Research On Breast Cancer Pathological Image Recognition System Based On Convolutional Neural Network

Breast cancer is a common tumor that currently threatens women's health.According to the latest data,the cancer with the highest incidence among Chinese women is breast cancer,and the cause of cancer death ranks sixth.The most reliable way to diagnose breast cancer is through pathological diagnosis.Therefore,it is of great significance and clinical application value to realize the classification and classification of breast cancer pathology images.and the increase in the incidence of breast cancer,these few years,with the extension of deep learning in the field of computer vision,pathological image recognition of breast cancer has become a research hotspot.This paper is based on the convolutional neural network for breast cancer pathology image recognition system research,the main work is as follows:(1)Most studies in recent years have focused on benign and malignant classifications,but with the increase in the incidence of breast cancer,in order to better help doctors diagnose which type of disease is specific,different types of benign and malignant breast tumors are classified.Different treatment methods and therapeutic significance are available to determine the treatment plan more quickly.Therefore,the main research in this study is the specific classification of breast cancer pathological images instead of the simple benign and malignant classification.(2)In order to solve the over-fitting problem caused by insufficient data volume,the data set is expanded by data enhancement.Make a series of transformations such as flipping,panning or rotating.This can improve generalization,by increasing the amount of data in the training process;it can also improve the robustness by adding noise data.(3)In order to improve the training effect,save training time,and adopt the method of migration learning,this paper uses the natural image classification ImageNet dataset to fine-tune the pre-trained Inception-Resnet V2 network model parameters to the breast cancer pathological image dataset.The network first learns some features and uses it in the target dataset to help it identify breast cancer pathology images.(4)This paper studies three kinds of convolutional neural networks popular in recent years,namely:GoogleNet,ResNet50,Inception-ResnetV2.By studying their structure,advantages and disadvantages,it is found that the deeper the depth of the network,the better the prediction effect,and the first time Inception-ResnetV2 network model was applied to the pathological image recognition of breast cancer.The results showed that the recognition rate of the method was more than 80%,which was much better than others.(5)In order to make doctors and patients more convenient and quick to apply breast cancer pathology image recognition,this paper based on the experimental results using Core ML and CNN network to establish breast cancer pathology image recognition APP,development tools for Xcode 9,programming language using Swift 4.0.The function is to take a pathological picture of breast cancer or select a pathology image of breast cancer from the album,and then the App will try to identify which disease is shown in the picture.I believe this will provide diagnostics and help for doctors and patients.