Breast Mass Detection And Diagnosis In Mammography Images

Mammography is one of the safest and most effective methods for early detection of breast cancer,but it is difficult for doctors to avoid missed diagnosis and misdiagnosis by hand.It is of great clinical value to research automatic detection and benign and malignant diagnosis of breast lesions based on mammograms.Based on the mothod of deep learning,the automatic detection of Mass in mammograms and the diagnosis of bengin and malignant breast Mass are studied.Aiming at the problem of narrow distribution range of mammary gland pixel value and blurred Mass boundary in images,the truncated normalization method is adopted,and combined with adaptive histogram equalization algorithm to preprocess the original images.Aiming at the problem of small size of data sets,a data expansion method based on local elastic deformation is proposed to increase the diversity of training samples and improve the training method of dunamic data sets updating,which makes the training process bases on RetinaNet of automatic Mass detection model more stable,more effective in mining gard samples and making full use of data from validation set to improve the robustness of detection model.In view of the small number of training dataset and fact that the transfer learning from natural images to medical image can not meet the requirements of medical image application,a hierarchical training and multistage transfer learning is adopted to make full use of the available datasets and improve the accuracy of the diagnosis model based on Inception-ResNet-V2 for benign and malignant Mass.After testing on the Inbreast dataset,Mass automatic detection model achieves a recall 0.913 when FPPI(False Positive Per Image)is 0.52;Mass diagnosis model achieves a accuracy 0.943 and the AUC(Area Under Receiver Operating Characteristic)reaches 0.953.At the same time,759 samples from DDSM(Digital Dataset for Screen Mammographys)were used to evaluate diagnosis model,and the accuracy of 0.931 and AUC 0.982 were obtained.The experimental results show that the method and models used in the detection of breast Mass and the benign and malignant judgment have a high accuracy,which has a certain reference value for clinical application.