Breast Cancer Detection of Small Sample Based on Data Augmentation and Corner Pooling

Abstract

Breast cancer is the most common cancer among women worldwide. The effective detection the location of breast cancer from the ultrasound images can assist doctors in diagnosing breast cancer. Diverse morphology, blurred edges, and small amount of data causes great difficulty in the detection of breast cancer. Deep learning is very advantageous when facing these problems. However, the problems of training on small sample datasets and the imbalance of positive and negative samples are problems that need to be solved. In order to improve the accuracy of ultrasound breast cancer detection, a small sample breast cancer detection method based on data augmentation and corner pooling is proposed in this paper. In this method, we propose a way for solving over-fitting of small samples and solving the imbalance problem of positive and negative samples. Data augmentation module based on geometric and noise transformation is proposed to solve the problem of small samples, and detection module based on focal loss and corner pooling is proposed to solve the problem of imbalance samples. The experiment found that the method used in this paper has more advantages than the mainstream methods in difficult to distinguish samples. The method used in this paper has an AP of 84.65%, which is higher than state-of-the-art methods.