Touchstone

"How Are We Doing on Medical Segmentation?" (NeurIPS 2024): a large-scale collaborative benchmark of 3D medical-segmentation AI trained on data from 76 hospitals and tested on 11 held-out hospitals, evaluating dozens of community-submitted architectures under a common protocol — a corrective to single-center leaderboard results. JHU CCVL, companion evaluation to AbdomenAtlas.