Application Of Deep Neural Network In The Prediction Of Benign And Malignant Breast Tumors And Metastasis

Breast cancer is the cancer disease with the largest number of deaths among women in China.The International Agency for Research on Cancer has made statistics on cancer related data.The data shows that the number of breast cancer has exceeded that of lung cancer,becoming the disease with the largest number of cancer patients in the world.In China,the early detection of breast cancer is less and the diagnosis rate is low,leading to the rapid growth of breast cancer mortality in China.Ultrasonography has become the most common diagnostic method in the diagnosis and treatment of breast cancer because of its convenience,non-invasive,low cost and other advantages.This method needs to interpret a large number of ultrasound images to give the diagnosis and treatment results,which is a painstaking thing for doctors.Therefore,research on automated diagnosis is imperative.This article designs three algorithms based on deep neural networks for the binary classification of benign and malignant breast tumor ultrasound images,the five classification of disease types in breast tumor ultrasound images,and the prediction of axillary lymph node metastasis status in breast tumor ultrasound images.The main work includes the following three points:(1)This article has jointly established a hospital breast ultrasound dataset with a cooperative affiliated hospital.A total of 389 cases of 2D and 3D DICOM files were obtained from cooperative hospitals.After screening,a total of 1172 breast ultrasound images were available for this study,including 574 benign tumor ultrasound images and 598 malignant tumor ultrasound images.In the ultrasound images of malignant tumors,there are 143 Luminal A type,163 Luminal B type,138 HER2 overexpression type,and 154 triple negative types.Through the pathological results provided by the cooperative hospital,ultrasound images of malignant tumors with axillary lymph node metastasis were also screened.Among them,271 had axillary lymph node metastasis and 327 had no metastasis.(2)A Dense Net classification model based on SKNet is proposed,which achieves the binary classification of benign and malignant breast tumor ultrasound images and the five classification of disease types in breast tumor ultrasound images.(3)Propose a YOLOv5 s detection model based on CBAM attention mechanism to predict the status of axillary lymph node metastasis in ultrasound images of malignant breast tumors.The accuracy of the improved model proposed in this study in the binary classification of benign and malignant breast tumor ultrasound images and the five classification of disease types in breast tumor ultrasound images is 97.1% and 96.2%,respectively.The m AP for detecting lymph node metastasis status in malignant breast tumor ultrasound images is 92.5%.The results showed that the improved model designed in this paper had good performance in the detection of benign and malignant breast tumors,five types of breast cancer,and axillary lymph node metastasis.