Research On Breast Cancer Detection Methods Based On Mammogram

Breast cancer has the characteristics of high morbidity and high mortality,and has become a major disease that threatens women's health and life.Early screening and early diagnosis of breast cancer is the most effective way to save patients' lives.Mammography is the most important imaging method in breast cancer screening.The imaging characteristics of early breast cancer are unclear.The density classification,microcalcification detection and benign and malignant classification of tumors are the main difficulties in screening for early breast cancer.It is susceptible to the subjective influence of doctors and is prone to misdiagnosis and missed diagnosis.With the continuous development of computer technology,computer-aided detection of breast cancer has become a hot area of medical imaging research.Effective computer-assisted detection methods can help doctors better analyze mammograms and improve the accuracy of breast cancer diagnosis.This research combines computer vision and machine learning to construct a breast cancer detection algorithm.The research work is summarized as follows:(1)A mammography classification algorithm based on wavelet transform for mammography is proposed.Women with high breast cancer density have a risk of breast cancer that is four to five times that of low-density breast cancer.Under the guidance of a professional physician,the analysis and study of breast density in this study is divided into three stages: first,the breast image is separately processed for denoising and adaptive histogram equalization,and then the wavelet technology is used to fuse the former to enhance the breast density of different densities Gray histogram features.Then,through multiple iterations to determine the threshold of the mammary gland region,and obtain a standard binarized mammogram,the ratio of the black and white pixels of the image is worth the breast density.We evaluate the performance of the breast density algorithm according to the rules of the Breast imaging reporting and data system(BI-RADS).The accuracy of our segmentation results is 85%,and the experimental results are better than those of radiologists and similar algorithms,providing clinicians with a more efficient classification algorithm for breast density.(2)An algorithm for detecting breast calcification lesions in mammograms based on adaptive support vector machines is proposed.Microcalcification is the most important marker of early breast cancer.At present,artificial morphological observation is the main method for clinical diagnosis of such diseases,but it is easy to cause misdiagnosis and missed diagnosis.We designed a detection algorithm based on the characteristics of calcification lesions.First,we used contour wave transformation and mathematical morphology(Contourlet transformation and morphology(CTM))to enhance the contrast features of mammograms.Then,a K-means clustering algorithm(K-means)is used to segment the region of interest(ROI),and then the gray features,shape features,and region histogram gradients of the ROI are calculated.Classification of coarse calcification points and false calcification points of the region of interest by adaptive support vector machine.The results show that the accuracy of the algorithm for detecting breast calcification lesions is 94%,and the experimental results are superior to radiologists and similar algorithms,and have great clinical application value.(3)A new algorithm for breast mass detection based on transfer learning finetuning network is proposed.Breast lump is the most important marker information of advanced breast cancer.With the emergence of deep learning applications in the medical field,transfer learning has received more and more attention.We designed a benign and malignant classification algorithm based on the characteristics of breast mass lesions: first,the learned ImageNet model parameters were migrated,the output layer of the original network was removed,and the remaining entire network was used as a fixed feature extraction machine.It is then applied to our breast mass data set,and then during the fine-tuning process,according to the characteristics of the breast mass lesion,the parameters of all layers in the network are frozen,the last few layers and the output layer are added,and only the last few layers of the network need to be fine-tuned to train.Continuously improve the model to improve the accuracy of the algorithm.The results show that the classification accuracy of the fine-tuned VGG algorithm reaches 94%,and the experimental results are better than the convolutional neural network algorithm and traditional machine learning algorithm trained from scratch.This algorithm is more efficient and provides a second reference point for radiologists.