Deep Learning For Computer Aidid Diagnosis Of Cervical Cancer

Cervical cancer is a major threat to women's health.However,it takes about 10 years or more for cervical precancerous lesions to develop into cervical cancer.Early detection of precancerous lesions may effectively reduce the incidence of cervical cancer as patients may get early diagnosis and treatment.It is of great significance to investigate the computer aided diagnosis method of cervical cancer based on deep learning to realize the early diagnosis of cervical cancer while reduce the unnecessary manpower and cost when facing a large number of colposcopy images.The content of this paper is as follows:Firstly,we analyzes the current research status of cervical cancer screening,and points out the problems and their causes of colposcopy.And then we review the development of deep learning,inspired by the successful application of deep learning in the field of medical diagnosis,this paper proposed a method based on deep learning for computer-aided diagnosis of cervical cancer.Secondly,we perform a series of operations on colposcopy images,mainly including the extraction of the region of interest and the removal of the specular reflection.First,a large number of colposcopy images which are repetitive,of poor quality,or even unrelated to disease diagnosis are excluded.Then,with the help of experimental data which was marked by colposcopy experts,we train the biomedical image segmentation network,U-Net,to extract the region of interest.Secondly,the colposcopy image with specular reflection was selected for specular reflection removal.Thirdly,we select a better augmentation scheme specially for colposcopy images by performing contrast experiments,and on this basis,the deep residual network is trained to classify the experimental data of 8 times augmentation.The experiment finally obtains the classification accuracy of 69.39% on the test set.Compared with the experimental results of Navdeep,etc.,there is a 7 percentage point improvement.Finally,we train a simple network based on the attention mechanism to classify our colposcopy images.In this paper,three kinds of attention region generation strategies are proposed for colposcopy images,we only choose the best effect one to classify colposcopy images.The results show that with the help of attention mechanism,the classification effect of the network has certain upgrade.Computer aided diagnosis of cervical cancer based on deep learning can effectively reduce the unnecessary manpower and cost when facing a large number of colposcopy images.At the same time,it can provide appropriate reference for doctors and provide a more sufficient basis for diagnosis and further treatment to guide doctors thus effectively reduce the incidence of cervical cancer.