Combining VGG16, Mask R-CNN and Inception V3 to identify the benign and malignant of breast microcalcification clusters

Abstract

Breast cancer has the highest incidence among women in Taiwan, which is mainly screened by mammography. When the doctor observes the mammogram and initially judges that the patient has malignant microcalcification (MC) clusters, the patient must undergo needle localization surgical biopsy. However, needle localization surgical biopsy makes the patient painful, and the color of breast tissue and MCs are all white, which makes it difficult for doctors to judge where MCs are clustered immediately. Thus, we use VGG16 to find out breast MC clusters from the image. Moreover, we use Mask RCNN to find MCs from the clusters to remove the noise from the background. Finally, we use Inception V3 to identify the benign and malignant of MC clusters. The accuracy of the cluster classification, MCs labeling and benign and malignant analysis are 93%, 95% and 91%. Furthermore, the precision, specificity and sensitivity of our proposed methods are about 87%, 89% and 90%, respectively. It proved that our system can effectively assist doctors in diagnosing and reduce the burden on patients and medical personnel.