Real-time Image Processing And Quality Control For High- Throughput Super-resolution Localization Microscopy

Super-resolution localization microscopy(SRLM)achieves nanoscale spatial resolution with a relatively simple setup and user-friendly experimental procedures,thus becomes an important tool for biomedical researches.In the past,the applications of SRLM are typically limited to a single field of view(FOV)that is smaller than a mammalian cell.In recent years,researchers are trying to apply SRLM into high-throughput microscopy applications.By imaging hundreds or even tens of thousands of FOVs,high-throughput localization microscopy can provide plentiful temporal and spatial information,thus is convenient to optimize experimental conditions,detect events occurred with low probability,and provide unbiased information for cell structures.However,because SRLM requires large data volume and high computational intensity,the large FOV and high activation density used in high-throughput localization microscopy make the computational intensity far exceeds the real-time processing capability of existing data processing methods.In other words,the existing data processing methods are not able to provide automatic image acquisition and analysis that are required in high-throughput localization microscopy.Moreover,current data processing methods cannot guarantee a uniform image quality among the large number of FOVs in high-throughput localization microscopy.To overcome these challenges,this thesis aimed to develop a series of data processing methods that are capable of providing real-time and high precision image processing and quality control for high-throughput localization imaging.The main works are shown as follows.(1)Real-time sparse emitter localization algorithm.This thesis modified the mathematical model of maximum likelihood estimation(MLE).By combining several other optimization strategies and graphics processing unit(GPU)acceleration,this thesis is able to provide one order of magnitude faster localization speed than the existing fastest MLE localization algorithm without sacrificing localization precision.This localization speed satisfies the real-time processing requirement of super-resolution localization microscopy with sCMOS camera,where the FOV is 200 ?m × 200 ?m and the exposure time is 10 ms.Moreover,this thesis developed a user-friendly plug-in,called QC-STORM,which provides a complete image processing workflow including real-time image preprocessing and molecule identification,super-resolution image rendering,and statistical information analyzing.Results showed that QC-STORM achieves two orders of magnitude faster overall image processing speed than similar software.(2)Real-time high-density emitter localization algorithm.This thesis developed a realtime weighted MLE localization algorithm and MLE localization algorithms for two and three emitters.These algorithms were further combined with a divide and conquer strategy to localize region-of-interest(ROI)with real-time.As a result,this thesis achieved similar localization precision and molecule detection rate,but three to four orders of magnitude faster image processing speed than existing software.Results proved that real-time processing of high activation density imaging at sCMOS camera is possible for 100 ?m × 100 ?m FOV and 10 ms exposure time.(3)Real-time quality control method.This thesis modelled the random activation of localization imaging by Poisson distribution,and generalized the integer dimension model of sample structures into fractional dimension.This treatment enabled a hyper-parameter free method for real-time calculating localization density and Nyquist resolution.Combining localization precision and Nyquist resolution calculation,this thesis developed a real-time method,termed ROMP,for real-time calculation of convolved spatial resolution.ROMP can process the imaging data in real-time from a representative sCMOS camera with 100 ?m × 100 ?m FOV and 10 ms exposure time.Furthermore,based on a real-time localization density feedback control and ROMP resolution monitoring,this thesis achieved homogenous spatial resolution across multiple FOVs.In summary,this thesis developed a series of real-time image processing and quality control methods for high-throughput localization microscopy.These methods can reduce the difficulties in high-throughput localization imaging,and thus pushes the applications of this promising technology in biomedical applications.