ViT-Based Classification of Mammogram Images: Impact of Data Augmentation Techniques

Abstract

Breast cancer is a major disease for women, and mammographic screening proved impressive benefits in reducing mortality risk. However, false positives and false negatives still occur due to human perception, differences in breast density, and the complexity of cancer itself. Convolutional Neural Networks (CNNs) have shown promise in medical imaging issues, but they struggle with understanding long-range spatial interactions in various image patches. Vision Transformers (ViTs) have emerged as a solution to this problem. This work used a subset of the Curated Breast Imaging Subset of the Digital Database for Screening Mammography (CBIS-DDSM) to implement a ViT-based framework for breast cancer classification. Geometric and Diffuser-based data augmentation (DA) methods were applied and compared to evaluate the resulting performance improvement. The obtained results show how diffuser-based DA improves the performance of geometric DA. However, their combination allows for higher performance (accuracy = 77.01%, sensitivity = 88.89%, specificity = 68.63%) and demonstrates the feasibility and effectiveness of this approach in enhancing the model’s capabilities for breast cancer classification.