Efficient Organ Segmentation with Minimal Bounding Boxes

This academic paper presents a novel approach to semantic segmentation in medical imaging, specifically focusing on organ identification using a limited amount of labeled data. The authors introduce a method that significantly reduces the need for extensive, pixel-level annotations by leveraging bounding box annotations for semi-supervised few-shot learning. Their proposed U-Net-based unbranched architecture utilizes prototype learning and a combined loss function to achieve fast 3D segmentation, demonstrating improved performance with fewer annotations compared to traditional supervised methods. The research highlights the potential of few-shot learning to address the challenge of data scarcity and the high cost of expert annotation in the medical domain.

CITE:

Parida A, Tran A, Navab N, Albarqouni S. Learn to segment organs with a few bounding boxes. arXiv preprint arXiv:1909.07809. 2019 Sep 17. https://arxiv.org/abs/1909.07809