BAD - Breast Area Detection
It is not rare to find printed labels on mammograms, especially when talking about digitized versions. Digital images are usually clear but, even in such cases, labels or annotations may exist. These labels are mostly related to the laterality of the breast or the view of the mammogram. Such elements can potentially influence the learning capability of the model during the training procedure; not only they tend to add another bias to the dataset but, also, the extraction of the actual region of interest is getting much more difficult.
To address this problem, the Breast Area Detection (BAD) tool is introduced which takes as input a mammogram and generates a cleared version as output, removing the useless information by the initial image. It consists of three main parts; a) the unsupervised breast area masks generation, b) the data augmentation and c) the supervised breast area mask generation.
Tzortzis, Ioannis N., et al. "Towards generalizable Federated Learning in medical imaging: A real-world case study on mammography data." Computational and structural biotechnology journal 28 (2025): 106-117.