Combination Ultrasound and Mammography for Breast Cancer Classification using Deep Learning

Abstract

The most widely used methods for early detection of breast cancer are Ultrasound and Mammography. However, single ultrasound or single mammography shows false classification that causes unnecessary biopsy. Therefore, the combination approach is proposed to improve breast cancer classification using the deep learning technique. The proposed method has been divided into two steps. First, images are randomly combined using the k-combination method. Second, deep learning based on MobileNet is used to classify breast tumors. The result demonstrated that the combination approach produces a variety of patterns and a large image dataset and improves the accuracy. In addition, the false positive tend to reduce by 13% and the false negative tend to reduce by 14%. It is useful to avoid unnecessary surgery and to plan aggressive treatment.