Automatic Breast Screening by Fully Automated Ultrasound Imaging Using CNN And Transformer Algorithm

Abstract

To investigate and evaluate the accuracy of automatic detection of breast cancer by fully automated ultrasound imagings(called “ring echo”) using deep learning. One hundred and ninety-four women diagnosed as breast cancer were performed by ring echo to obtain 917 images with breast cancer and 3321 images without breast cancer. These datasets were trained with the novel algorithm using convolutional neural network and transformer. The result showed accuracy 87.9%, sensitivity 87.9%, specificity 85.3% per images, and accuracy 78.6%, sensitivity 85.7%, specificity 71.4% per case. Ring echo imaging could help contribute and change whole system of breast screening with ultrasound.