Research On General Algorithm Platform For Microscopic Cell Images

With the continuous development of modern microscopic imaging technology,researchers can collect thousands of high-resolution cell images every day.For the processing and analysis of these images,the existing cell image analysis software based on local computing resources can use computer image processing technology to automatically identify and quantitatively analyze cell images.However,these software have problems such as difficulty in data sharing,lack of support for team collaboration,and difficulty in customizing algorithms for new cell types.Therefore,taking into account the needs of the development of biomedical research,this thesis studies a Browser/Server(B/S)architecture-oriented algorithm platform for microscopic cell images,which provides researchers with a feature-rich,easy-to-learn,and flexible Cell image processing software tools that are changeable and support team collaboration and user-defined algorithms.(1)Abstract the needs of researchers into the basic functional modules of the system,and design the hierarchical structure and database of the system according to the functional modules.The platform supports teamwork,provides complete team authority management and project schedule management functions,and users can form teams to share data algorithms for collaborative processing.(2)The platform integrates traditional algorithms and deep learning algorithms for cell images,and encapsulates them as algorithm components with adjustable parameters that users can call dynamically.The platform supports user-defined algorithm solutions with a certain image processing foundation,and the algorithm scheme can be built by dragging and connecting the algorithm components of the package,so as to achieve rapid algorithm verification.Custom algorithms can be saved for analysis,re-use,and team sharing.(3)Verify the function of the platform through the detection experiment of cilia and nuclei in the fluorescence microscopy image and the segmentation experiment of mitochondria and membrane structure in the electron microscopy image.The detection experiment uses traditional image processing algorithms,completes algorithm construction and testing on the custom algorithm interface,and completes data analysis through the platform's visualization tools.The segmentation experiment uses a deep learning algorithm,in which the mitochondrial segmentation experiment annotates small batch data,completes the model training on the deep learning training platform provided by the platform,and applies the trained segmentation model to subsequent data processing.