Breast, thorax and chest-muscle segmentation on breast MRI
Three-label 3D segmentation (breast, chest muscles, thorax) on T1 breast MRI with correction of a pre-existing algorithm.
A medical imaging vendor was iterating on its breast MRI algorithm and needed a training set with a new chest-muscle label added to the existing breast and thorax masks, on axial T1 series (with and without fat saturation), across a mix of prostheses, lesions and densities.
Deliver clean three-label segmentations — breast (skin, glands, nipples, axillary hollows, breast arteries), chest muscles (pectoralis, intercostal, back and shoulder muscles, sternum, ribs) and thorax (lungs, heart, liver) — on all axial slices, with no gaps or overlaps between masks.
Annotators worked on the client's RedBrick project using the fast tool for initial masks and brush / pen for refinement, correcting the pre-existing segmentation where available or starting from scratch on poor-quality cases. Cases with strong artifacts on the heart or thorax contours were rejected with a comment.
The client received a coherent three-label dataset following anatomical delimitations, with smooth contours and no gaps between masks — ready to retrain the next algorithm version.