Research On The Method Of Nucleus Segmentation And Classification In Cervical Cell Clusters Images

With the wide application of computer technology in different fields,image processing and pattern recognition technology combined with pathological knowledge are used to assist doctors to complete diagnosis,which has become one of the research hotspots in the field of machine vision.Cervical cancer is the most common female malignancy with definite etiology.Its occurrence and development is a gradual process of evolution,and the cure rate of early and middle stage lesions is very higher.Therefore,regular screening and early detection of some precancerous lesions can effectively avoid the risk of cervical cancer.In recent years,many computer-aided cervical cell image analysis algorithms for different types Pap smear images have been designed by scholars at home and abroad.In this paper,we take the nucleus in cervical cell clusters images as the research object,and combine the steps of image segmentation,feature extraction and classification and recognition to realize the screening of diseased cervical cell nucleus.The specific work is as follows:1.Firstly,adaptive threshold Otsu algorithm combined with mathematical morphological operations is used to locate cell clusters in Pap smear.Then,the regions of interest(ROIs)are extracted from the whole image by selective search.Through the above two steps,ROIs in the cell cluster can be obtained.Finally,a feature named MAXSection is defined,which combined with Back Propagation(BP)neural network is used for nucleus screening from all ROIs.2.On the basis of the extracted nucleus regions,different contrast nucleus images are divided into high contrast group and low contrast group based on gray histogram distribution.Nucleus images in high contrast group are segmented by Chan-Vese(CV)model directly.While nucleus images in low contrast group are enhanced by a method of Canny operator combined with mathematical morphology operation before segmentation.3.Nucleus morphological parameters based on the segmentation results as well as the MAXSection features and texture information based on gray level co-occurrence matrix are combined with BP neural network to classify the cell nuclei into normal class and abnormal class,thus completing the diseased nuclei screening.The experimental results show that the nucleus image region extraction proposed in this paper for cluster cervical cell image largely guarantees the complete acquisition of the nuclei of cell clusters in whole Pap smear.In the segmentation task,the algorithm can not only segment the high contrast nucleus images with high accuracy,but also the low contrast images on the basis of contrast enhancement.In classification tasks,the parameters of nucleus area based on the results of selective image enhancement are more accurate and estimated height and width based on MAXSection features are easier to obtain.The highest test accuracy of the classifier based on BP neural network can reach 100%.This research can provide a reference for the study of computer-assisted nucleus screening for cluster cervical cell images.