Aided Diagnosis Of Cervical Cancer Lesions Based On Machine Learning

Cervical cancer is one of the most common gynecological malignancies.The colposcopy can observe the surface of the cervix through the microscopic biopsy and pathological examination to diagnose lesions,improve the diagnostic rate and ensure that patients get rapid and effective treatment.There are many disadvantages such as large amount of data,long time and inaccurate classification if the manual screening is used to diagnose cervical lesions.With the development of science and technology,computer-aided diagnosis has become the trend which can not only reduce the intensity of doctors’ work,but also improve the accuracy and efficiency of diagnosis,so that patients can get more effective treatment.In this paper,various methods of machine learning are used to conduct research on the classification and prediction of cervical lesions in women.The research works are following:(1)Aiming at the research on rough segmentation of cervical lesions and binary classification of mild and severe canceration,proposing an AlexNet-SVM model which combines the traditional machine learning algorithm and neural network to realize classification of cervical lesions.It is divided into the following four parts:Firstly,design and implement a series of preprocessing operations for rough segmentation of cervical images.Secondly,the traditional machine learning algorithm and the neural network method are used to classify the cervical lesions.Thirdly,the AlexNet-SVM model uses the AlexNet network in the front end to extract depth images’ features and inputs the extracted feature parameters into the support vector machine(SVM)in the back end to realize classification.Fourthly,using transfer learning method of frozen model convolution layer can improve the classification accuracy and compare with the classification performance of the original model.The experimental results show that the accuracy of AlexNet-SVM model is 91.77%,the precision is 93.72%,the sensitivity is 86.15% and the specifificity is 95.83%.(2)Aiming at the research on the fine segmentation of cervical lesions and multi-grade prediction of tripartite and quadripartite classification,proposing the model PD-ResNet which uses the modified ResidualNetwork(ResNet)by using pyramid convolution and depthwise separable convolution to realize multigrade prediction of cervical lesions.In view of the inaccurate segmentation phenomenon caused by the fact that most of the cervical images are not in the center of colposcopy images,proposing the extraction algorithm of the center movement of the region of interest(ROI)and accurate segmentation.The algorithm is conducive to the subsequent classification effect.In the three classification experiments of cervical lesions,the method can not only reduce the network parameters to achieve the lightweight model,but also obtain the classification accuracy of cervical images up to 91.29%,the precision is 89.70%,the sensitivity is 88.75%,the specificity is 94.98%,the rate of missed diagnosis(α)is 11.25% and the rate of misdiagnosis(β)is 5.02%through the multiple comparative experiments.The effectiveness of the modified module has been proved by the results of ablation experiments.Finally,dividing the three-classification datasets into four categories to verify the superiority of model classification performance and obtaining the results of the four-classification of PD-ResNet model with the accuracy of 83.06%,the precision is 80.39%,the sensitivity and specificity are 81.17% and 93.94% respectively.