Skip to main content

๐Ÿฉป AMOS 2022

๐Ÿฉป AMOS 2022

Abdominal multi-organ segmentation ยท CT and MRI

Official dataset sourceRelated challenge

๐Ÿ“š Publications and records

NeurIPS D&B 2022arXiv 2206.08023Zenodo DOI 10.5281/zenodo.7262581

Quick facts
Input500 CT and 100 MRI scans
Targets15 abdominal organs
Institutional originThe benchmark teamโ€™s affiliations include the University of Hong Kong, CUHK-Shenzhen and the Shenzhen Research Institute of Big Data, Longgang Central Hospital, and Longgang Peopleโ€™s Hospital. The paper describes the images as a multi-center collection; these affiliations should not be read as a complete list of acquisition hospitals. NeurIPS paper.
Official sourceAMOS 2022 challenge ยท Post-challenge Zenodo record

๐Ÿงพ Dataset overviewโ€‹

AMOS was designed to evaluate abdominal organ segmentation across multi-center, multi-vendor, multi-modality and multi-disease imaging. The dataset paper describes voxel-level annotations for 15 organs, including the spleen, kidneys, gallbladder, esophagus, liver, stomach, aorta, inferior vena cava, pancreas, adrenal glands, duodenum, bladder, and prostate/uterus.

The challenge has a CT-only task and a mixed CT/MRI task. The official Zenodo record links a labeled 500-CT/100-MRI archive and additional unlabeled data.

๐Ÿ—‚๏ธ Official split and label availabilityโ€‹

Release / cohortOfficial information
Challenge development dataTraining and validation data were released with reference labels. A later peer-reviewed description reports 300 CT and 60 MRI across these accessible sections.
Challenge test dataThe challenge portal describes test images without ground-truth labels during the challenge.
Post-challenge Zenodo v2The Zenodo record describes its 500-CT/100-MRI archive as labeled. This is broader than the original challenge-time test release; check the archive contents and version before treating test references as available.
Additional unlabeled dataLinked separately from the Zenodo record; these are not the 600 labeled scans.

The challenge split and the post-challenge archive are different release contexts. Do not infer a train/validation subdivision from a merged post-challenge package.

๐Ÿ”’ Access and termsโ€‹

AMOS 2022 Challenge benchmarked segmentation of 15 abdominal organs across CT and MRI, using data collected across multiple centers and vendors.

๐Ÿ“š Publication and citationโ€‹

Ji et al. โ€œAMOS: A Large-Scale Abdominal Multi-Organ Benchmark for Versatile Medical Image Segmentation.โ€ Advances in Neural Information Processing Systems (NeurIPS), Datasets and Benchmarks Track, vol. 35, 2022, pp. 36722โ€“36732.

The challenge portal and Zenodo archive represent different release contexts; see the split and label notes above when selecting a release.


๐Ÿ•˜ Sources last checked: 27 October 2026 ยท Back to the dataset catalog