A Novel Transparency Strategy-based Data Augmentation Approach for BI-RADS Classification of Mammograms

Abstract

Image augmentation techniques have been extensively studied to enhance the performance of deep learning (DL) algorithms in mammography classification tasks. Recent advancements have demonstrated the effectiveness of image augmentation in addressing issues of data deficiency and imbalance. In this paper, we introduce a novel transparency strategy aimed at improving the Breast Imaging Reporting and Data System (BI-RADS) scores for mammogram classifiers. This new method leverages Region of Interest (ROI) information to create additional high-risk training examples (BI-RADS 3, 4, 5) from the original images. Our comprehensive experiments across three different datasets reveal that this approach significantly boosts mammogram classification performance, outperforming the state-of-the-art data augmentation technique known as CutMix. This research also underscores that our transparency method is more effective than other augmentation strategies for BI-RADS classification and has the potential for broad application in various computer vision tasks.