Research On Key Technology In Automated Pathological Scanning And Analysis System

Pathological scanning and analysis system can automatically acquire the digital images of massive samples and do a preliminary analysis,which helps doctors to diagnosis more effi-ciently.Cervical screening is an important application of the pathological scanning and analysis system.Due to the low efficiency of manual screening,the prevention of cervical cancer and the popularity of cervical screening require automatic cervical screening methods.Deep research of automatic pathological scanning and analysis system is conducive to enhancing the effect of computer-aided diagnosis and analysis,improving the efficiency of cervical screening,which is of great significance to popularize the cervical screening.The key algorithms of pathological scanning and analysis system are studied in this the-sis,and the Feulgen-Eosin stained cervical cells are mainly used for experiment and analysis.The application of MOG algorithm and GMG algorithm as background subtraction method is studied in image illumination balancing part,in order to facilitate the subsequent analysis and processing.Aiming at obtaining the full image of samples,this thesis analyzes a variety of image registration and splicing methods.Registration method based on SURF or ORB features is also improved for dual camera image,thus we can get the corresponding relationship between the monochrome image and color image.For the extraction of nuclei,at first the region of interest was obtained and the possible overlapping nuclei were decomposed by Otsu threshold and mor-phological operations,and then the morphological features,optical density features and texture features were calculated in nuclei,convex hull of nuclei and peripheral regions,finally random forest is used to accurately distinguish valid nuclei from impurities.In the image enhancement part,this thesis presents a probability model to fuse the monochrome images and color images,which enhances the nuclei regions.The cervical cells are quantitatively analyzed by a nuclei segmentation followed with the calculation of DNA index.In addition,a polar coordinate edge fitting method which uses stable edges to interpolate the whole boundaries is proposed for rapid segmentation of cytoplasm and thus the nucleus cytoplasm ratio can be calculated.Experiments result shows that the proposed segmentation method is not only very fast,but also have a good effect for both independent cells and mild overlapped cells.