Breast Image Classification And Target Detection Based On Deep Learning

According to the 2015 China Cancer Statistics data,from 2003 to 2014,the average annual number of cancer incidences was more than 3 million.For female friends,the incidence of breast cancer ranks first.Early detection and treatment of breast cancer can greatly improve the cure rate of breast cancer and relieve the physical and mental suffering of patients and their families as soon as possible.With the development of computer technology and artificial intelligence technology,the use of computer-assisted diagnosis for breast cancer screening has gradually become a trend.Because early breast tumors are usually small,with different shapes,blurred edges,and some dense tissues,these factors bring great challenges to the classification and detection of breast images.Most traditional breast image classification and target recognition use machine learning methods,the network level is not deep enough,many high-level image semantics cannot be mined,and image features often need to be manually extracted based on human experience.Such feature extraction is not only troublesome but also Not objective enough,resulting in low accuracy of final classification and recognition.In recent years,computer-aided diagnosis based on deep learning has gradually become a research focus in the field of auxiliary diagnosis.Compared with general machine learning algorithms,deep learning has a deeper and more complex network structure,and performs better for many highly nonlinear classification and recognition and target detection tasks.Therefore,this paper uses deep learning to classify and detect target mammography X-ray images.The main work done in this paper is as follows:(1)Breast image classification based on convolutional neural networkFirst,pre-processing the breast images in the original data set,including image binarization,removal of image noise,removal of artificial images,and removal of breast muscle images,etc.,basically eliminating the impact of redundant and irrelevant images on breast image classification.After that,the data set is augmented by data to expand the data capacity to 20 times.The convolutional neural network model in this paper is based on the VGG-16 network,and the number of network blocks is reduced,and the configuration of-the network parameters is improved.The improved VGG-16 network model.Later,the improved VGG-16 model is applied to the classification of breast images.The experimental results show that the average classification accuracy of the improved model is improved by about 2%,and the average test time of a single image is also reduced by 25.7ms.(2)Breast tumor target detection based on region proposal networkThe experimental data still uses the pre-processed breast image data set and is expanded to 40 times by the new data augmentation.The target detection network is mainly Faster R-CNN.Based on this,the multi-scale feature fusion method is combined with region proposal networks in Faster R-CNN.At the same time,the ROI pooling layer of the network is improved to adapt to multi-scale fusion Feature map,the final breast tumor target detection results show that the improved Faster R-CNN model has higher recognition accuracy of each category of target compared to the original algorithm,and mAP is increased by 4%.Observation of the detected image also basically proves this However,the average detection time of a single image of the improved algorithm is 15ms longer than the original algorithm.Figure[40]Table[8]Reference[56].