Data Processing Methods For Improving The Spatial Resolution Of Super-resolution Localization Microscopy

Super-resolution localization microscopy (LM) has improved the spatial resolution of fluorescence microscopy by more than an order of magnitude, and thus has been an important tool for various biology application scenarios to reveal the working mechanisms of cellular processes at molecular level. Although the spatial resolution of LM has achieved at tens of nanometer, further improvement in spatial resolution would greatly enhance the measurement accuracy and extend the application scope of LM. Particularly, the performance of data processing methods in LM is a determining factor of spatial resolution. By improving the accuracy of post-acquisition data processing methods, the spatial resolution of LM can be effectively improved. However, the demands in improving data processing accuracy will bring in new challenges that the complexity and difficulty of data processing will significantly increase. Against this challenge, a series of method have been developed by employing novel algorithms in this dissertation, on the basis of key data processing steps to generate a super-resolution image.(1) High accuracy and fast high-density molecule localization method. We analyzed the bottleneck issue in limiting the computational efficiency of current high-density localization algorithms, optimized the multi-emitter model estimation and localization steps, and thus developed a high accuracy high-density molecule localization algorithm with relatively low computational amount. By combining this algorithm with the GPU platform, we presented the PALMER (PArallel Localization of Multiple Emitters via Bayesian information criterion Recommendation) method and its Image J software-based plug in, which successfully solves the problem that the high localization performance is not compatible with fast localization speed in existing methods. The simulation and experimental results indicate that the PALMER method both has high localization performance and fast localization speed. The localization speed has been improved by an order of magnitude compared to the DAOSTORM method. Meanwhile, compared to the traditional LM which is based on sparse emitter data acquisition and localization, high-density LM based on high-density emitter data acquisition and the PALMER method allows a Nyquist resolution gain of 4within the same data acquisition time.(2) High accuracy drift correction method. We deeply analyzed the mathematical model in localization events-based cross-correlation drift correction method. By utilizing the redundancy characteristic between sub-localization datasets, which exactly describe the same underlying structure, we developed redundant cross-correlation drift correction method (RCC). The simulation and experimental results indicate that the RCC method has high accuracy and robust performance. The major advantage of RCC is the robustness when the number of localization events is low, under which conditions that the traditional cross-correlation drift correction methods are not applicable. The RCC method has allowed us to significantly improve the effective resolution of the final super-resolution image. Compared with other cross-correlation drift correction methods, the RCC method provides an improvement in spatial resolution of-10%.(3) High-performance localization event filtering and image visualization method. Based on the idea of introducing new dimensionality of localization dataset, we quantified the structural anisotropy information of localization dataset using prior knowledge and thus developed an anisotropy coefficient-based localization event filtering method (SALEF). The simulation and experimental results indicate that the SA_LEF method can be comprehensively filter out the background localization events, including the non-specific labels appeared as clusters which traditional filtering method fails to process. The SA_LEF method can be effectively improve the signal-to-noise ratio of the localization dataset. Meanwhile, we discussed the structural anisotropy can be combined with the Gaussian image rendering method to non-symmetrically enhance the spatial structure in the final super-resolution image.In summary, we present a series of high-performance algorithms based on the inherent feature of LM. The presented methods further enhanced the spatial resolution of LM and will promote the wide spread use of LM in the study of structure and function of fine cellular components.