Computational Specific Microscopy Studies For Morphological Analysis Of Subcellular Structures

The intracellular subcellular structure plays an important functional role in life activities,and the study of its morphological function is of great significance for analyzing the life process and the occurrence of important diseases.Current dynamic studies of subcellular structures rely on fluorescent fluorescence microscopy,but the additional stress on cells caused by photobleaching and phototoxicity of fluorescent dyes,makes it difficult to study long-term dynamic processes in the living cell.The phase-based label-free imaging developed in recent years can successfully visualize these subcellular structures without disturbing their normal physiological state,but label-free images also lacks specificity,making it difficult to automate analysis the specific target structures.This hinders its application to the study of dynamic processes in subcellular structures.To address these issues,in this thesis we proposes a computational specific microscopy technique which is suitable for long-term subcellular structure morphological analysis and takes the analysis of mitochondrial network as an example to perform our study.Mitochondria provide energy for cells,and their network structure is complex and sensitive.The nucleus controls the expression of genetic information,and they are related to all life processes in the cell.Computational specific microscopy is able to obtain specific morphological information of mitochondria and nucleus in a gentle label-free method,and its image resolution and contrast allow accurate morphological analysis studies of complex subcellular structures.First,we used a recently developed high-contrast phase microscope to obtain a panoramic view of the cell,which contains rich physical information of subcellular structures such as mitochondria and nucleus.A trained information extraction neural network is then used to obtain mitochondriaspecific and nucleus-specific images from the unlabeled images.Our method is able to resolve individual mitochondria and complex 3D mitochondria network structures with sufficient resolution and accuracy for downstream morphological analysis.We also validated the accuracy and generalization ability of this method by comparing the information extraction results of computational specificity under different morphological cells,different cell line data and different imaging system data.Since the results of computational specificity are derived from endogenous information of the cell,its signal intensity does not correlate the staining result.This is in contrast to fluorescence-based approaches,where heterogeneous staining may cause researchers or morphological analysis tools to select only mitochondria with high signal values,while mitochondria with low signal values are missed or fail to be analyzed.This reflects the unbiased analytical advantage of the computational specificity approach over the fluorescence approach in that it does not suffer from different image intensity due to heterogeneous dye absorption.In the experiment on mitochondrial dynamics,we found that not every cell responds to stimuli in the same way,which underscoring the importance of using a computationally specific approach based on endogenous signals to study the time course of mitochondrial dynamics.Computationally specific also enables simultaneous information extraction of multiple subcellular structures,such as mitochondria and nucleus.By analyzing the correlation distribution of mitochondria and nucleus in dynamic experiments,we have discovered additional dynamic change processes.In this thesis we demonstrate the effectiveness,robustness,and utility of computational specific in multiple ways,providing an attractive alternative to study subcellular structural dynamics with the potential to link subcellular structure behavior to function in a more reliable way than traditional fluorescence-based methods.