Automatic Segmentation And Recognition Of Microscopic Images In Cytology Slide

Cervical cancer is a common malignancy in women,and whether it can be cured mainly depends on the early detection and early diagnosis.With the rapid development of computer technology,it is applied for the analysis of microscopic images of cervical Pap smear to assist the pathological diagnosis of cervical cancer.Microscopic images of cervical Pap smear have the characteristics as inconspicuous contrast between nuclei and background,uneven dyeing and overlapped nuclei so that the accurate segmentation of nuclei is difficult.The classification and recognition of cervical images have some problems as one texture method for feature extraction and no feature selection for the classification so that the quantitatively analysis of microscopic images of cervical Pap smear is challenging.For the above existing problems in the microscopic images of cervical Pap smear,this paper studied and proposed a two-scale automatic segmentation algorithm based on multi-feature Mean-shift clustering with elastic mathematical morphology and classification and recognition algorithms based on multi-type feature extraction and selection.The research work of this paper mainly includes:(1)A two-scales automatic segmentation algorithm based on multi-features Mean-shift clustering with elastic mathematical morphology was proposed to solve the problem of the segmentation of nuclei in microscopic images of cervical Pap smear.The algorithm used top-bottom hat transform to enhance the contrast between nuclei and background.Mean-shift clustering based on the gray and spatial location information was applied to achieve the localization nuclei.The roundness and area threshold were used to determine the overlapped of nuclei and the elastic mathematical morphology algorithm was applied to realize the segmentation of nuclei with different degree of overlap.(2)Aiming at the feature extraction of cervical Pap smear images,this paper studied and proposed a multi-types feature extraction algorithm of cervical nuclei.The feature extraction algorithm could extract morphological features,color features and statistical texture features.In addition,the Gabor features analyzed from time domain and frequency domain and the MRF features described as a structural relation of the pixels themselves in the images were also extracted.Various types of features were used for the classification and recognition of microscopic images of cervical Pap smear.(3)A filter feature selection algorithm based on p-value and a wrapper feature selection algorithm based on GA were proposed to reduce the classification time and improve the classification performance of cervical images.Those features extracted by proposed feature extraction algorithm were used for the classification and recognition of normal cervical images,uninvolved cervical images and abnormal cervical images,which verified the effectiveness of the feature extraction,feature selection and classification algorithms proposed in this paper.The proposed automatic segmentation and recognition algorithms for cervical Pap smear images provided a new solution,and provide a new theoretical and method basis for diagnosis of cervical cancer based on cervical cytology images.It can be applied in the clinical application of cervical cytology images analysis.The proposed method has certain theoretical significance and application value.