Lesion annotation on Digital Breast Tomosynthesis
Lesion segmentation and BI-RADS classification across large volumes of 3D DBT, 2D tomosynthesis and 2D mammography images.
A medical AI company building breast imaging models needed a large annotated dataset combining 3D digital breast tomosynthesis, 2D tomosynthesis reconstructions and standard mammography views, with several thousand studies per modality.
Provide lesion segmentation and BI-RADS classification on the 3D DBT volumes, with optional layers of annotation (2D lesion segmentation, calcifications, architectural distortions, density, descriptors and lymph nodes) available on request.
Dedicated radiologists trained on breast imaging were assigned to the project and worked on the client's platform under a modular protocol. Annotation was performed in tiers, starting with the mandatory 3D DBT layer, then adding optional layers as the client's model iterations progressed.
Delivery followed a predictable weekly cadence across all three modalities. The tiered approach let the client scale annotation depth as the model matured, without renegotiating scope or timelines.