Cellular Structure Detection And Reconstruction For Light Microscope Images

One of the important methods for cell analysis is to detect and reconstruct cellular structure for microscopical cell images. High resolution microscopic imaging technology is needed to obtain the structure ofsome kinds of cells due to their small scale. With the development of experimental methods, especially the light microscopy image technique, it is possible to obtain image data containing those structure. Besides, the data type and capacity are increasing rapidly. However, it is still hard to achieve fast and reliable cellular structure detection and reconstruction with the large scale light microscopy image analysis.The present work proposed the algorithm of structure detection and nonspherical surface reconstruction of cells(somas) for three-dimensional light microscopy images. The specific content is as follows:In preprocessing, an improved multi-scale Laplacian of the Gaussian filter was proposed and used in combination with the Otsu's method for gray threshold segmentation and the volumetric threshold segmentation method, to achieve the structural enhancement of multi-scale cells and the segmentation of the foreground of cells and background.The results show that the improved filter has a good adaptability to the scale variation of cells, while the Otsu's method and the volumetric threshold segmentation method could meet the basic demand of foreground segmentation. The filtering operation spends the most time of the whole algorithm, and thereupon the proper filtering window is helpful to cut down the time consumption.To locate the cells, the preprocessing was followed by the distance transformation, and the regional maximum points of distance map whose jitter was eliminated were corresponding to real cellular centers. The results show that the elimination of jitter could eliminate undesirable regional maximum points which are produced by irregular structure to some extent and those center points are mainly close to real cellular centers.In order to detect the cellular surface structure and achieve the nonspherical digital reconstruction with a high degree of automation after the location, a double-boundary ray-burst sampling algorithm was proposed and used in combination with ellipsoid fitting. An operation of recovering missing cells was used to improve accuracy of reconstruction results. The results show that most reconstructed objects are closer to real cells than sphere model, and the precision are fairy high.As the computing hardware and software demand for cellular structure detection and reconstruction is increasing, it is essential to extend original research methods, and absorbing new methods from other fields or the combination of various algorithms should be the trend of research. Studies in this dissertation have added a set of alternative methods for this field.