The Research On Medical Microscopy Cell Image Segmentation

Digital image processing is widely applied in medical image analysis, particularly in medical microscopic image processing, such as tissue cell classification, chromosome analysis, tissue recognition and cell mutation research, etc. Cell area segmentation&extraction is the premise of judgment abnormalities in quantity, morphology of cells, which is directly related to the reliability of the diagnosis. Therefore, the study of medical cell image segmentation has important theoretical significance and practical value. The original images analysised in this dissertation are neuron stem cells and white blood cells acquired by Karin Althoff and Johan Degerman of Chalmers University of technology in Sweden.The main work in this thesis can be summarized as follows:(1) Review the technology of cell image segmentation. Some traditional methods and new technologies are introduced, and analyses their own characteristics.(2) To solve the problem of less detection precision in order morphology, the thesis proposes the edge detection method based on multiple order morphology. First, two-dimensional histogram oblique segmentation method is adopted to locate the image edge region, then the edge region is processed with fuzzy enhancement, and finally the multiple order morphology edge detection algorithm uses different direction linear structuring elements and two kind percentiles to obtain edge image. The results of experimentation prove that the method is valid and practical for segmentation and extraction of cell area.(3) To overcome the limitation of obtaining the driving force by setting thresholds manually in the fast level set, the thesis proposes the fast level set partition algorithm without setting thresholds. In this new approach, the external velocity acquired from image data indirectly, and therefore, the driving force does not come from thresholds any more. The improved algorithm preserves the global segmentation characteristic of C-V model, and adopts fast level set based on two lists to realize the curve evolution. This method is proved by experimental results to have higher computering speed and better denoising performance than traditional level set method.(4) To solve the problems of limited suitable object and high computation complexity in the traditional chain code mode, the thesis proposes two improved chain code modes for oval cell and irregularly shaped cell. For approximate oval cells, ECCC(Eleven components Chain Code) is proposed. For irregular shape cells, FPCC(Feature Preferences Chain Code) is proposed. Experimental results show that ECCC and FPCC perform better than the traditional chain code segmentation mode.(5) To overcome the limitation of failing to adequately utilize gray level information of an image in isoperimetric cut, the thesis proposes the multilevel thresholding method based on isoperimetric. The proposed method uses isoperimelric ratio of the isoperimetric cut as criterion for threshold selection and finds multiple thresholds by a fast and effective iterative scheme, simplifies computation of isoperimetric ratio, and introduces a way of automatically determining cluster number to adaptively choose reasonable threshold number. Experimental results on aseries of cell images show the effectiveness of our multilevel method.(6) Finally, the algorithm of medical microscopy cell image segmentation presented in this thesis is estimated and summaried.