Self-supervised 3D medical image segmentation by flow-guided mask propagation learning

Adeleh Bitarafan, Mohammad Mozafari, Mohammad Farid Azampour, Mahdieh Soleymani Baghshah, Nassir Navab, Azade Farshad

Abstract: Highlights•Flow2Mask is a self-supervised mask propagation method for volume segmentation.•Segmentation is done by propagating a single annotated slice to the unlabeled slices.•The inter-slice flow fields are learned by the L2G loss based on curriculum learning.•The ISS loss ensures smoother changes between consecutive slices within a volume.•Systematic selection of the most representative slice for manual annotation.