{"success":true,"data":{"author":"M. Kronke, C. Eilers, N. Navab, T. Wendler et al. (TUM); mirrored by MedOtter","bucketAccessType":"public","bucketName":"MedOtter/SegThy","bucketPrefix":"","category":"Health & Biomedical","cloudProvider":"huggingface","createdAt":"2026-11-21T20:45:56.541Z","description":"SegThy pairs electromagnetically tracked freehand 3D ultrasound with T1-VIBE MRI of the neck in 28 healthy volunteers from TUM and Klinikum rechts der Isar, Munich, with manual segmentation of the thyroid gland, both common carotid arteries and both internal jugular veins. The ultrasound sweeps are reconstructed to roughly 0.12 mm isotropic volumes, which is rare among public ultrasound sets that usually hold 2D B-mode frames; MRI volumes are 320x320x80 at 0.625x0.625x1.0 mm. Fourteen subjects carry multi-class labels in both modalities, which makes the set usable as an MR to ultrasound registration benchmark. Folders us/, mri_thyroid/ and mri_multiclass/ hold NIfTI volumes and labels with JSONL indexes and previews, originals/ keeps the upstream files and the upstream readmes and LICENSE.txt are included; the CNN-predicted network_results tier is not mirrored. About 1.4 GB (measured 2026-11-21); read with nibabel or SimpleITK. Use for thyroid segmentation, neck vessel segmentation and multimodal registration. The TUM CAMP group releases SegThy under CC BY; cite Kronke et al., PLOS ONE 2022 (doi 10.1371/journal.pone.0268550).","displayTitle":"SegThy: tracked 3D freehand ultrasound and MRI of the thyroid gland, carotid arteries and jugular veins in 28 volunteers","huggingFaceUrl":"https://huggingface.co/datasets/MedOtter/SegThy","id":"4d80a3f6-422d-5dd7-b900-be33ff7cd927","license":"CC-BY-4.0","licenseEvidenceUrl":"https://www.cs.cit.tum.de/camp/publications/segthy-dataset/","objectCountEstimate":162,"probeStatus":"ok","referenceUrl":"https://doi.org/10.1371/journal.pone.0268550","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"MedOtter/SegThy\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"MedOtter/SegThy\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=MedOtter, license:cc-by-4.0, size_categories:n<1K, format:parquet, downloads=429"],"format":"mixed","generatedAt":"2026-11-21T20:48:43.922Z","layout":{"example":"metadata/manifest.csv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"config,case_id,subject_id,role,repo_path,bytes,file_sha256,payload_sha256,original_payload_sha256,shape_xyz,dtype,source_container,repaired\r\nus,001_P1_1_left,001,image,us/train/images/001_P1_1_left.nii.gz,19453790,d0d47c66c7556c4b4db612686c590728c3d506cec436f6b556f1b2b465b65bed,a121ae56daad64b91183190faa08dc53fa71a27608e41919034076fc4538b293,,486x453x370,uint8,.nii,False\r\nus,001_P1_1_left,001,mask,us/train/masks/001_P1_1_left.nii.gz,266859,b0dc9b664fd413b0cb776baa5b3f3186686f8d465e45112339c6a1fb2d8a33ba,a77","encoding":"utf8","key":"metadata/manifest.csv","truncatedAt":512}],"sampleObjects":["Readme_MRI.txt","Readme_US.txt","metadata/manifest.csv","metadata/subject_crossref.csv","mri_multiclass/train/images/001.nii.gz"],"sidecars":[{"bytes":"SegThy is distributed by the Chair for Computer Aided Medical Procedures (CAMP),\nTechnical University of Munich, together with Klinikum rechts der Isar, from the\nauthor-hosted deposit:\n\n  https://www.cs.cit.tum.de/camp/publications/segthy-dataset/\n  https://www.campar.in.tum.de/public_datasets/2022_plosone_eilers/\n\nLICENSE: CC BY (Creative Commons Attribution)\n\nThe license statement ships INSIDE both official archives, as Readme.txt, and is\nreproduced verbatim from the MRI archive:\n\n  \"This dataset is licensed under a CC BY license. This means that users of it can\n  be distributed, shuffled, adapted, and built upon any material with the sole\n  condition that the creators are acknowledged by citing the following article:\n  Tracked 3D Ultrasound and deep neural network-based Thyroid Segmentation reduce\n  Interobserver Variability in Thyroid Volumetry; Markus Kroenke, Christine Eilers,\n  Desislava Dimova, Melanie Koehler, Gabriel Buschner, Lilit Mirzojan, Lemonia\n  Konstantinidou, Marcus R. Makowski, James Nagarajah, Nassir Navab, Wolfgang Weber,\n  Thomas Wendler; PLOS ONE 2022 (in press)\n  The license permits commercial use as long as the authors are cited as described\n  above.\"\n\nThe identical statement appears in the US archive's Readme.txt. Both Readme files\nare preserved unmodified at the root of this mirror as Readme_MRI.txt and\nReadme_US.txt.\n\nConsequences:\n  * Attribution  -- the citation above is mandatory.\n  * Commercial use is EXPLICITLY PERMITTED by the authors (\"The license permits\n    commercial use as long as the authors are cited\").\n  * Redistribution and adaptation are explicitly permitted, which is what makes\n    this mirror possible.\n\nIMPORTANT -- no CC BY version number is stated anywhere upstream. Neither the\nwebsite nor either in-archive Readme names 1.0/2.0/3.0/4.0; they say only \"a CC BY\nlicense\". Third-party listings that assert \"CC BY 4.0\" are inferring. This mirror\nis tagged cc-by-4.0 because the Hub's license vocabulary has no unversioned CC BY\noption, and 4.0 is the current default -- but the authoritative statement is the\nunversioned one quoted above.\n\nThis mirror redistributes the gold-standard portion of the official release with\nits NIfTI payloads byte-identical to source (payload_sha256 in metadata/manifest.csv\nrecords the digest of the uncompressed NIfTI bytes for every file, so fidelity is\ncheckable independently of gzip container framing). The upstream `network_results/`\nfolder -- QuickNAT model predictions, not ground truth -- is deliberately excluded;\nsee the README. Added files are metadata only (jsonl indices, manifest.csv,\nsubject_crossref.csv, this notice and the dataset card).\n","encoding":"utf8","key":"LICENSE.txt","name":"LICENSE.txt","size":2664,"truncated":false},{"bytes":"---\nlicense: cc-by-4.0\ntask_categories:\n- image-segmentation\ntags:\n- medical\n- ultrasound\n- 3d-ultrasound\n- mri\n- thyroid\n- neck\n- carotid-artery\n- jugular-vein\n- vessel-segmentation\n- segthy\n- multimodal\npretty_name: SegThy - Thyroid and Neck Vessel Segmentation (3D Ultrasound + MRI)\nsize_categories:\n- n<1K\nconfigs:\n- config_name: preview_mri_multiclass\n  data_files:\n  - split: train\n    path: preview_mri_multiclass/train-*\n- config_name: preview_mri_thyroid\n  data_files:\n  - split: train\n    path: preview_mri_thyroid/train-*\n- config_name: preview_us\n  data_files:\n  - split: train\n    path: preview_us/train-*\ndataset_info:\n- config_name: preview_mri_multiclass\n  features:\n  - name: case_id\n    dtype: string\n  - name: subject_id\n    dtype: string\n  - name: config\n    dtype: string\n  - name: modality\n    dtype: string\n  - name: side\n    dtype: string\n  - name: shape\n    dtype: string\n  - name: thyroid_volume_ml\n    dtype: float32\n  - name: annotation_gap\n    dtype: string\n  - name: mask_repaired\n    dtype: bool\n  - name: image\n    dtype: image\n  - name: mask\n    dtype: image\n  - name: overlay\n    dtype: image\n  - name: slice_index\n    dtype: int32\n  - name: n_slices\n    dtype: int32\n  - name: classes_present\n    dtype: string\n  splits:\n  - name: train\n    num_bytes: 1979055\n    num_examples: 14\n  download_size: 1989056\n  dataset_size: 1979055\n- config_name: preview_mri_thyroid\n  features:\n  - name: case_id\n    dtype: string\n  - name: subject_id\n    dtype: string\n  - name: config\n    dtype: string\n  - name: modality\n    dtype: string\n  - name: side\n    dtype: string\n  - name: shape\n    dtype: string\n  - name: thyroid_volume_ml\n    dtype: float32\n  - name: annotation_gap\n    dtype: string\n  - name: mask_repaired\n    dtype: bool\n  - name: image\n    dtype: image\n  - name: mask\n    dtype: image\n  - name: overlay\n    dtype: image\n  - name: slice_index\n    dtype: int32\n  - name: n_slices\n    dtype: int32\n  - name: classes_present\n    dtype: string\n  splits:\n  - name: train\n    num_bytes: 3940848\n    num_examples: 28\n  download_size: 3950988\n  dataset_size: 3940848\n- config_name: preview_us\n  features:\n  - name: case_id\n    dtype: string\n  - name: subject_id\n    dtype: string\n  - name: config\n    dtype: string\n  - name: modality\n    dtype: string\n  - name: side\n    dtype: string\n  - name: shape\n    dtype: string\n  - name: thyroid_volume_ml\n    dtype: float32\n  - name: annotation_gap\n    dtype: string\n  - name: mask_repaired\n    dtype: bool\n  - name: image\n    dtype: image\n  - name: mask\n    dtype: image\n  - name: overlay\n    dtype: image\n  - name: slice_index\n    dtype: int32\n  - name: n_slices\n    dtype: int32\n  - name: classes_present\n    dtype: string\n  splits:\n  - name: train\n    num_bytes: 5853237\n    num_examples: 32\n  download_size: 5869481\n  dataset_size: 5853237\n---\n\n# SegThy — Thyroid and Neck Vessel Segmentation (3-D Ultrasound + MRI)\n\n**SegThy** pairs **electromagnetically tracked freehand 3-D ultrasound** with\n**T1-VIBE MRI** of the same healthy volunteers, and annotates the **thyroid gland,\nboth common carotid arteries and both internal jugular veins** by hand in each.\n\nTwo things make it unusual. It is one of very few public datasets with **3-D\nreconstructed freehand ultrasound** rather than 2-D B-mode frames — the sweeps are\ntracked and resampled to a 0.12 mm isotropic volume. And **14 subjects carry\nmulti-class labels in both modalities**, which is what makes it a genuine MR/US\nregistration benchmark rather than two unrelated collections.\n\n## What this mirror contains — read first\n\n> **⚠️ The 504-volume `network_results/` tier is deliberately NOT mirrored.**\n> It is 94 % of the upstream archive (17.3 GiB) and it is **not ground truth**.\n> Upstream's own Readme: *\"contains the US scans and corresponding thyroid\n> segmentations **from the trained QuickNAT**\"* — these are CNN predictions, and the\n> object of study in the paper, not a reference standard. Three independent reasons\n> to exclude them:\n> 1. They are model output, so evaluating against them measures agreement with\n>    QuickNAT, not with a hu","encoding":"utf8","key":"README.md","name":"README.md","size":18128,"truncated":true}],"sourceObjectKey":"metadata/manifest.csv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/MedOtter/SegThy/metadata/manifest.csv\")\ndf.head()","starterSamplePath":"metadata/manifest.csv","totalObjects":162},"summary":"SegThy pairs electromagnetically tracked freehand 3D ultrasound with T1-VIBE MRI of the neck in 28 healthy volunteers from TUM and Klinikum rechts der Isar, Munich, with manual…","title":"SegThy: tracked 3D freehand ultrasound and MRI of the thyroid gland, carotid arteries and jugular veins in 28 volunteers","updatedAt":"2026-11-22T01:04:00.032Z","visibility":"public","websiteUrl":"https://www.cs.cit.tum.de/camp/publications/segthy-dataset/"}}