Study On Multi Parameters Fusion Recognition Of Cervical Cancer Cells

Statistics show that the incidence rate of cervical cancer ranks fourth in the world.The early screening is of great significance for reducing the incidence rate and mortality rate of cervical cancer and saving the lives of patients.At present,the early screening of cervical cancer mainly depends on manual work.Pathologists need to look for cervical cancer cells on a smear containing thousands of cervical cells by naked eye.In large hospitals,pathologists have to deal with hundreds of similar cell smears every day,which is of great risk of missed diagnosis.To solve this problem,this paper proposed a new method of cervical cancer cell recognition,established a multi parameters fusion discriminant model including 75 characteristic parameters,and carried out the related research on cervical cancer cell recognition using machine learning and deep learning,which lays the foundation for the development of intelligent cervical cancer cell recognition system.The specific research contents are as follows:(1)A multi parameter fusion recognition model of cervical cells was established.The different characteristics of different types of cervical cells were analyzed.In order to fully reflect these characteristics in the recognition model and describe the characteristics of different types of cervical cells,three kinds of parameters including geometry,texture and chromaticity were summarized.Firstly,the recognition model reduces the dimension of the three kinds of parameter information extracted from the cell image to remove the noise information and improve the information density.Then,the classification algorithm in machine learning is used to judge the category of cervical cells according to the parameter information after decomposition.Finally,the specific discrimination method of multi parameter fusion recognition model of cervical cancer cells was analyzed and determined.(2)The automatic segmentation of nucleus,cytoplasm and background in cervical cell image was realized.1834 cervical cell images were used to train the neural network.In order to solve the problem of different sizes of cervical cell images used in this paper,some parameters of neural network are optimized.After training,the neural network can automatically segment the nucleus,cytoplasm and background in the target image.(3)Feature extraction and dimension reduction of cervical cells were carried out.Because the cervical cell image segmented by neural network may have tiny holes,Open CV was used to conduct the close operation on segmented image to smooth the contour edge and fill the tiny holes in the foreground area.The contour detection function of image was realized.Aiming at the problem that the contour detection may fail in some images,the contour detection process was optimized.For the extracted feature data,the results of four different decomposition algorithms were compared to determine the specific decomposition algorithm used in this paper and the dimension of the feature after decomposition.(4)The optimal combination algorithm was established,and the classification effect was verified by the cervical cell images from the Second Hospital of Dalian Medical University.Building single algorithm classifiers using K-Nearest Neighbors,Decision Tree,Random Forest,Support Vector Machine,Logistic Regression and Naive Bayes algorithm.Integrate these single algorithm classifiers using soft voting and hard voting methods.The false negative rate was used as a performance measure to test the classification results,and the optimal combination algorithm of cervical cells was found.In the cervical cell images from Second Hospital of Dalian Medical University,the false negative rate was 1.57%.