Segmentation And Recognition For Three-dimensional Lfuorescence Images Of T Cells

Through the use of fluorescence images of cells under a microscope, we canautomatically analyze and identify the status of the cells, this technology has animportant significance in the field of medical image HCS (High-content screening).First, we need to split cells as foreground objects, then we get the binary image.Second, we need to achieve the separation of the adhesion of cells, and finally get asingle cell region. Identification of cells refers to the statistics and analysis of thecharacteristics of the single cell area by cell division, allowing the computer todetermine the current state of the cell automatically.Presently, most of research about division and identification for cells on thetwo-dimensional cell image, but the actual distribution of the cells is inthree-dimensional space, two-dimensional images miss part of the information,leading to kinds of problems of the segmentation result. Recently, obtainingthree-dimensional cell image in the medical imaging device becomes possible,therefore, segmentation and recognition on three-dimensional cell image becomemore and more important.This paper does the research on the segmentation and recognition of thethree-dimensional fluorescence images of cell protein, and automatically traces thecells in activated state.In this paper, I do research about the conventional technology of the imagesegmentation, and propose an improved algorithm on the segmentation andrecognition on three-dimensional cell image.This paper studies the conventional two-dimensional image segmentationalgorithm and analyses their feasibility and advantages, finally, chooses the watershedalgorithm for three-dimensional space. Then we compare the merits of theconventional methods of binarization. Because the fluorescent noise of the image isserious, we can’t get an ideal result by using the conventional methods. In order tosolve this problem, this paper proposes a self-adapting level-set method to achieve thebinarization. And the experimental results show that this method is effective. Usingthe conventional watershed method on markers will lead to the problem of over-segmentation. To solve this problem, this paper presents a process using inscribesphere to get the markers, which improve the precision of the segmentation of thethree-dimensional cell image. At last, we can recognise the cells in activated state, by using the standard deviation and entropy from the brightness histogram of the regio nof each cell.