecular localization technology requires the use of complex analytical models, which greatly increases the burden of data analysis.Aiming at the above problems, this paper will solve the problem of fast data processing in super-resolution localization imaging from two ways: efficient data processing algorithm and advanced computing architecture.The main tasks include:1) a fast sparse molecular localization method.Based on the radiative symmetry of the point diffusion function of the optical imaging system, a fast and high precision sparse molecular localization algorithm, MrSE, is developed.The simulation and experimental results show that the accuracy of Mr SE algorithm is close to the theoretical limit, and the localization speed is more than 1000 times faster than that of the traditional high precision algorithm.By optimizing the data analysis flow based on MrSE algorithm and combining with the GPU parallel processing architecture, the GPU parallelization of the whole data analysis process is realized. The speed of data processing is increased to 650 Mpixel / s, which meets the demand of fast detection for real-time data processing.2) efficient data reduction method.An efficient data reduction algorithm is developed based on the molecular sparsity in super-resolution localization imaging.Simulation and experimental results show that the proposed algorithm can reduce the amount of data by about 20 times without affecting the quality of super-resolution reconstruction in conventional sparse molecular localization imaging.Furthermore, the algorithm is combined with the FPGA embedded computing platform in the detector to realize the embedded data reduction up to 532Mpix/s, which reduces the pressure of data transmission and data storage brought by the fast detection.Rapid high density molecular localization method.A fast high density localization algorithm is developed by deduplication of high density molecular images.Simulation and experimental results show that the proposed algorithm can significantly improve the spatial and temporal resolution of the super-resolution localization imaging technology, and its effect is similar to that of the traditional high-density molecular localization algorithm (DAOSTORM), but the complexity of the algorithm is reduced by two orders of magnitude.Based on this algorithm, a special hardware processor, POWERs.POWERs, is further developed to accomplish data processing independently and relieve the pressure of computer system.Its data processing speed can reach 200Mpixel / s, which meets the requirement of conventional high-density molecular positioning imaging for data on-line processing.In this paper, a series of efficient data processing methods are developed based on the characteristics of super-resolution localization imaging technology. By further combining with advanced computing architecture, the fast data analysis in super-resolution positioning imaging is solved.The problems of data transmission and data storage are of great significance to the application and popularization of fast imaging detection and high density molecular localization technology.
【学位授予单位】：华中科技大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：R310

【共引文献】