Automatic Staging System For Whole Slide Images Of Breast Lymph Nodes

Breast cancer is the most common malignant tumor in women all over the world,with a mortality rate of around 6%.Compared with other malignant tumors,breast cancer has a better.Early detection,early diagnosis and early treatment can make the cure rate of early breast cancer reach over 96%.The staging of breast cancer metastases is an important component of breast cancer diagnosis.In clinical,this work is mainly carried out by pathologists using microscopes.It is not only time-consuming and laborious,but also lacks unified evaluation criteria,and has low reproducibility.The diagnosis results are subjective and easy to be affected by factors such as experience and environment.In recent years,computer-aided diagnosis has become one of the fastest-growing directions in the field of medical image computing.However,due to the large size and high complexity of pathological images,the analysis of pathological images is still a challenging problem.The automatic staging method of breast cancer lymph node metastasis based on image computing proposed in this paper can provide a set of objective,high accuracy and repeatable diagnosis results.It is very difficult to automatically detect and locate cancer metastasis areas in highly complex lymph node images.In this paper,A novel training strategy is proposed based on sliding window to train the automatic positioning model for cancer metastasis.Firstly,A small amount of data is used to train the initial convolutional network,which is used to extract the false positive and false negative image blocks.Secondly,the false positive blocks are reclassified by manual screening combined with network screening to improve the interclass difference of negative categories.Moreover,the robustness of the model is improved by data enhancement such as rotation,mirroring and H&E dye transformation.Thirdly,a deeper convolutional neural network is trained to realize the automatic classification of image blocks.In the testing phase,the white background regions were removed.Then,the cancer metastasis area is automatically located based on the sliding window and the deep convolutional network and a probability calorific value map for cancer metastasis is obtained.Finally,the pathological features concerned by pathologists were extracted to train a random forest classifier,which can be used to predict the type of cancer metastasis.Then,the staging result was determined according to the predicted values of multiple lymph node sections of patients.In the test set,the classification accuracy rate can reach 0.94 and the FROC on cancer metastasis area is 0.9464.The AUC value is 0.96 and the kappa coefficients is 0.8192.In addition,a software system based on automatic analysis of panoramic images for breast lymph nodes is designed,which can help pathologists quickly locate areas suspected of cancer metastasis and give the suspected type of cancer metastasis,so as to assist pathologists get corresponding staging results.