Classification Of Breast Tumors Based On Pathological Images

Breast cancer is a malignant tumor with the highest mortality rate in women.To improve the diagnostic efficiency and provide doctors with more objective and accurate diagnosis results,we conducted an indepth study in a public data set Brea KHis of pathological images of breast tumors.The main contents are as follows:(1)In the paper,the nuclei of the pathological image of the breast tumor are segmented.Here we carried out a radiomic analysis of 585-dimensional features quantifying breast tumors’ texture,color and the nuclei’ shape and intensity.Relief F is used for feature selection,and then we used three different classifiers such as Random Forest(RF),Extreme Learning Machine(ELM),and Support Vector Machine(SVM)to build a diagnostic model of breast tumors.The results show that the classification of SVM is the best.For binary classification,the accuracy rate reaches95.96% at the patient level and 95.82% at the image level.For eight categories,the accuracy rate reached 8 6.15% at the patient level and86.68% at the image level.(2)This paper proposes a cascade convolutional neural network(CNN)based on Res Net50,which is used for benign and malignant classification of breast tumors and multiple classification of subtypes.First,this paper divided images into patches and the pre-processed breast tumor pathological image was input to the first level network for benign and malignant classification with the help of transfer learning and data augmentation methods.Then,to achieve eight classification of each subtype under benign and malignant,the output of the network was used as the input of the subsequent level networks.Finally,the classification results of each patch are integrated,and the integrated results are used as the classification results of the images.For binary classification,the accuracy rate reaches 99.26% at the patient level and 99.18% at the image level.For eight categories,the accuracy rate reaches 98.63% at the patient level and 98.39% at the image level.It proves the effectiveness of the framework in the classification of breast tumors.