Research On Cervical Lesion Image Classification Based On Convolutional Neural Network

Cervical cancer not only ranks first among female genital tumors,but is also the most common of various malignant tumors in women.Through early cervical screening,women's risk of cervical cancer can be effectively reduced,and early cervical lesions can be observed and diagnosed in a targeted manner.The acetic acid test is the main screening method for early cervical cancer.Clinicians need to observe the results of the acetic acid test with the naked eye.Because there are many interfering factors in the acetic acid experiment of the artificial colposcopy,the main manifestations are: the subjective diagnosis of professional doctors is too strong,the qualifications and levels are different,the diagnosis is too complicated,and it is easy to cause misjudgment.In the above research background,it is of great significance to realize an intelligent cervical lesion image screening system.In recent years,with the development of cervical screening technology,a large amount of cervical lesion image data has been accumulated,providing a rich data source for the research of cervical lesion medical images.At present,deep learning technology has been successfully applied in the field of medical assisted diagnosis.It is expected that the accuracy of diagnosis can be effectively improved by performing deep learning on cervical lesion images.This paper first uses convolutional neural networks to classify and predict cervical lesion images,combines the advantages of Inception and ResNet,and proposes an improved model Inception-ResNet,and uses transfer learning to train the model,and then compares several common convolution The effect of the neural grid model.Aiming at the shortcomings of convolutional neural network,this paper introduces a capsule neural network.Based on the fusion of convolutional neural network,a CNN-CapsNet prediction method is proposed.This paper adopts two experimental schemes.The first one uses transfer learning technology to train the improved model Inception-ResNet and compares it with the accuracy of ResNet and InceptionV3 models.The second one uses the CNN-CapsNet method proposed in this paper.training.The experimental results show that the InceptionV3 method has achieved a prediction accuracy of 68.1%,and the prediction accuracy of the ResNet method is 1.2% higher than that of the InceptionV3.Compared with the above two methods,the CNN-CapsNet method has achieved the highest prediction accuracy.In terms of classification accuracy,the prediction accuracy of the CNN-CapsNet prediction method for cervical lesion images is 71.2,which is 1.9% and 3.1% higher than the ResNet and InceptionV3 methods,respectively.This verifies that the CNN-CapsNet model is used in cervical lesion image detection Compared with other test models.In view of the lack of intelligent factors in the current hospital diagnosis system,a cervical lesion image classification system based on convolutional neural network is designed.The main application scenario of this system is that the user uploads an image of a colposcopy device check on the web terminal and automatically gives it The classification level is for the doctor's reference and assists the doctor's diagnosis.Considering that colposcopy doctors are easily fatigued when reading cervical lesion images for a long time,which leads to unnecessary omissions and misjudgments,and the system has advantages in this regard.