Segmentation And Counting Of Retina Cells In Fluorescence Microscopic Image Based On Shape Classification

Nowadays,more and more people are suffering from various ophthalmic diseases especially retinopathy.The damages of retinal tissue caused by retinal diseases are usually hard to be cured by medicine merely.The biological tissue repair based on cell transplantation technology is expected to become a powerful tool for retinal injury repair.In the cell transplantation therapy,the biomaterial as carrier must be compatible with human body.So,the cytotoxicity of the material should be evaluated.In our research,to evaluate the cytotoxicity of a polymer based biomaterial,several groups of retina cells culture experiments（with and without material）are set and the cell images of different experiments are obtained at different time points.Because of the noise and overlapping cells in the cell images,the segmentation and counting of retina cells is challenging.To improve the efficiency and accuracy of segmentation and counting of the retina cells,we propose a method based on shape classification in this thesis.The main steps of the proposed method are as follows:First,the preprocessing of the original cell images.It includes converting the original images into grayscale images,dynamic thresholding segmentation and morphological operations to reduce noise and obtain the connected components of cells for segmentation.Second,the recognition of the overlapping cells based on shape classification.A series of shape features are extracted from the connected components according to the shape characteristics of retina cells.An AdaBoost classifier is trained based on these features and used to recognize the overlapping cells.Finally,the segmentation of the overlapping cells.The bottleneck points are detected on the contours of the overlapping cells and taken as the segmentation points.The accelerated Dijkstra algorithm is then used to find the optimal connection of the pair of segmentation points and finish the segmentation.When new connected components are generated by segmentation,the above steps of overlapping cells recognition and segmentation will be repeated until no more connected components need to be segmented.The number of the final connected components is counted as the number of cells.The proposed method is tested on 50 fluorescence microscopic images.Compared with the ground truth,the average accuracy,recall rate,precision rate andF₁-measure of the proposed method are 95.52%,91.11%,88.64%and 0.895,respectively.The result shows the feasibility of the method which can be used in the evaluation of the cytotoxicity of biological materials.