A Data Augmentation Approach to Enhance Breast Cancer Segmentation

The early detection of breast tumors is a critical concern for healthcare professionals, including oncologists and radiologists. While Artificial Intelligence (AI) has demonstrated potential in early breast cancer diagnosis, the efficacy of these models is often constrained by the limited size and lack of diversity in medical training sets. Although data augmentation techniques are explored to enlarge and enhance training sets, many such methods neglect the crucial aspect of sample diversity, leading to suboptimal tumor identification. Among the prevalent data augmentation techniques, the MixUp method is commonly employed to increase the size and diversity of data sets. However, its application in ultrasound image enhancement can introduce extraneous noise and may result in the loss of vital image features. This paper presents a novel data augmentation strategy termed Cluster and MixUP (Cluster MixUP) Augmentation, designed to enrich the diversity of training data while retaining essential image features. The approach combines K-means clustering with the MixUp Augmentation technique to group and mix images effectively. The efficacy of the proposed strategy is validated using the Breast Ultrasound Images database (BUSI), demonstrating superior performance and generalizability in breast cancer detection relative to existing data augmentation methods.