{"success":true,"data":{"author":"Kanwar et al. (TCIA collection); mirrored by MedOtter","bucketAccessType":"public","bucketName":"MedOtter/Prostate-Anatomical-Edge-Cases","bucketPrefix":"","category":"Health & Biomedical","cloudProvider":"huggingface","createdAt":"2026-11-21T20:45:56.564Z","description":"A TCIA collection of pelvic radiotherapy simulation CT for prostate cancer, curated to stress-test automatic contouring: 112 of the 131 patients have anatomy known to break autosegmentation (hip arthroplasty hardware, median lobe hypertrophy, drooping seminal vesicles, Foley catheters, SpaceOAR hydrogel, brachytherapy seeds and calcifications, narrow rectum, bowel in the field, morbid obesity) and 19 are normal controls. Each patient has one CT series and one DICOM RTSTRUCT with manually drawn organs at risk around the prostate, 262 series and 23,359 CT slices in all, acquired on a Philips Brilliance Big Bore scanner. It is CT only, with no prostate MRI zones and no lesion labels. The mirror holds images/ and segmentations/ (DICOM), a converted data/ tier, series_to_patient.json and the TCIA LICENSE.txt; about 17 GB (measured 2026-11-21). Read with pydicom, SimpleITK or rt-utils. Use for robustness testing of pelvic organ-at-risk segmentation and radiotherapy auto-contouring QA. TCIA lists the collection under CC BY 4.0 with DOI 10.7937/QSTF-ST65; cite Kanwar et al., Physics and Imaging in Radiation Oncology 2023.","displayTitle":"Prostate-Anatomical-Edge-Cases: pelvic radiotherapy planning CT of 131 patients with organ-at-risk contours for stress-testing autosegmentation (TCIA)","doi":"10.7937/QSTF-ST65","huggingFaceUrl":"https://huggingface.co/datasets/MedOtter/Prostate-Anatomical-Edge-Cases","id":"9b0f6504-fa9d-5ae3-a239-dd2eb1b88bb1","license":"CC-BY-4.0","licenseEvidenceUrl":"https://www.cancerimagingarchive.net/collection/prostate-anatomical-edge-cases/","objectCountEstimate":23495,"probeStatus":"ok","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"MedOtter/Prostate-Anatomical-Edge-Cases\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"MedOtter/Prostate-Anatomical-Edge-Cases\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=MedOtter, license:cc-by-4.0, size_categories:n<1K, format:parquet, downloads=5294"],"format":"mixed","generatedAt":"2026-11-21T20:48:44.310Z","layout":{"example":"data/preview-00000-of-00001.parquet","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"50415231150415f82815ca0a4c1586021500120000bc144c1000000050726f73746174652d4145432d3030314a140000324a140000334a140000344a140000354a140000364a140000374a140000384a140000394614000431304a14004ac80000314ac80000314ac80000314ac80000314ac80000314ac80000314ac80000314ac80000314ac80000324ac80000324ab40000324ab40000324ab40000324ab40000324ab40000324ab40000324ab40000324ab40000334ab40000334a7c0100334ab40000334ab40000334ab40000334ab40000334ab40000334ab40000334ab40000344ab40000344ab40000344a7c0100344ab40000344ab40000344ab40000344ab40000344ab40000344ab40000354ab40000354ab40000354ab40000354a7c0100354ac80000354ac80000354ac80000354ac80000354ac80000354ac80000364ac80000364ac80000364ac80000364ac80000364ac80000364ac80000364ac80000364ac80000364ac80000364ac80000374ac80000374ac80000374ac80000374ac80000374ac80000374ac80000374ac80000374ac80000374ac80000374ac80000384ac80000384ac80000384ac80000384ac80000384ac80000384ac80000384ac80000384ac80000384ac80000384ac80000394ac80000394ac80000394ac80000394ac80000394ac80000394ac80000394ac80000394ac80000","encoding":"hex","key":"data/preview-00000-of-00001.parquet","truncatedAt":512}],"sampleObjects":["data/preview-00000-of-00001.parquet","images/Prostate-AEC-001/1.3.6.1.4.1.14519.5.2.1.90648498715999918797883179860499513578/00000001.dcm","images/Prostate-AEC-001/1.3.6.1.4.1.14519.5.2.1.90648498715999918797883179860499513578/00000002.dcm","images/Prostate-AEC-001/1.3.6.1.4.1.14519.5.2.1.90648498715999918797883179860499513578/00000003.dcm","images/Prostate-AEC-001/1.3.6.1.4.1.14519.5.2.1.90648498715999918797883179860499513578/00000004.dcm"],"sidecars":[{"bytes":"Prostate-Anatomical-Edge-Cases is distributed by The Cancer Imaging Archive (TCIA)\nunder the Creative Commons Attribution 4.0 International License (CC BY 4.0).\n\nhttps://creativecommons.org/licenses/by/4.0/\n\nData citation:\n  Thompson, R. F., Kanwar, A., Merz, B., Cohen, E., Fisher, H., Rana, S.,\n  Claunch, C., & Hung, A. (2023). Stress-Testing Pelvic Autosegmentation\n  Algorithms Using Anatomical Edge Cases [Data set]. The Cancer Imaging Archive.\n  https://doi.org/10.7937/QSTF-ST65\n\nThis mirror redistributes the collection unmodified. Attribution to the original\nauthors and TCIA is required by the license.\n","encoding":"utf8","key":"LICENSE.txt","name":"LICENSE.txt","size":614,"truncated":false},{"bytes":"---\nlicense: cc-by-4.0\ntask_categories:\n- image-segmentation\ntags:\n- medical\n- ct\n- pelvis\n- prostate\n- radiotherapy\n- organs-at-risk\n- rtstruct\n- dicom\n- tcia\n- autosegmentation\n- edge-cases\npretty_name: Prostate Anatomical Edge Cases (pelvic OAR CT)\nsize_categories:\n- n<1K\ndataset_info:\n  features:\n  - name: patient_id\n    dtype: string\n  - name: ct_series_uid\n    dtype: string\n  - name: rt_series_uid\n    dtype: string\n  - name: num_ct_slices\n    dtype: int32\n  - name: slice_index\n    dtype: int32\n  - name: n_structures\n    dtype: int32\n  - name: missing_structures\n    dtype: string\n  - name: prostate_voxels\n    dtype: int64\n  - name: rectum_voxels\n    dtype: int64\n  - name: bladder_voxels\n    dtype: int64\n  - name: femur_head_l_voxels\n    dtype: int64\n  - name: femur_head_r_voxels\n    dtype: int64\n  - name: image\n    dtype: image\n  - name: mask\n    dtype: image\n  - name: overlay\n    dtype: image\n  splits:\n  - name: preview\n    num_bytes: 28895143\n    num_examples: 131\n  download_size: 28894404\n  dataset_size: 28895143\nconfigs:\n- config_name: default\n  data_files:\n  - split: preview\n    path: data/preview-*\n---\n\n# Prostate-Anatomical-Edge-Cases\n\n**Stress-Testing Pelvic Autosegmentation Algorithms Using Anatomical Edge Cases** —\na TCIA collection of **pelvic radiotherapy planning CT** with manually contoured\norgans at risk, curated so that most cases contain anatomy known to break\nautosegmentation algorithms (Kanwar et al., *Phys Imaging Radiat Oncol* 2023).\n\n> **Read before using — the name is misleading in two ways:**\n> 1. **This is CT, not MRI.** Despite \"Prostate\" in the name it is *not* a prostate\n>    mpMRI/zonal dataset: there is no T2W/DWI/ADC, no peripheral or transition\n>    zone, no urethra, and **no lesion labels**. It is RT simulation CT with\n>    pelvic organ-at-risk contours, and 4 of the 5 classes are not the prostate.\n> 2. **\"Edge cases\" means difficult *anatomy*, not difficult lesions.** The cohort\n>    was selected for hip arthroplasty hardware, prostatic median-lobe hypertrophy,\n>    \"droopy\" seminal vesicles, Foley catheters, SpaceOAR hydrogel, brachytherapy\n>    seeds/calcifications, extracapsular surface irregularity, narrow rectum,\n>    in-field bowel and morbid obesity — features that degrade atlas-, model- and\n>    deep-learning-based contouring.\n\n## Dataset Details\n\n| Field | Value |\n|---|---|\n| Modality | CT (RT simulation) + DICOM RTSTRUCT contours |\n| Body part | Pelvis — prostate and surrounding organs at risk |\n| Task | 3D multi-organ segmentation |\n| Patients | 131 (112 anatomical edge cases + 19 normal controls) |\n| Series | 262 = 131 CT + 131 RTSTRUCT (1:1 paired) |\n| CT slices | 23,359 (min 86 / median 167 / max 379 per case) |\n| Scanner | Philips Brilliance Big Bore (100%); RTSTRUCT exported from Varian ARIA RTM |\n| Format | DICOM (CT) + DICOM RTSTRUCT (contours) |\n| Size | ~17 GB |\n| License | CC BY 4.0 |\n| DOI | `10.7937/QSTF-ST65` |\n| Source | Oregon Health & Science University (single institution) |\n\nThere is **no train/val/test split** — this is a stress-test cohort, not a\nchallenge. Splitting is left to the consumer.\n\n## Classes\n\nFive foreground structures, present in the `StructureSetROISequence` of every\ncase, named per TG-263 and all typed `ORGAN`:\n\n| Label | ROI name |\n|---|---|\n| 0 | background |\n| 1 | `Prostate` |\n| 2 | `Rectum` |\n| 3 | `Bladder` |\n| 4 | `Femur_Head_L` |\n| 5 | `Femur_Head_R` |\n\n## Ground Truth — single manual tier\n\nEvery ROI in every case carries `ROIGenerationAlgorithm = MANUAL`. The contours\nwere drawn by a single radiation oncologist, peer-reviewed, and used clinically\nfor treatment planning — so there is exactly one annotation tier and no rater\nambiguity.\n\nThe three autosegmentation outputs benchmarked in the paper (atlas-based/MIM,\nmodel-based/RayStation, deep-learning U-Net/RayStation v9B) were **not** deposited;\nall 262 series report `ThirdPartyAnalysis = NO`. The RTSTRUCTs here are the gold\nstandard, not algorithm output.\n\n## Important Notes for Loaders\n\nThese were verified by parsing all 131 RTSTRUCT objects and are easy to get wron","encoding":"utf8","key":"README.md","name":"README.md","size":9315,"truncated":true}],"sourceObjectKey":"data/preview-00000-of-00001.parquet","starterCell":"from datasets import load_dataset\nds = load_dataset(\"MedOtter/Prostate-Anatomical-Edge-Cases\")\nprint(ds)","starterSamplePath":"data/preview-00000-of-00001.parquet","totalObjects":23495},"summary":"A TCIA collection of pelvic radiotherapy simulation CT for prostate cancer, curated to stress-test automatic contouring: 112 of the 131 patients have anatomy known to break…","title":"Prostate-Anatomical-Edge-Cases: pelvic radiotherapy planning CT of 131 patients with organ-at-risk contours for stress-testing autosegmentation (TCIA)","updatedAt":"2026-11-22T00:51:51.841Z","visibility":"public","websiteUrl":"https://www.cancerimagingarchive.net/collection/prostate-anatomical-edge-cases/"}}