Study On The Classification Of Cervical Cytopathological Image Based On Deep Learning

Cervical cancer is a common gynecological malignancy.Cervical cancer screening and diagnosis is one of the effective means to prevent and treat the disease.Among them,the classification of cervical cytopathology image is an important factor affecting the accuracy and efficiency of auxiliary screening and diagnosis of this disease.This article is based on deep learning to study the classification of single-cell and multi-cell image of cervical pathology.For the problem of cervical single-cell image classification,the research was carried out on the public data set SIPaKMeD(abbreviated as data set 1)and cervical single-cell image data set(abbreviated as data set 2)derived from the original cervical pathological cell image and diagnostic information provided by various hospitals in Xiangtan City are obtained through cutting,screening,and labeling.The above two kinds of data sets were enhanced by the flip and rotation methods,and the convolutional neural network VGG16 combined with transfer learning,model parameter tuning and numerical experimental analysis to build the two kinds of cervical single cell image classification models.Numerical experiments show that the two models designed have high accuracy for the above two data sets,which are 98.59%and 97.50%,respectively.Since it is difficult to screen and diagnose various complicated cervical lesions only using cervical single-cell image,research has been conducted on the classification of cervical single-cell image based on the classification of cervical single-cell image.Firstly,a cervical multi-cell image data set(abbreviated as data set 3)was obtained by using a similar method to obtain data set 2 from the data provided by hospitals.Then,based on the VGG16 network model,the classification model of cervical multi-cell image was established by comparing and analyzing various data sampling methods and selecting the focus loss function.Numerical experiments show that the proposed model has a good accuracy for the problem of data imbalance.