Research On Feature Analysis And Automatic Recognition Method Of Cervical Cell Image

Cervical cancer is the second most common malignancies in women,account ing for approximately one-tenth of the mortality rate of cancers.According to the statistics of WHO,there were 518 thousand new cases and 273 thousand deaths all over the world in 2015.In 2015,there were 98.9 thousand new cases and 30.5thousand deaths in China,while there were 12.9 thousand new cases and4.1thousand deaths in the United States.It is estimated that there will be 187 thousand new cases in China by 2050.Therefore,it is urgent to find a suitable screening method for cervical cancer in China.The recognition technology of cervical cell image is a new method of cervical cell recognition in recent years.This method overcomes the defects in the traditional manual screening method,such as high cost,large workload,reliability and accuracy,etc.First,the exfoliated cells were produced by liquid based thin film production technique,and then the cells were captured and analyzed by the industrial camera.It not only can be used to detect abnormal epithelial cells,but also can be applied to the screening and diagnosis of cervical cancer.In this paper,the cervical cell image recognition technology is used to assist the pathology.The ultimate goal is to identify the abnormal cervical epithelial cells.This work can reduce the workload of doctors,and reduce false negative and false positive.Our method of cervical cell image recognition mainly includes four stages: cervical cell image acquisition,cervical cell image preprocessing and segmentation,cervical cell image feature extraction and cervical cell image classification.We improved the recognition method of cervical cell image as following:In the classification of cervical cell image: We proposed two methods based on the characteristics of cervical materials.The first classification method firstlydivided the cervical cell image into epithelial cells,lymphocytes,neutrophils and garbage cells.Then it divided epithelial cells into normal and abnormal epithelial cells.This method was applied to the SMMCFBCCI classifier.The second classification methods divided the cervical cell image into normal epithelial cells,abnormal epithelial cells,neutrophils,lymphocytes and garbage cells.This method was applied to the PMMCFBCCI classifier.In the feature selection of cervical cell image: We built NF sets and PF sets based on the previous studies and the characteristics of cervical cytology.NF is a set of common features with a total of 22 dimensions,which is defined according to the pathological features of cervical cells.PF is the potential feature sets with a total of 14 dimensions,which includes the features traditional method does not take into account.Then,the Relief F algorithm was used to select the 24 dimensional features with high correlation.In the design of cervical cell image classifier: we proposed SMMCFBCCI classifier and PMMCFBCCI classifier,which used multiple classifiers fusion method.SMMCFBCCI classifier is a multi-classifier fusion method based on two-level cascade.The first level classifier makes use of C4.5 to realize rough classification.The second level classifier achieves fine classification using LR.The PMMCFBCCI is a fusion classifier based on the majority voting method.It first achieves the prediction results by using NB,C4.5 and KNN classifier.Then majority voting method is used to obtain the final results.