CORAL: COlonography Radiology Annotation Lab

Abstract

CT colonography (CTC) interpretation hinges on three intertwined structures: the colonic lumen, intracolonic fluid pockets, and polyps. Constructing voxel-level ground truth for these structures is a persistent bottleneck for computer-aided diagnosis. We present CORAL, a browser-based CT colonography annotation environment that delivers an end-to-end workflow for joint lumen, fluid-pocket, and polyp annotation, and provides a slice-level free-text comment panel for radiologists to describe their findings. CORAL parses DICOM and NIfTI volumes entirely in the browser; all image processing runs on the client and no annotation data is uploaded to a remote server. It offers six label-aware brushes, four of which encode HU-based domain knowledge (edge-aware and a fluid brush family), four in-browser refinements (GrabCut-style, Random-Walker-style, Combined, and Fluid-aware GrabCut-style), slice-to-slice mask propagation, 3D mesh visualisation, and on-image measurement. The resulting environment pairs voxel-level CTC annotations with radiologists' free-text reasoning, supporting downstream research in dataset construction and explainable medical AI. A live demo and walkthrough video are linked from our open-source repository at https://github.com/catarina-moreira/CORAL. A demo video is publicly available at https://youtu.be/hq48kqLNLD0.