SOUSA
A unified weakly and semi-supervised segmentation framework that learns from sparse labels and unlabeled medical images.
Overview
SOUSA is built for realistic medical imaging workflows where dense expert annotation is scarce. The framework combines sparse scribble supervision with large pools of unlabeled data in a teacher-student setting, reducing the annotation burden required for segmentation models to become useful.
Within the broader imaging direction, SOUSA serves as a data-efficiency method: a way to make segmentation pipelines trainable when high-quality labels are expensive or slow to obtain.