Appendix A Extensive Datasets in Touchstone 20 A.1 Construction of AbdomenAtlas 1.0 20 A.2 Domain Shift in TotalSegmentator

The naive aggregation of these public datasets results in a database with partial and incomplete labels, e.g., LiTS only had labels for the liver and its tumors, and KiTS only had labels for the kidneys and its tumors. Conversely, our AbdomenAtlas 1.0 is fully-annotated, offering detailed per-voxel labels for all 9 organs within each CT scan. We detected and removed duplicated CT scans across public datasets like LiTS and FLARE'23. Duplicate scans were identified by generating a 3D perceptual hash [83] for each image in the dataset. By comparing the similarity of these hashes, duplicates were reliably detected, a finding that was further confirmed through manual inspection of CT scans with high perceptual hash similarities. After aggregating all datasets and removing duplicates, we obtained a total of 5,195 fully-annotated CT scans. Part of TotalSegmentator is included in the AbdomenAtlas dataset (N=485), because it is contained in FLARE, one of the AbdomenAtlas constituents.