Self-supervised learning via inter-modal reconstruction and feature projection networks for label-efficient 3D-to-2D segmentation | Publicación