One of the most frequent ways to interact with the surrounding environment occursudas a visual way. Hence imaging is a very common way in order to gain informationudand learn from the environment. Particularly in the field of cellular biology, imagingudis applied in order to get an insight into the minute world of cellular complexes. Asuda result, in recent years many researches have focused on developing new suitableudimage processing approaches which have facilitates the extraction of meaningfuludquantitative information from image data sets. In spite of recent progress, but due toudthe huge data set of acquired images and the demand for increasing precision, digitaludimage processing and statistical analysis are gaining more and more importance inudthis field.udThere are still limitations in bioimaging techniques that are preventing sophisticatedudoptical methods from reaching their full potential. For instance, in the 3DudElectron Microscopy(3DEM) process nearly all acquired images require manual postprocessingudto enhance the performance, which should be substitute by an automaticudand reliable approach (dealt in Part I). Furthermore, the algorithms to localize individualudfluorophores in 3D super-resolution microscopy data are still in their initialudphase (discussed in Part II). In general, biologists currently lack automated and highudthroughput methods for quantitative global analysis of 3D gene structures.udThis thesis focuses mainly on microscopy imaging approaches based on MachineudLearning, statistical analysis and image processing in order to cope and improve theudtask of quantitative analysis of huge image data. The main task consists of buildinguda novel paradigm for microscopy imaging processes which is able to work in anudautomatic, accurate and reliable way.ududThe specific contributions of this thesis can be summarized as follows:ud• Substitution of the time-consuming, subjective and laborious task of manualudpost-picking in Cryo-EM process by a fully automatic particle post-pickingudroutine based on Machine Learning methods (Part I).ud• Quality enhancement of the 3D reconstruction image due to the high performanceudof automatically post-picking steps (Part I).ud• Developing a full automatic tool for detecting subcellular objects in multichannelud3D Fluorescence images (Part II).ud• Extension of known colocalization analysis by using spatial statistics in orderudto investigate the surrounding point distribution and enabling to analyze theudcolocalization in combination with statistical significance (Part II).udAll introduced approaches are implemented and provided as toolboxes which areudfree available for research purposes.
展开▼
