Research On Segmentation And Recognition Technology For Biomedical Image

As the accelerated pace of modern life and sports entertainment, human peripheral nerve injury is very common, but nerve injury has still been surgical clinical problem.During peripheral nerve repair surgery, accurately docking neurological beam of the same nature is the best, but human neural structures is tiny, substance is transparent and shape is changing, these physiological characteristics can not meet the doctor qualitative docking directly by the naked eye, the effect after nerve injury repair and regeneration is not ideal, it has yet to find a satisfactory efficacy repair method for human nerve. With the rapid development of computer technology, the study of the computer three-dimensional reconstruction and visualization of the structure of the human body advance rapidly.The internal structure image got by computer three-dimensional technology can reflect the internal structure and variation of the nerve at the view of its entire length. This has an important significance of counseling treatment to peripheral nerve repair surgery. Currently, the study of the three-dimensional visualization of peripheral nerve focuses on the preparation and registration of the image, and it has made a good solution. By means of off-the-shelf3D reconstruction software on the market, three-dimensional reconstruction has been effectively resolved after segmented by hand. The automatic segmentation techniques and automatic recognition technology of nerve bundles are rare reported at home and abroad, the automatic segmentation and recognition of the nerve bundles still being investigated.The computer image processing technology is used to segment nerve bundles and nerve fibers positive stains in nerve bundles, and pattern recognition knowledge is used to design classifier to identify the positive stains. According to the rate of the nerve fibers in the nerve bundles to identify the nerve bundles function.This thesis studies the following major work:1. Analyse the lack of the generic Karnorvsky-Roots staining neural slice on the image features.A new useful re-staining method which is Karnovsky-Roots-Toluidine Blue-Water-Soluble Scarlet is introduced to nerve bundles segmentation and recognition.At the same time, and analyse staining characteristics of the image.2. The segmentation of nerve bundles.As the characteristics of the nerve sectioning microscopy images is not conducive to the classic method of segmentation, spatial fuzzy C-means clustering segmentation method (SFCM) is designed, segment nerve bundles in the S component of the HSI color model, the histogram Fuzzy C-Means is used to determine the best classification number and the initialization parameter for SFCM performance improvement. This method not only overcome the shortcomings of the original image with colors overlap and low target contrast, but also avoid the characteristics,which are true color, color mixing and uneven illumination,of original image from the S component diagram.3. The segmentation of nerve fibers positive stains within the nerve bundle.Firstly, design watershed segmentation algorithm with inner and outer marker to segment all the nerve fibers positive stain areas.And then, set adhesions stain conditions through the area and circular degree to filter none adhesions stains.At last, Fuzzy C-Means is used to segment adhesion stains.4. Recognition and statistics of nerve fibers positive stains within the nerve bundle. BP neural network is used as a classifier to achieve classification and statistics of positive nerve fibers stains.Extract the morphological characteristics and color characteristics of the nerve fibers positive stains. After training and testing by samples extracted by hand, a designed three-layer BP neural network is constructed for classifying unknown nerve fibers positive stains.Next the statistics of all kinds of nerve fibers positive stains is obtained.5. Identify the function of the nerve bundl. According to the ratio calculated for each nerve fiber within the nerve bundle to identify the function of the nerve bundle.The studied spatial fuzzy C-means clustering segmentation method based HSI color space model better meets the requirements of complex staining nerve sectioning microscopy image segmentation.Compared with the traditional segmentation method, the split nerve bundles more complete and the edge more continuous. The technical application of pattern recognition classifier to identify the function of the nerve bundle, not only improve the efficiency of neural three-dimensional reconstruction, but also improve the accuracy of recognition.It has certain theoretical significance and practical value.