Research On Breast Mass Detection And Classification Algorithm Based On Deep Learning

According to the latest data released by Global caner:Breast cancer is one of the most common malignant tumors among women worldwide: in 2012,there were approximately 1.671 million new cases of breast cancer among women worldwide,ranking first in the incidence of female malignancies.By 2021,the number of breast cancer patients in China will reach 2.5 million,and the incidence will increase from less than 60 cases/100,000 women(aged from 55 to 69)to more than 100cases/100,000 women.Studies have shown that early screening,accurate diagnosis,and precise and aggressive treatment can significantly reduce breast cancer mortality.Therefore,early screening of breast cancer research has very important practical significance.Early X-ray signs of breast cancer include mass and calcification.Usually the mass edges are blurred,the shapes and sizes are different,and the features are diverse.In addition,there is a large amount of dense tissue in the mammary gland,which makes the detection of breast lumps more difficult.In recent years,with the big data,the artificial intelligence technology based on deep learning has become more and more mature,and with the simplicity of this detection process,the advantage of high automation has become a new development direction in the field of image target detection.How to use deep learning technology combined with medical imaging to assist reading doctors has become a hot issue in the field of medical image processing.Therefore,this article will focus on the use of regional convolutional neural network technology to achieve the radiology review doctor's lump detection evaluation classification process,reduce the workload of reading doctors,help clinicians early screening breast cancer,reliable,stable,highly accurate Breast cancer screening.The main research in this paper is the detection of breast masses based on regional convolutional neural network and the classification of malignancy evaluation.Firstly,this paper reviewed the status quo and development direction of deep learning in the field of image target detection,and then Faster R-CNN network,a classical target detection network,was improved to detect breast lumps.Finally,the evaluation and classification of breast masses were realized based on RI-RADS standard.The main work of this paper is as follows:1.Breast mass detection based on Faster R-CNN improved networkIn order to quickly and accurately detect the breast lump from the molybdenum target image,we proposed a breast lump detection method based on the improvedFaster R-CNN network.Firstly,the data of experimental training was expanded by using data enhancement technology,and then the image data was marked according to the information marked by experts provided by the database,that is,the location information of breast was marked on the image.The annotated data was then sent to the improved Faster R-CNN network for training.In this improved network,we reset the size of anchor to make it more consistent with the size of breast lump.In addition,we added feature fusion technology in the last two layers of the Shared convolutional layer,and then sent the fused convolutional feature map into the RPN network to generate candidate regions.The experimental results show that the improved network model is 8.5 percentage points more sensitive than the original model,and 18.5% less sensitive than the original model.2.Classification of breast tumor malignancy based on BI-RADSAccurate qualitative analysis of the mammary gland is very necessary for doctors to make accurate and timely diagnosis and treatment plans for patients.At present,a large number of literatures only classify breast masses as benign or malignant(either good or evil).This is very unscientific in clinical practice,and it is easy to misdiagnose the patient's condition.Moreover,in clinical practice,the malignancy of the mass is finally determined by case analysis,not by molybdenum imaging alone.The general practice of doctors reading the film is to classify the masses(BI – RADS,0,1,2,3,4,5)according to the breast cancer diagnosis and treatment guidelines and norms issued by the breast cancer expert committee of the anti-cancer association,rather than simply classify them as either evil or good.We redesigned the shared convolutional layer based on the previous improvement of the network model,and finally determined the shared convolution of the seven convolutional layers and the two pooled layers.The experimental results show that the detection and classification accuracy of our model for BI-RADS 0,1,2,3,4 and 5 reaches 39.5%,91.3.4%,64.3%,73.6%,73.9 % and 79.3%,respectively,with an average accuracy of 70.3%.Although the accuracy of the classification is still to be improved,the model basically realizes the classification of the malignancy evaluation of breast mass based on BI-RADS.Compared with the previous classification of benign and malignant breast masses,it has clinical significance.