Research On BI-RADS Density Classification And Object Detection Of Breast Digital X-ray Based On Deep Learning

Part I Study on density classification of breast digital X-ray BI-RADS based on deep learningObjective:To ensure the consistency and accuracy based on ACR BI-RADS mammography density assessment,this study constructs an automatic classification model of mammographic digital X-ray density based on deep learning to achieve accurate classification of mammography density.Methods:In the research,the classic model Res Net 50 based on deep learning convolutional neural network(CNN)was constructed and optimized,and 18152 images of all-digital mammography from August 2015 to February 2018 were collected,The breast density of all the images was assessed by two experienced radiologists according to the ACR BI-RADS criteria.To obtain the classification accuracy,the fine-tuned classification models were used to evaluate the classification of breast density in small datasets(n=4000)and original datasets(n=18,152),respectively.The classification performance of the model was evaluated by the receiver operating characteristic curve and the area under the curve.Results:When the CNN model was trained in small data sets,the classification accuracy of each type was 91% for class a,86% for class b,84% for class c,and 90% for class d.When training in the original data sets,the classification accuracy of the class a and d did not change significantly.The accuracies of b and c were 89% and 88%,respectively.With the increase of data volume,the accuracy rate was significantly improved.The AUC comparison also showed that the classification performance was significantly improved.Conclusion:The deep learning-based convolutional neural network(CNN)classification model can classify breast density with high accuracy.In clinical work,it can assist radiologists to accurately and consistently classify breast density.Part II Preliminary study on digital X-ray object detection of breast based on deep learningObjective:In order to improve the accuracy of detection and diagnosis of breast lesions on breast X-ray images,this study constructs a object detection model for breast X-ray lesions based on deep learning.The value of object detection algorithm based on deep learning in the localization and classification of breast lesions in full-field digital mammography is discussed.Methods:This study retrospectively analyzed 2218 patients who underwent FFDM before operation in our hospital.The radiologists classified the lesions according to the pathological results.Using a object detection model constructed by combining the Resnet network model and the YOLO algorithm.The training data set consisting of 1775 cases marked by manual was trained to obtain an optimized object detection model,and 443 cases constituted a test data set for testing the optimized model positioning and classification accuracy.The test data set included 892 lesions,there were 562 benign lesions and 330 malignant lesions.The accuracy of lesion detection and localization was evaluated by IOU,the accuracy of lesion classification was evaluated by AUC,and the performance of object detection model for benign and malignant classification of breast lesions was evaluated using m AP.Results:The object detection accuracy evaluation index IOU is 87%,the classification sensitivity is 89.1%,the specificity is 87.9%,and the classification performance AUC value is 89.2%.The m AP value was 90.4%,indicating that the object detection model we constructed has better classification performance for benign and malignant breast lesions.Conclusion:The object detection algorithm based on deep learning can locate and classify lesions on breast X-ray images with high accuracy,providing an auxiliary diagnosis for the radiologist’s lesion identification and classification.And for the further study of deep learning in the further application of medical image lesion detection.