Connecting modalities: Semi-supervised segmentation and annotation of images using unaligned text corpora | Synapse