| { |
| "schema_version": "1.0.0", |
| "generated_at": "2026-09-16", |
| "verification_policy": { |
| "rule": "Every entry must be verified against its official landing page before download_by_default may be true.", |
| "unverified_fields_are_null": true, |
| "commercial_use_default": "deny" |
| }, |
| "training_use_values": [ |
| "approved_with_attribution", |
| "approved_with_attribution_share_alike", |
| "noncommercial_research_only", |
| "per_file_license_audit_required", |
| "resources_only", |
| "blocked_pending_license_review", |
| "blocked_pending_access_and_license", |
| "blocked_no_derivatives" |
| ], |
| "attribution_policy": { |
| "rule": "Every source that contributes a single frame to a published artifact MUST appear in the rendered ATTRIBUTION.md and in the model and dataset cards. scripts/push_to_hf.py builds that list from prepared-training-data/build-manifest.json counts_by_source - the sources actually used, not the whole inventory - and refuses to upload if any of them lacks a verified attribution block.", |
| "required_fields": ["title", "creators", "source_url", "license", "license_url", "citation", "modifications"], |
| "why_modifications": "CC BY 4.0 s3(a)(1)(B) and CC BY-SA 4.0 s3(a)(1)(B) require an indication that the material was modified. Every source here is modified - sliced to 2D, windowed, resized to 224px, capped per patient and relabelled into this taxonomy - so the field is never empty.", |
| "citation_verified_at": "A citation is a legal statement, not a convenience. Publishing a wrong one is itself an attribution failure, so each block carries its own verification date and the publisher refuses to upload while any contributing source is still null.", |
| "license_compatibility": { |
| "apache-2.0_weights_requires": ["approved_with_attribution"], |
| "note": "If any contributing source is share-alike or non-commercial, the weights cannot be published under Apache-2.0. push_to_hf.py computes the effective licence from the contributing set and fails rather than mislabelling the artifact." |
| } |
| }, |
| "sampling_policy": { |
| "rule": "Count patients, not frames (plan section 5). scripts/build_frame_dataset.py MUST enforce these caps before writing prepared-training-data/*.jsonl.", |
| "default_max_frames_per_patient": 40, |
| "ultrasound_frames_per_loop": { "min": 5, "max": 20, "selection": "temporal_stride" }, |
| "ct_mr_slices_per_volume": 32, |
| "split_by": ["patient", "site"], |
| "patient_groups": { |
| "lits-201": [ |
| "medmnist-plus-organamnist-224", |
| "medmnist-plus-organcmnist-224", |
| "medmnist-plus-organsmnist-224" |
| ] |
| }, |
| "patient_group_rule": "Sources sharing a patient_group are different views of the same volumes and MUST be assigned to the same split. Splitting them independently leaks patients across train and test." |
| }, |
| "sources": [ |
| { |
| "id": "medmnist-plus-organamnist-224", |
| "name": "MedMNIST+ OrganAMNIST 224px", |
| "category": "dataset", |
| "kind": "zenodo_file", |
| "url": "https://zenodo.org/records/10519652", |
| "filename": "organamnist_224.npz", |
| "revision": "3.0", |
| "md5": "50747347e05c87dd3aaf92c49f9f3170", |
| "approximate_bytes": 1803859544, |
| "destination": "open/medmnist-plus/organamnist_224", |
| "attribution": { |
| "title": "MedMNIST v2 (OrganAMNIST, 224px)", |
| "creators": "Jiancheng Yang, Rui Shi, Donglai Wei, Zequan Liu, Lin Zhao, Bilian Ke, Hanspeter Pfister, Bingbing Ni", |
| "year": 2023, |
| "source_url": "https://zenodo.org/records/10519652", |
| "license": "CC-BY-4.0", |
| "license_url": "https://creativecommons.org/licenses/by/4.0/", |
| "doi": "10.1038/s41597-022-01721-8", |
| "citation": "Yang, J., Shi, R., Wei, D., Liu, Z., Zhao, L., Ke, B., Pfister, H., Ni, B. MedMNIST v2 - A large-scale lightweight benchmark for 2D and 3D biomedical image classification. Scientific Data 10, 41 (2023).", |
| "upstream_attribution": "OrganA/C/S MNIST are derived from the Liver Tumor Segmentation Benchmark (LiTS), Bilic et al., Medical Image Analysis 84 (2023), doi 10.1016/j.media.2022.102680.", |
| "modifications": "Decoded from .npz, converted to 3-channel RGB PNG at 224px, capped per patient, and relabelled from the 11 MedMNIST class indices into the OrganScan modality/region/organ taxonomy.", |
| "citation_verified_at": "2026-09-25" |
| }, |
| "access": "open", |
| "download_by_default": true, |
| "training_use": "approved_with_attribution", |
| "license": "CC-BY-4.0", |
| "commercial_use": "permitted_with_attribution", |
| "modalities": ["CT"], |
| "label_levels": ["modality", "region", "organ"], |
| "body_regions": ["abdomen"], |
| "classes": 11, |
| "patients": 201, |
| "images": 58830, |
| "patient_group": "lits-201", |
| "sampling": { "max_frames_per_patient": 40, "raw_frames_per_patient": 293, "max_frames_total": 4000 }, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://raw.githubusercontent.com/MedMNIST/MedMNIST/main/medmnist/info.py", |
| "notes": "Axial abdominal CT slices, 11 organ classes. 34,561/6,491/17,778 train/val/test from only 115/16/70 LiTS CT scans - ~293 images per patient, so the effective sample size is 201, not 58,830. Cap hard. Day-one pipeline bootstrap per plan section 10.2. DermaMNIST is the only CC-BY-NC subset and is not used." |
| }, |
| { |
| "id": "medmnist-plus-organcmnist-224", |
| "name": "MedMNIST+ OrganCMNIST 224px", |
| "category": "dataset", |
| "kind": "zenodo_file", |
| "url": "https://zenodo.org/records/10519652", |
| "filename": "organcmnist_224.npz", |
| "revision": "3.0", |
| "md5": "050f5e875dc056f6768abf94ec9995d1", |
| "approximate_bytes": 760231860, |
| "destination": "open/medmnist-plus/organcmnist_224", |
| "attribution": { |
| "title": "MedMNIST v2 (OrganCMNIST, 224px)", |
| "creators": "Jiancheng Yang, Rui Shi, Donglai Wei, Zequan Liu, Lin Zhao, Bilian Ke, Hanspeter Pfister, Bingbing Ni", |
| "year": 2023, |
| "source_url": "https://zenodo.org/records/10519652", |
| "license": "CC-BY-4.0", |
| "license_url": "https://creativecommons.org/licenses/by/4.0/", |
| "doi": "10.1038/s41597-022-01721-8", |
| "citation": "Yang, J., Shi, R., Wei, D., Liu, Z., Zhao, L., Ke, B., Pfister, H., Ni, B. MedMNIST v2 - A large-scale lightweight benchmark for 2D and 3D biomedical image classification. Scientific Data 10, 41 (2023).", |
| "upstream_attribution": "Derived from the Liver Tumor Segmentation Benchmark (LiTS), Bilic et al., Medical Image Analysis 84 (2023), doi 10.1016/j.media.2022.102680.", |
| "modifications": "Decoded from .npz, converted to 3-channel RGB PNG at 224px, capped per patient, and relabelled into the OrganScan taxonomy.", |
| "citation_verified_at": "2026-09-25" |
| }, |
| "access": "open", |
| "download_by_default": true, |
| "training_use": "approved_with_attribution", |
| "license": "CC-BY-4.0", |
| "commercial_use": "permitted_with_attribution", |
| "modalities": ["CT"], |
| "label_levels": ["modality", "region", "organ"], |
| "body_regions": ["abdomen"], |
| "classes": 11, |
| "patients": 201, |
| "images": 23583, |
| "patient_group": "lits-201", |
| "sampling": { "max_frames_per_patient": 40, "raw_frames_per_patient": 117, "max_frames_total": 4000 }, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://raw.githubusercontent.com/MedMNIST/MedMNIST/main/medmnist/info.py", |
| "notes": "Coronal view of the same 201 LiTS volumes as OrganAMNIST. 12,975/2,392/8,216 train/val/test. Shares patients with OrganA/S via patient_group lits-201 - must be assigned to the same split." |
| }, |
| { |
| "id": "medmnist-plus-organsmnist-224", |
| "name": "MedMNIST+ OrganSMNIST 224px", |
| "category": "dataset", |
| "kind": "zenodo_file", |
| "url": "https://zenodo.org/records/10519652", |
| "filename": "organsmnist_224.npz", |
| "revision": "3.0", |
| "md5": "b354719e553fbbb2513d5533f52a4cb1", |
| "approximate_bytes": 802713625, |
| "destination": "open/medmnist-plus/organsmnist_224", |
| "attribution": { |
| "title": "MedMNIST v2 (OrganSMNIST, 224px)", |
| "creators": "Jiancheng Yang, Rui Shi, Donglai Wei, Zequan Liu, Lin Zhao, Bilian Ke, Hanspeter Pfister, Bingbing Ni", |
| "year": 2023, |
| "source_url": "https://zenodo.org/records/10519652", |
| "license": "CC-BY-4.0", |
| "license_url": "https://creativecommons.org/licenses/by/4.0/", |
| "doi": "10.1038/s41597-022-01721-8", |
| "citation": "Yang, J., Shi, R., Wei, D., Liu, Z., Zhao, L., Ke, B., Pfister, H., Ni, B. MedMNIST v2 - A large-scale lightweight benchmark for 2D and 3D biomedical image classification. Scientific Data 10, 41 (2023).", |
| "upstream_attribution": "Derived from the Liver Tumor Segmentation Benchmark (LiTS), Bilic et al., Medical Image Analysis 84 (2023), doi 10.1016/j.media.2022.102680.", |
| "modifications": "Decoded from .npz, converted to 3-channel RGB PNG at 224px, capped per patient, and relabelled into the OrganScan taxonomy.", |
| "citation_verified_at": "2026-09-25" |
| }, |
| "access": "open", |
| "download_by_default": true, |
| "training_use": "approved_with_attribution", |
| "license": "CC-BY-4.0", |
| "commercial_use": "permitted_with_attribution", |
| "modalities": ["CT"], |
| "label_levels": ["modality", "region", "organ"], |
| "body_regions": ["abdomen"], |
| "classes": 11, |
| "patients": 201, |
| "images": 25211, |
| "patient_group": "lits-201", |
| "sampling": { "max_frames_per_patient": 40, "raw_frames_per_patient": 125, "max_frames_total": 4000 }, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://raw.githubusercontent.com/MedMNIST/MedMNIST/main/medmnist/info.py", |
| "notes": "Sagittal view of the same 201 LiTS volumes as OrganAMNIST. 13,932/2,452/8,827 train/val/test. Shares patients with OrganA/C via patient_group lits-201 - must be assigned to the same split." |
| }, |
| { |
| "id": "medmnist-plus-breastmnist-224", |
| "name": "MedMNIST+ BreastMNIST 224px", |
| "category": "dataset", |
| "kind": "zenodo_file", |
| "url": "https://zenodo.org/records/10519652", |
| "filename": "breastmnist_224.npz", |
| "revision": "3.0", |
| "md5": "b56378a6eefa9fed602bb16d192d4c8b", |
| "approximate_bytes": 30903564, |
| "destination": "open/medmnist-plus/breastmnist_224", |
| "attribution": { |
| "title": "MedMNIST v2 (BreastMNIST, 224px)", |
| "creators": "Jiancheng Yang, Rui Shi, Donglai Wei, Zequan Liu, Lin Zhao, Bilian Ke, Hanspeter Pfister, Bingbing Ni", |
| "year": 2023, |
| "source_url": "https://zenodo.org/records/10519652", |
| "license": "CC-BY-4.0", |
| "license_url": "https://creativecommons.org/licenses/by/4.0/", |
| "doi": "10.1038/s41597-022-01721-8", |
| "citation": "Yang, J., Shi, R., Wei, D., Liu, Z., Zhao, L., Ke, B., Pfister, H., Ni, B. MedMNIST v2 - A large-scale lightweight benchmark for 2D and 3D biomedical image classification. Scientific Data 10, 41 (2023).", |
| "upstream_attribution": "Derived from the Breast Ultrasound Images dataset (BUSI), Al-Dhabyani et al., Data in Brief 28 (2020), doi 10.1016/j.dib.2019.104863. The upstream BUSI licence is itself unresolved - see the busi-original entry - so this subset is used only through MedMNIST's CC BY 4.0 release.", |
| "modifications": "Pathology class labels discarded; every frame relabelled with a constant modality=US, region=chest, organ=BREAST anatomy label. Converted to 3-channel RGB PNG at 224px.", |
| "citation_verified_at": "2026-09-25" |
| }, |
| "access": "open", |
| "download_by_default": true, |
| "training_use": "approved_with_attribution", |
| "license": "CC-BY-4.0", |
| "commercial_use": "permitted_with_attribution", |
| "modalities": ["US"], |
| "label_levels": ["modality", "region"], |
| "body_regions": ["breast"], |
| "classes": 2, |
| "patients": 600, |
| "images": 780, |
| "sampling": { "max_frames_per_patient": 40, "raw_frames_per_patient": 2 }, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://raw.githubusercontent.com/MedMNIST/MedMNIST/main/medmnist/info.py", |
| "notes": "546/78/156 train/val/test. Derived from BUSI. Class labels are benign/malignant pathology, NOT anatomy - use only for the modality=US and region=breast levels. Already ~1.3 images per patient, so the cap never binds. Upstream BUSI licence is unresolved; see the busi-original entry." |
| }, |
| { |
| "id": "medmnist-plus-chestmnist-224", |
| "name": "MedMNIST+ ChestMNIST 224px", |
| "category": "dataset", |
| "kind": "zenodo_file", |
| "url": "https://zenodo.org/records/10519652", |
| "filename": "chestmnist_224.npz", |
| "revision": "3.0", |
| "md5": "45bd33e6f06c3e8cdb481c74a89152aa", |
| "approximate_bytes": 3900000000, |
| "destination": "open/medmnist-plus/chestmnist_224", |
| "attribution": { |
| "title": "MedMNIST v2 (ChestMNIST, 224px)", |
| "creators": "Jiancheng Yang, Rui Shi, Donglai Wei, Zequan Liu, Lin Zhao, Bilian Ke, Hanspeter Pfister, Bingbing Ni", |
| "year": 2023, |
| "source_url": "https://zenodo.org/records/10519652", |
| "license": "CC-BY-4.0", |
| "license_url": "https://creativecommons.org/licenses/by/4.0/", |
| "doi": "10.1038/s41597-022-01721-8", |
| "citation": "Yang, J., Shi, R., Wei, D., Liu, Z., Zhao, L., Ke, B., Pfister, H., Ni, B. MedMNIST v2 - A large-scale lightweight benchmark for 2D and 3D biomedical image classification. Scientific Data 10, 41 (2023).", |
| "upstream_attribution": "Derived from NIH ChestX-ray14, Wang et al., CVPR 2017, doi 10.1109/CVPR.2017.369.", |
| "modifications": "The 14 pathology labels are discarded; every frame is relabelled with a constant modality=XR, region=chest label and the organ head fully masked. Subsampled well below the published 112,120 images to stay inside the per-region budget.", |
| "citation_verified_at": "2026-09-25" |
| }, |
| "access": "open", |
| "download_by_default": false, |
| "training_use": "approved_with_attribution", |
| "license": "CC-BY-4.0", |
| "commercial_use": "permitted_with_attribution", |
| "modalities": ["XR"], |
| "label_levels": ["modality", "region"], |
| "body_regions": ["chest"], |
| "classes": 14, |
| "patients": 30805, |
| "images": 112120, |
| "sampling": { "max_frames_per_patient": 4, "raw_frames_per_patient": 4, "max_frames_total": 8000 }, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://raw.githubusercontent.com/MedMNIST/MedMNIST/main/medmnist/info.py", |
| "notes": "78,468/11,219/22,433 train/val/test from 30,805 unique patients (NIH ChestXray14). Only source of XR modality so far. 14 labels are pathologies, not organs - region level only. A per-patient cap cannot bound 30,805 patients: uncapped this source was 112,094 frames, 43% of the whole corpus, for one region and no organ labels. max_frames_total thins it deterministically on the content hash so the subset keeps the upstream split proportions." |
| }, |
| { |
| "id": "msu-abdominal-ultrasound", |
| "name": "Abdominal Ultrasound Image Dataset for Organ Classification and Disease Detection", |
| "category": "dataset", |
| "kind": "manual_access", |
| "url": "https://scholarsjunction.msstate.edu/context/research-data/article/1004/type/native/viewcontent", |
| "landing_page": "https://scholarsjunction.msstate.edu/research-data/5/", |
| "filename": "dataset_publish_2026_June.zip", |
| "revision": "10.54718/LZXF6315", |
| "md5": "5be76e9cbce5ca994efc97e07e194dee", |
| "approximate_bytes": 1494219274, |
| "destination": "open/msu-abdominal-ultrasound", |
| "attribution": { |
| "title": "Abdominal Ultrasound Image Dataset for Organ Classification and Disease Detection", |
| "creators": "Sifat Zina Karim", |
| "creator_orcid": "https://orcid.org/0009-0005-8413-4149", |
| "year": 2025, |
| "source_url": "https://scholarsjunction.msstate.edu/research-data/5/", |
| "license": "CC-BY-4.0", |
| "license_url": "https://creativecommons.org/licenses/by/4.0/", |
| "doi": "10.54718/LZXF6315", |
| "citation": "Karim, Sifat Zina. Abdominal Ultrasound Image Dataset for Organ Classification and Disease Detection (2025). Research Data 5, Mississippi State University Scholars Junction. https://doi.org/10.54718/LZXF6315", |
| "related_publication": "Karim, S. Z., Biswas, S., Ball, J. Learnable 2D Gaussian filters for computationally efficient abdominal organ classification. Proc. SPIE 13458, Real-Time Image Processing and Deep Learning 2025, 1345802 (2025). https://doi.org/10.1117/12.3053222", |
| "data_provenance": "5,468 unique images from 563 patients at MH Samorita Medical College and Hospital, Dhaka, Bangladesh, read by two radiologists. Collected under IERB ethical clearance and de-identified before release.", |
| "upstream_attribution": null, |
| "modifications": "Only organ_classification_1 and organ_classification_2 are read; images deduplicated by content hash across folders; patient ids joined from Patient_Wise/Metadata.xlsx; resized to 224px and relabelled from the 10 folder names into the OrganScan taxonomy, with kidney laterality masked because the source does not state it.", |
| "citation_verified_at": "2026-09-25" |
| }, |
| "access": "open_manual_browser_download", |
| "download_by_default": false, |
| "training_use": "approved_with_attribution", |
| "license": "CC-BY-4.0", |
| "commercial_use": "permitted_with_attribution", |
| "modalities": ["US"], |
| "label_levels": ["modality", "region", "organ"], |
| "body_regions": ["abdomen"], |
| "classes": 10, |
| "patients": 563, |
| "images": 5468, |
| "sampling": { "max_frames_per_patient": 20, "raw_frames_per_patient": 10 }, |
| "archive_contents": { |
| "verified_at": "2026-09-17", |
| "file_entries": 13088, |
| "unique_images": 5468, |
| "distinct_basenames": 2905, |
| "duplication_factor": 2.39, |
| "folders": { |
| "organ_classification_1": 2775, |
| "organ_classification_2": 1334, |
| "organ_classification+anomaly_detection": 2920, |
| "Anomaly_detection_1": 2808, |
| "Anomaly_detection_2": 925, |
| "Patient_Wise": 2326 |
| }, |
| "metadata_files": [ |
| "dataset publish/Patient_Wise/Metadata.xlsx", |
| "dataset publish/Patient_Wise/Update Version 01_USG.txt" |
| ] |
| }, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://scholarsjunction.msstate.edu/research-data/5/", |
| "notes": "HIGHEST-VALUE SOURCE. The only verified open, commercially usable, patient-attributed abdominal ULTRASOUND organ-classification set found. MANUAL STEP: the host (bepress/Digital Commons) returns HTTP 403 to scripted requests, so use the Download button on the landing page in a browser. Do not work around the bot filter. LEAKAGE HAZARD verified 2026-09-17: the archive holds 13,088 image entries but only 5,468 unique images - the six folders are reorganizations of the same pictures, so a naive glob duplicates 2.39x ACROSS folders and a random split puts the same image in train and test. Worse, there are only 2,905 distinct basenames, so filenames collide between folders and CANNOT be used as an identity key - deduplicate by content hash. Build from organ_classification_1 + organ_classification_2 for organ labels and take patient IDs from Patient_Wise; never concatenate all six folders. Classes: aorta, gallbladder, hepatic vein, kidney, liver, ovary, pancreas, portal vein, spleen, urinary system. Karim, Biswas & Ball, Proc. SPIE 13458 (2025). Single site (MH Samorita, Dhaka): cannot serve as the multi-vendor test set on its own. Do NOT use the Hugging Face copy - see rejected_sources." |
| }, |
| { |
| "id": "fetal-planes-db", |
| "name": "FETAL_PLANES_DB - Common maternal-fetal ultrasound planes", |
| "category": "dataset", |
| "kind": "zenodo_file", |
| "url": "https://zenodo.org/records/3904280", |
| "filename": "FETAL_PLANES_ZENODO.zip", |
| "revision": "1.0", |
| "md5": "2a5fcc2cefb789bcc0f6c1f73e0ea43f", |
| "approximate_bytes": 2088522169, |
| "destination": "open/fetal-planes-db", |
| "attribution": { |
| "title": "FETAL_PLANES_DB: Common maternal-fetal ultrasound images", |
| "creators": "Xavier P. Burgos-Artizzu, David Coronado-Gutierrez, Brenda Valenzuela-Alcaraz, Elisenda Bonet-Carne, Elisenda Eixarch, Fatima Crispi, Eduard Gratacós", |
| "year": 2020, |
| "source_url": "https://zenodo.org/records/3904280", |
| "license": "CC-BY-4.0", |
| "license_url": "https://creativecommons.org/licenses/by/4.0/", |
| "doi": "10.1038/s41598-020-67076-5", |
| "citation": "Burgos-Artizzu, X. P., Coronado-Gutierrez, D., Valenzuela-Alcaraz, B., Bonet-Carne, E., Eixarch, E., Crispi, F., Gratacós, E. Evaluation of deep convolutional neural networks for automatic classification of common maternal fetal ultrasound planes. Scientific Reports 10, 10200 (2020).", |
| "upstream_attribution": null, |
| "modifications": "Resized to 224px; the six plane classes are mapped to region labels only - fetal anatomy is not claimed, so the organ head is masked for every fetal row.", |
| "citation_verified_at": "2026-09-25" |
| }, |
| "access": "open", |
| "download_by_default": true, |
| "training_use": "approved_with_attribution", |
| "license": "CC-BY-4.0", |
| "commercial_use": "permitted_with_attribution", |
| "modalities": ["US"], |
| "label_levels": ["modality", "region"], |
| "body_regions": ["obstetric"], |
| "classes": 6, |
| "patients": null, |
| "images": null, |
| "sampling": { "max_frames_per_patient": 20, "raw_frames_per_patient": null }, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://zenodo.org/records/3904280", |
| "notes": "Planes: abdomen, brain, femur, thorax, maternal cervix, other; brain sub-split into trans-thalamic/trans-cerebellum/trans-ventricular. Image and patient counts are NOT stated on the Zenodo record - read FETAL_PLANES_DB_data.csv after download and write the verified figures back here. CSV carries patient number, US machine and operator - gives a free vendor axis for the site/vendor split. 'Abdomen' here is FETAL abdomen; it must not be mapped onto the adult abdomen region code." |
| }, |
| { |
| "id": "totalsegmentator-ct-sample", |
| "name": "TotalSegmentator CT 102-subject sample", |
| "category": "dataset", |
| "kind": "zenodo_file", |
| "url": "https://zenodo.org/records/10047263", |
| "filename": "Totalsegmentator_dataset_small_v201.zip", |
| "revision": "10.5281/zenodo.8367169", |
| "md5": "6b5524af4b15e6ba06ef2d700c0c73e0", |
| "approximate_bytes": 3244617817, |
| "destination": "open/totalsegmentator-ct-sample", |
| "attribution": { |
| "title": "TotalSegmentator CT dataset (102-subject sample, v2.0.1)", |
| "creators": "Jakob Wasserthal, Hanns-Christian Breit, Manfred T. Meyer, Maurice Pradella, Daniel Hinck, Alexander W. Sauter, Tobias Heye, Daniel T. Boll, Joshy Cyriac, Shan Yang, Michael Bach, Martin Segeroth", |
| "year": 2023, |
| "source_url": "https://zenodo.org/records/10047263", |
| "license": "CC-BY-4.0", |
| "license_url": "https://creativecommons.org/licenses/by/4.0/", |
| "doi": "10.1148/ryai.230024", |
| "citation": "Wasserthal, J. et al. TotalSegmentator: Robust Segmentation of 104 Anatomic Structures in CT Images. Radiology: Artificial Intelligence 5(5) (2023).", |
| "upstream_attribution": null, |
| "modifications": "Volumes sliced axially at fixed stride into 2D frames, windowed and resized to 224px; the 117 segmentation masks are reduced to the 30-slot OrganScan organ taxonomy and used as exhaustive multi-label targets.", |
| "citation_verified_at": "2026-09-16" |
| }, |
| "access": "open", |
| "download_by_default": true, |
| "superseded_by": "totalsegmentator-ct-v300", |
| "superseded_note": "Verified 2026-09-24 that all 102 sample subjects are also in v3.0.0. Building both attributes those 102 subjects to the older v2.0.1 release and leaves v300 contributing only 1,834 of its 1,939 subjects after content-hash dedup. Excluded from default builds; pass --source totalsegmentator-ct-sample explicitly for fast iteration.", |
| "training_use": "approved_with_attribution", |
| "license": "CC-BY-4.0", |
| "commercial_use": "permitted_with_attribution", |
| "modalities": ["CT"], |
| "label_levels": ["modality", "region", "organ"], |
| "body_regions": ["head", "neck", "chest", "abdomen", "pelvis", "extremity"], |
| "classes": 117, |
| "patients": 102, |
| "images": null, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://zenodo.org/records/22688904", |
| "notes": "102-subject sample of the v2.0.1 release. Use this to develop the mask-to-multi-label slicer before pulling 37 GB. DOI 10.5281/zenodo.8367169 is a version alias that resolves to canonical record 10047263 - building a download URL from 8367169 returns 404." |
| }, |
| { |
| "id": "totalsegmentator-ct-v300", |
| "name": "TotalSegmentator CT v3.0.0", |
| "category": "dataset", |
| "kind": "zenodo_file", |
| "url": "https://zenodo.org/records/22688904", |
| "filename": "Totalsegmentator_dataset_v300.zip", |
| "revision": "3.0.0", |
| "md5": "0728beea76475e92cdb3160e228265d1", |
| "approximate_bytes": 37415622057, |
| "destination": "open/totalsegmentator-ct-v300", |
| "attribution": { |
| "title": "TotalSegmentator CT dataset v3.0.0", |
| "creators": "Jakob Wasserthal, Hanns-Christian Breit, Manfred T. Meyer, Maurice Pradella, Daniel Hinck, Alexander W. Sauter, Tobias Heye, Daniel T. Boll, Joshy Cyriac, Shan Yang, Michael Bach, Martin Segeroth", |
| "year": 2023, |
| "source_url": "https://zenodo.org/records/22688904", |
| "license": "CC-BY-4.0", |
| "license_url": "https://creativecommons.org/licenses/by/4.0/", |
| "doi": "10.1148/ryai.230024", |
| "citation": "Wasserthal, J. et al. TotalSegmentator: Robust Segmentation of 104 Anatomic Structures in CT Images. Radiology: Artificial Intelligence 5(5) (2023).", |
| "upstream_attribution": null, |
| "modifications": "Volumes sliced axially at fixed stride into 2D frames, windowed and resized to 224px; the 117 segmentation masks are reduced to the 30-slot OrganScan organ taxonomy and used as exhaustive multi-label targets.", |
| "citation_verified_at": "2026-09-16" |
| }, |
| "access": "open", |
| "download_by_default": false, |
| "training_use": "approved_with_attribution", |
| "license": "CC-BY-4.0", |
| "commercial_use": "permitted_with_attribution", |
| "modalities": ["CT"], |
| "label_levels": ["modality", "region", "organ"], |
| "body_regions": ["head", "neck", "chest", "abdomen", "pelvis", "extremity"], |
| "classes": 117, |
| "patients": 1939, |
| "images": null, |
| "sampling": { "slices_per_volume": 32, "expected_frames": 62048 }, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://zenodo.org/records/22688904", |
| "notes": "The exhaustive-multi-label backbone of the CT side: masks give true multi-label organ targets, which is exactly what the sigmoid heads need. v3.0.0 adds 291 pediatric CTs over v2.0.1 (1,228 subjects, record 10047292, md5 fe250e5718e0a3b5df4c4ea9d58a62fe). Opt-in because of size. Wasserthal et al., doi 10.1148/ryai.230024." |
| }, |
| { |
| "id": "amos22", |
| "name": "AMOS22 abdominal multi-organ segmentation (labelled)", |
| "category": "dataset", |
| "kind": "zenodo_file", |
| "url": "https://zenodo.org/records/7262581", |
| "filename": "amos22.zip", |
| "revision": "v2", |
| "md5": "67717b2a483ac0744c89c3016b7aaef7", |
| "approximate_bytes": 24234336519, |
| "companion_files": [ |
| { |
| "filename": "labeled_data_meta_0000_0599.csv", |
| "md5": "524fdc33d32ca06b37ae4b5ff4ddf5f4", |
| "bytes": 35206, |
| "url": "https://zenodo.org/records/7262581/files/labeled_data_meta_0000_0599.csv?download=1", |
| "notes": "Per-case metadata for the 600 labelled scans; needed for patient- and scanner-aware splits." |
| } |
| ], |
| "destination": "open/amos22", |
| "attribution": { |
| "title": "AMOS22: A Large-Scale Abdominal Multi-Organ Benchmark for Versatile Medical Image Segmentation", |
| "creators": "Yuanfeng Ji, Haotian Bai, Chongjian Ge, Jie Yang, Ye Zhu, Ruimao Zhang, Zhen Li, Lingyan Zhang, Wanling Ma, Xiang Wan, Ping Luo", |
| "year": 2022, |
| "source_url": "https://zenodo.org/records/7262581", |
| "license": "CC-BY-4.0", |
| "license_url": "https://creativecommons.org/licenses/by/4.0/", |
| "doi": "10.5281/zenodo.7262581", |
| "citation": "Ji, Y. et al. AMOS: A Large-Scale Abdominal Multi-Organ Benchmark for Versatile Medical Image Segmentation. Advances in Neural Information Processing Systems 35, Datasets and Benchmarks Track (2022).", |
| "upstream_attribution": null, |
| "modifications": "Volumes sliced axially at fixed stride, windowed and resized to 224px; the 15 label indices are mapped to the OrganScan taxonomy, with the merged prostate/uterus index masked because the source does not publish per-case sex.", |
| "citation_verified_at": "2026-09-25" |
| }, |
| "access": "open", |
| "download_by_default": false, |
| "training_use": "approved_with_attribution", |
| "license": "CC-BY-4.0", |
| "commercial_use": "permitted_with_attribution", |
| "modalities": ["CT", "MR"], |
| "label_levels": ["modality", "region", "organ"], |
| "body_regions": ["abdomen"], |
| "classes": 15, |
| "patients": 600, |
| "images": null, |
| "sampling": { "slices_per_volume": 32, "expected_frames": 19200 }, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://zenodo.org/records/7262581", |
| "notes": "CORRECTION to plan section 3.2: the licence is CC-BY-4.0, not CC-BY-NC-SA. Commercial use is permitted. 500 CT + 100 MR, 15 organ classes. Only verified source of MR abdominal organ labels. labelsTs is withheld by the challenge, so imagesTs is unusable for supervised training. Unlabelled companion records 7262757 / 7295661 / 7295816 have NOT been licence-verified." |
| }, |
| { |
| "id": "medical-segmentation-decathlon", |
| "name": "Medical Segmentation Decathlon", |
| "category": "dataset", |
| "kind": "manual_access", |
| "url": "http://medicaldecathlon.com/", |
| "filename": null, |
| "revision": null, |
| "md5": null, |
| "approximate_bytes": null, |
| "destination": "pending-access/medical-segmentation-decathlon", |
| "attribution": { |
| "title": "The Medical Segmentation Decathlon", |
| "creators": "Michela Antonelli, Annika Reinke, Spyridon Bakas, Keyvan Farahani, Annette Kopp-Schneider, Bennett A. Landman, Geert Litjens, Bjoern Menze, Olaf Ronneberger, Ronald M. Summers, Bram van Ginneken, Michel Bilello, Patrick Bilic, Patrick F. Christ, Richard K. G. Do, Marc J. Gollub, Stephan H. Heckers, Henkjan Huisman, William R. Jarnagin and others", |
| "year": 2022, |
| "source_url": "http://medicaldecathlon.com/", |
| "license": "CC-BY-SA-4.0", |
| "license_url": "https://creativecommons.org/licenses/by-sa/4.0/", |
| "doi": "10.1038/s41467-022-30695-9", |
| "citation": "Antonelli, M. et al. The Medical Segmentation Decathlon. Nature Communications 13, 4128 (2022).", |
| "upstream_attribution": null, |
| "modifications": "Volumes sliced axially at fixed stride and resized to 224px; each task's single organ label is mapped to the OrganScan taxonomy and treated as non-exhaustive, so unlabelled organs are masked rather than counted as negatives.", |
| "share_alike_warning": "CC BY-SA 4.0. If share-alike propagates to trained weights, every artifact built with this source must itself be released CC BY-SA 4.0 - it cannot be Apache-2.0. push_to_hf.py enforces this rather than mislabelling the artifact.", |
| "citation_verified_at": null |
| }, |
| "access": "open_but_unresolved_host", |
| "download_by_default": false, |
| "training_use": "approved_with_attribution_share_alike", |
| "license": "CC-BY-SA-4.0", |
| "commercial_use": "permitted_with_share_alike_review", |
| "modalities": ["CT", "MR"], |
| "label_levels": ["modality", "region", "organ"], |
| "body_regions": ["head", "chest", "abdomen", "pelvis"], |
| "classes": null, |
| "patients": 2633, |
| "images": null, |
| "verified_at": "2026-09-16", |
| "verified_from": "http://medicaldecathlon.com/", |
| "notes": "10 tasks, 2,633 volumes. Share-alike: get written guidance on whether CC-BY-SA propagates to model weights before this enters a shippable run. Download host (AWS/Drive mirror) did not resolve during verification - resolve manually, then pin it here as a zenodo_file/http_file entry." |
| }, |
| { |
| "id": "rocov2", |
| "name": "ROCOv2 - Radiology Objects in COntext v2", |
| "category": "dataset", |
| "kind": "zenodo_file", |
| "url": "https://zenodo.org/records/10821435", |
| "filename": null, |
| "revision": "2.0.1", |
| "md5": null, |
| "approximate_bytes": 6400000000, |
| "destination": "quarantine/rocov2", |
| "attribution": { |
| "title": "ROCOv2: Radiology Objects in COntext Version 2", |
| "creators": "Johannes Ruckert, Louise Bloch, Raphael Brungel, Ahmad Idrissi-Yaghir, Henning Schafer, Cynthia S. Schmidt, Sven Koitka, Obioma Pelka, Asma Ben Abacha, Alba Garcia Seco de Herrera, Henning Muller, Peter A. Horn, Felix Nensa, Christoph M. Friedrich", |
| "year": 2024, |
| "source_url": "https://zenodo.org/records/10821435", |
| "license": "CC-BY-NC-4.0", |
| "license_url": "https://creativecommons.org/licenses/by-nc/4.0/", |
| "doi": "10.1038/s41597-024-03496-6", |
| "citation": "Ruckert, J. et al. ROCOv2: Radiology Objects in COntext Version 2, an Updated Multimodal Image Dataset. Scientific Data 11, 688 (2024).", |
| "upstream_attribution": null, |
| "modifications": "Pseudo-labelled with a teacher model and filtered to a confident subset. Teacher and ablation experiments only.", |
| "noncommercial_warning": "CC BY-NC 4.0. Weights trained on this source may not be used commercially, whether directly or through distillation. It may never enter a shipped run.", |
| "citation_verified_at": null |
| }, |
| "access": "open", |
| "download_by_default": false, |
| "training_use": "noncommercial_research_only", |
| "license": "CC-BY-NC-4.0", |
| "commercial_use": "prohibited", |
| "modalities": ["CT", "MR", "XR", "US", "OTHER"], |
| "label_levels": ["modality", "region"], |
| "body_regions": null, |
| "classes": 1947, |
| "patients": null, |
| "images": 79789, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://zenodo.org/records/10821435", |
| "notes": "Caption/CUI pairs for the BiomedCLIP pseudo-label bootstrap in plan section 3.2. NON-COMMERCIAL: usable for ablations and teacher experiments only, never for shippable weights. Ships license_information.csv with per-image provenance. 59,958/9,904/9,927 train/valid/test." |
| }, |
| { |
| "id": "radimagenet", |
| "name": "RadImageNet", |
| "category": "dataset", |
| "kind": "manual_access", |
| "url": "https://www.radimagenet.com/", |
| "filename": null, |
| "revision": null, |
| "md5": null, |
| "approximate_bytes": null, |
| "destination": "pending-access/radimagenet", |
| "access": "registration_and_agreement_required", |
| "download_by_default": false, |
| "training_use": "blocked_pending_access_and_license", |
| "license": null, |
| "commercial_use": "separate_commercial_licence_offered", |
| "modalities": ["CT", "MR", "US"], |
| "label_levels": ["modality", "region"], |
| "body_regions": null, |
| "classes": 165, |
| "patients": null, |
| "images": null, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://www.radimagenet.com/", |
| "notes": "Gates the strongest teacher (plan section 6 Stage A) - the only large public backbone trained on ultrasound. SUBMIT THE REQUEST NOW; it is the long-lead item. Licence text is NOT published and the T&C page 404s - obtain the DUA in writing. Size claims conflict: site says 5M key images / 1.35M annotated per the GitHub README. The MIT licence on github.com/BMEII-AI/RadImageNet covers the CODE AND PRETRAINED WEIGHTS ONLY, not the image database." |
| }, |
| { |
| "id": "camus", |
| "name": "CAMUS cardiac ultrasound", |
| "category": "dataset", |
| "kind": "manual_access", |
| "url": "https://humanheart-project.creatis.insa-lyon.fr/database/#collection/6373703d73e9f0047faa1bc8", |
| "filename": null, |
| "revision": null, |
| "md5": null, |
| "approximate_bytes": null, |
| "destination": "pending-access/camus", |
| "access": "registration_required", |
| "download_by_default": false, |
| "training_use": "blocked_pending_license_review", |
| "license": null, |
| "commercial_use": "unverified", |
| "modalities": ["US"], |
| "label_levels": ["modality", "region"], |
| "body_regions": ["cardiac"], |
| "classes": 2, |
| "patients": 500, |
| "images": null, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://www.creatis.insa-lyon.fr/Challenge/camus/databases.html", |
| "notes": "Best available open-ish cardiac echo source now that EchoNet is blocked. A4C + A2C views, GE Vivid E95. NO licence text is published on either the challenge site or the Girder collection - treat as unusable until CREATIS confirms terms in writing. Cite Leclerc et al., IEEE TMI 38(9):2198-2210, 2019." |
| }, |
| { |
| "id": "echonet-dynamic", |
| "name": "EchoNet-Dynamic", |
| "category": "dataset", |
| "kind": "manual_access", |
| "url": "https://echonet.github.io/dynamic/", |
| "filename": null, |
| "revision": null, |
| "md5": null, |
| "approximate_bytes": null, |
| "destination": "pending-access/echonet-dynamic", |
| "access": "signed_research_use_agreement", |
| "download_by_default": false, |
| "training_use": "blocked_no_derivatives", |
| "license": "Stanford EchoNet-Dynamic Dataset Research Use Agreement", |
| "commercial_use": "prohibited", |
| "modalities": ["US"], |
| "label_levels": ["modality", "region"], |
| "body_regions": ["cardiac"], |
| "classes": null, |
| "patients": null, |
| "images": 10030, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://stanford.redivis.com/datasets/66s1-2hsmzj5rn", |
| "notes": "HARD BLOCK. The RUA prohibits commercial use, redistribution, AND creation of derivative works - trained weights are a derivative work. Frames are 112x112, below the 224 input size anyway. Do not download into this repo. Listed so the decision is recorded, not repeated." |
| }, |
| { |
| "id": "echonet-lvh", |
| "name": "EchoNet-LVH", |
| "category": "dataset", |
| "kind": "manual_access", |
| "url": "https://echonet.github.io/lvh/", |
| "filename": null, |
| "revision": null, |
| "md5": null, |
| "approximate_bytes": null, |
| "destination": "pending-access/echonet-lvh", |
| "access": "signed_research_use_agreement", |
| "download_by_default": false, |
| "training_use": "blocked_no_derivatives", |
| "license": "Stanford EchoNet-LVH Dataset Research Use Agreement", |
| "commercial_use": "prohibited", |
| "modalities": ["US"], |
| "label_levels": ["modality", "region"], |
| "body_regions": ["cardiac"], |
| "classes": null, |
| "patients": null, |
| "images": 12000, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://stanford.redivis.com/datasets/cchq-0srz1fy9a", |
| "notes": "HARD BLOCK, same RUA terms as EchoNet-Dynamic. PLAX views, native resolution." |
| }, |
| { |
| "id": "busi-original", |
| "name": "BUSI - Breast Ultrasound Images Dataset", |
| "category": "dataset", |
| "kind": "manual_access", |
| "url": "https://www.kaggle.com/datasets/aryashah2k/breast-ultrasound-images-dataset", |
| "filename": null, |
| "revision": null, |
| "md5": null, |
| "approximate_bytes": 266190000, |
| "destination": "quarantine/busi", |
| "access": "mirror_only", |
| "download_by_default": false, |
| "training_use": "blocked_pending_license_review", |
| "license": null, |
| "commercial_use": "unverified", |
| "modalities": ["US"], |
| "label_levels": ["modality", "region"], |
| "body_regions": ["breast"], |
| "classes": 3, |
| "patients": 600, |
| "images": 780, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://www.kaggle.com/datasets/aryashah2k/breast-ultrasound-images-dataset", |
| "notes": "The CC0 claim comes from a third-party Kaggle re-uploader, NOT the authors. The Cairo University source page and the Data in Brief article (28:104863, 2020) could not be reached to confirm terms. Multiple HF re-uploads assert mutually contradictory licences. BreastMNIST is the licence-clean route to the same imagery. Also carries documented duplicate/leakage issues." |
| }, |
| { |
| "id": "tn3k", |
| "name": "TN3K thyroid nodule ultrasound", |
| "category": "dataset", |
| "kind": "manual_access", |
| "url": "https://github.com/haifangong/TRFE-Net-for-thyroid-nodule-segmentation", |
| "filename": null, |
| "revision": null, |
| "md5": null, |
| "approximate_bytes": null, |
| "destination": "quarantine/tn3k", |
| "access": "third_party_drive", |
| "download_by_default": false, |
| "training_use": "blocked_pending_license_review", |
| "license": null, |
| "commercial_use": "unverified", |
| "modalities": ["US"], |
| "label_levels": ["modality", "region", "organ"], |
| "body_regions": ["neck"], |
| "classes": 2, |
| "patients": null, |
| "images": 3493, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://github.com/haifangong/TRFE-Net-for-thyroid-nodule-segmentation", |
| "notes": "Would be the only thyroid organ coverage. The repo MIT licence covers the CODE only; no data licence is published. Distributed via Google Drive / Baidu Pan (code 'trfe'), which is not a pinnable, auditable host." |
| }, |
| { |
| "id": "ddti", |
| "name": "DDTI thyroid ultrasound", |
| "category": "dataset", |
| "kind": "manual_access", |
| "url": "http://cimalab.unal.edu.co/applications/thyroid/", |
| "filename": null, |
| "revision": null, |
| "md5": null, |
| "approximate_bytes": 18750000, |
| "destination": "quarantine/ddti", |
| "access": "source_site_unavailable", |
| "download_by_default": false, |
| "training_use": "blocked_pending_license_review", |
| "license": null, |
| "commercial_use": "unverified", |
| "modalities": ["US"], |
| "label_levels": ["modality", "region", "organ"], |
| "body_regions": ["neck"], |
| "classes": null, |
| "patients": 99, |
| "images": 134, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://www.kaggle.com/datasets/dasmehdixtr/ddti-thyroid-ultrasound-images", |
| "notes": "CIMALAB source site did not respond. Only Kaggle mirrors remain, with licence 'Other'. 134 images is below the plan section 5 minimum-viable-demo bar anyway." |
| }, |
| { |
| "id": "mmotu", |
| "name": "MMOTU ovarian tumour ultrasound", |
| "category": "dataset", |
| "kind": "manual_access", |
| "url": "https://github.com/cv516Buaa/MMOTU_DS2Net", |
| "filename": null, |
| "revision": null, |
| "md5": null, |
| "approximate_bytes": null, |
| "destination": "quarantine/mmotu", |
| "access": "third_party_drive", |
| "download_by_default": false, |
| "training_use": "blocked_pending_license_review", |
| "license": null, |
| "commercial_use": "unverified", |
| "modalities": ["US"], |
| "label_levels": ["modality", "region", "organ"], |
| "body_regions": ["pelvis"], |
| "classes": 8, |
| "patients": null, |
| "images": 1639, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://github.com/cv516Buaa/MMOTU_DS2Net", |
| "notes": "OTU_2d 1,469 images + OTU_CEUS 170 images. Repo Apache-2.0 covers the mmsegmentation-derived CODE only; the images have no separate licence. Google Drive distribution only. The MSU abdominal set already supplies an 'ovary' class with a clean licence." |
| }, |
| { |
| "id": "pocus-covid19-ultrasound", |
| "name": "POCUS / covid19_ultrasound lung ultrasound", |
| "category": "dataset", |
| "kind": "manual_access", |
| "url": "https://github.com/jannisborn/covid19_ultrasound", |
| "filename": null, |
| "revision": null, |
| "md5": null, |
| "approximate_bytes": null, |
| "destination": "quarantine/pocus-covid19-ultrasound", |
| "access": "open", |
| "download_by_default": false, |
| "training_use": "per_file_license_audit_required", |
| "license": "mixed per-file (CC-BY-4.0, CC-BY-NC-4.0, all-rights-reserved)", |
| "commercial_use": "partially_permitted_after_per_file_filter", |
| "modalities": ["US"], |
| "label_levels": ["modality", "region"], |
| "body_regions": ["chest"], |
| "classes": 4, |
| "patients": null, |
| "images": null, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://github.com/jannisborn/covid19_ultrasound", |
| "notes": "A licence patchwork, not a dataset. The README explicitly pushes licence verification onto the user. If it is ever used, the build must filter dataset_metadata.csv by the License column and drop every CC-BY-NC and all-rights-reserved row; the Butterfly and scraped subsets are not redistributed at all. The authors now point to COVID-BLUES (github.com/NinaWie/COVID-BLUES, 371 videos / 63 patients) - licence unverified." |
| }, |
| { |
| "id": "biomedclip", |
| "name": "BiomedCLIP PubMedBERT + ViT-B/16", |
| "category": "teacher_model", |
| "kind": "huggingface_model", |
| "repository": "microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224", |
| "url": "https://huggingface.co/microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224", |
| "revision": null, |
| "md5": null, |
| "approximate_bytes": null, |
| "destination": "quarantine/biomedclip", |
| "attribution": { |
| "title": "BiomedCLIP-PubMedBERT_256-vit_base_patch16_224", |
| "creators": "Sheng Zhang, Yanbo Xu, Naoto Usuyama, Hanwen Xu, Jaspreet Bagga, Robert Tinn, Sam Preston, Rajesh Rao, Mu Wei, Naveen Valluri, Cliff Wong, Andrea Tupini, Yu Wang, Matt Mazzola, Swadheen Shukla, Lars Liden, Jianfeng Gao, Matthew P. Lungren, Tristan Naumann, Sheng Wang, Hoifung Poon", |
| "year": 2024, |
| "source_url": "https://huggingface.co/microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224", |
| "license": "MIT tag, but the model card restricts use", |
| "license_url": "https://huggingface.co/microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224", |
| "doi": "10.48550/arXiv.2303.00915", |
| "citation": "Zhang, S. et al. BiomedCLIP: a multimodal biomedical foundation model pretrained from fifteen million scientific image-text pairs. arXiv:2303.00915 (2023).", |
| "upstream_attribution": null, |
| "modifications": "Used as a frozen teacher for logit and feature distillation, and for zero-shot bootstrap labelling.", |
| "teacher_warning": "A distilled student is a derivative of its teacher in the most direct sense - it is trained to reproduce the teacher's outputs. The MIT tag and the model card disagree; the card says deployed use is out of scope, and the card is the more specific statement. Do not distil into shippable weights without written clarification.", |
| "citation_verified_at": null |
| }, |
| "access": "open", |
| "download_by_default": false, |
| "training_use": "noncommercial_research_only", |
| "license": "MIT tag, but the model card restricts use", |
| "commercial_use": "prohibited_by_model_card", |
| "modalities": null, |
| "label_levels": null, |
| "body_regions": null, |
| "classes": null, |
| "patients": null, |
| "images": null, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://huggingface.co/microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224", |
| "notes": "Zero-shot pseudo-labeller from plan section 6 Stage A. LICENCE CONFLICT: the HF tag is MIT but the card states 'Any deployed use case of the model - commercial or otherwise - is currently out of scope' and restricts use to research and reproducibility. Treat as research-only. Pin a revision SHA before first use. Loaded via open_clip." |
| }, |
| { |
| "id": "usfm", |
| "name": "USFM - Universal Ultrasound Foundation Model", |
| "category": "teacher_model", |
| "kind": "manual_access", |
| "url": "https://github.com/openmedlab/USFM", |
| "filename": "USFM_latest.pth", |
| "revision": null, |
| "md5": null, |
| "approximate_bytes": null, |
| "destination": "quarantine/usfm", |
| "attribution": { |
| "title": "USFM: A universal ultrasound foundation model", |
| "creators": "Jing Jiao, Jin Zhou, Xiaokang Li, Menghua Xia, Yi Huang, Lihong Huang, Na Wang, Xiaofan Zhang, Shichong Zhou, Yuanyuan Wang, Yi Guo", |
| "year": 2024, |
| "source_url": "https://github.com/openmedlab/USFM", |
| "license": "CC-BY-NC-4.0", |
| "license_url": "https://creativecommons.org/licenses/by-nc/4.0/", |
| "doi": "10.1016/j.media.2024.103202", |
| "citation": "Jiao, J. et al. USFM: A universal ultrasound foundation model generalized to tasks and organs towards label efficient image analysis. Medical Image Analysis 96, 103202 (2024).", |
| "upstream_attribution": null, |
| "modifications": "Candidate frozen teacher for distillation.", |
| "teacher_warning": "CC BY-NC 4.0. A student distilled from it inherits the non-commercial restriction. Ablations only.", |
| "citation_verified_at": null |
| }, |
| "access": "third_party_drive", |
| "download_by_default": false, |
| "training_use": "noncommercial_research_only", |
| "license": "CC-BY-NC-4.0", |
| "commercial_use": "prohibited", |
| "modalities": ["US"], |
| "label_levels": null, |
| "body_regions": null, |
| "classes": null, |
| "patients": null, |
| "images": null, |
| "verified_at": "2026-09-16", |
| "verified_from": "https://github.com/openmedlab/USFM", |
| "notes": "Ultrasound-specific SSL ViT, weights on Google Drive, no registration. CC-BY-NC rules it out of any shipped artifact. Its 2M-image pre-training corpus is not released. Jiao et al., Medical Image Analysis 96:103202, 2024." |
| } |
| ], |
| "rejected_sources": [ |
| { |
| "id": "sif92-ultrasound-organ-dataset", |
| "url": "https://huggingface.co/datasets/Sif92/ultrasound-organ-dataset", |
| "reason": "Contains no data files (5.95 kB, README only); licence tag cc-by-nc-4.0 contradicts the CC BY 4.0 in its own body text; image counts (5,005/2,784/1,293) contradict the authoritative Scholars Junction record (5,468/2,775/1,334). Use msu-abdominal-ultrasound instead.", |
| "verified_at": "2026-09-16" |
| }, |
| { |
| "id": "binaryy-all-ultrasound-images", |
| "url": "https://huggingface.co/datasets/Binaryy/all-ultrasound-images", |
| "reason": "20,378 images / 5.42 GB with no licence, no dataset card and no provenance of any kind. Patient consent chain unknown. Unusable for a clinical product.", |
| "verified_at": "2026-09-16" |
| }, |
| { |
| "id": "infobayai-ultrasound-usg-scans", |
| "url": "https://huggingface.co/datasets/InfoBayAI/Ultrasound-USG-Scans-With-Findings-Dataset", |
| "reason": "Commercial vendor upload claiming 13,070 images / 4,793 patients under cc-by-4.0, gated. Provenance and patient-consent chain unverified. Treat as suspect.", |
| "verified_at": "2026-09-16" |
| }, |
| { |
| "id": "jacksonyu123-miccai25-flare-ultrasound", |
| "url": "https://huggingface.co/datasets/jacksonyu123/MICCAI25_FLARE_Ultrasound", |
| "reason": "Personal mirror of MICCAI 2025 FLARE ultrasound data, empty README, cc-by-nc-4.0, not an official distribution.", |
| "verified_at": "2026-09-16" |
| }, |
| { |
| "id": "novel-biomedai-amos22-nnunet", |
| "url": "https://huggingface.co/datasets/Novel-BioMedAI/AMOS22_nnunet", |
| "reason": "Declares cc-by-nc-nd-4.0, which contradicts the upstream AMOS22 CC BY 4.0, and is gated. Use the Zenodo original.", |
| "verified_at": "2026-09-16" |
| }, |
| { |
| "id": "abdomenus", |
| "url": null, |
| "reason": "No dataset by this name exists. Searched and not found; do not re-add from memory.", |
| "verified_at": "2026-09-16" |
| } |
| ] |
| } |
|
|
