Breast Image Classification Based On Deep Learning

Breast cancer is one of the leading causes of cancer death in women,which seriously threatens the physical and mental health of women worldwide.Studies have shown that detection,diagnosis,and treatment early can significantly improve the survival rate of breast cancer and reduce mortality.As the preferred method for early diagnosis of breast cancer,mammography has the advantages of fast imaging,low cost,and can display early lesions that can not be perceived by palpation.In recent years,computer-aided diagnosis(CAD)based on mammograms has been widely used in the clinic,which has effectively solved the problem of lack of medical resources to a certain extent.At the same time,the CAD technology based on mammography can provide advice for clinical diagnosis of radiologists and improve the accuracy of early diagnosis of breast cancer.By exploring the inherent characteristics of mammograms,this paper designs and implements the mammograms classification research based on deep learning image recognition algorithm,which effectively avoids the dependence of traditional methods on prior knowledge and the subjectivity and one-sidedness of manual features of traditional methods.It is a more objective and intelligent automatic method.Specifically,the work of this paper mainly includes the following aspects:1.Using Convolution Neural Network(CNN)to classify mammograms,the algorithm was tested on Mammographic Image Analysis Society(MIAS)database.The mammograms were classified into three categories: normal(N),benign(B)and malignant(M).The highest classification result was 66.9%.2.Considering the influence of gland type on classification accuracy,a method to improve the classification accuracy of breast lesions is proposed,that is,to classify breast lesions in the case of differentiating the gland type of mammogram.The accuracy of image classification based on CNN algorithm structure is effectively improved.The effect of gland density on classification is demonstrated again.3.In order to explore the necessity of image preprocessing for deep learning based on small-scale datasets,the mammogram is preprocessed firstly in the algorithm design,and then mammogram is classified by using CNN model.The experimental results show that the preprocessing can effectively reduce interference factors and improve the network learning effect when using deep learning in small-scale datasets.4.In this paper,we designed a gland density classification experiment based on deep learning,and obtained 84.9% overall classification accuracy,94.5% single fat classification accuracy,80.9% single fat-gland classification accuracy and 82.1% single dense classification accuracy.The experimental results show that the deep learning method is feasible for breast density classification.In summary,the work of this paper focuses on the classification of mammograms,providing some analysis and reference for the follow-up mammographic lesion detection and classification tasks,and promoting the further study of breast cancer.