BCSDBT-seg

Publicly available DBT datasets remain scarce, and existing datasets predominantly only provide lesion-level labels or bounding-box annotations. To facilitate the development and evaluation of segmentation algorithms for DBT, we present DBT-Seg dataset, a voxel-level lesion segmentation dataset derived from the publicly available Breast Cancer Screening Digital Breast Tomosynthesis (BCS-DBT) dataset.

Expert annotators manually delineated lesion boundaries on DBT slices containing biopsy-proven benign and malignant findings. The dataset includes voxel-wise breast-lesion masks in standard image formats, linked to the original DBT examinations. In addition, BreastImaging Reporting and Data System (BI-RADS) scores are also provided per case. The dataset consists of 201 patients with 396 DBT images, including MLO and CC views with biopsy confirmation.