{"success":true,"data":{"author":"Maximilian Haist, John Hickey et al.; curated by SPORA","bucketAccessType":"public","bucketName":"csmp-hub/haist-hnscc","bucketPrefix":"","category":"Health & Biomedical","cloudProvider":"huggingface","createdAt":"2026-11-17T19:29:33.139Z","description":"H&E images of head and neck squamous cell carcinoma (HNSCC) primary tumors and lymph nodes from the study of Haist, Hickey et al., in which 59-channel CODEX imaging showed that lymph node colonization remodels tissue through immunosuppressive fibroblast-myeloid niches. The selected cohort has 389 tissue regions from 79 patients, cut into 13,615 overlapping 512x512 RGB crops at 0.377 micrometers per pixel with unchanged H&E values. Cell regions are VirTues model predictions supplied by SPORA rather than the original segmentation; original CODEX cells are linked to them by centroid, and a region receives one of 21 source cell types (tumor, immune, vascular and stromal populations) when its linked labels agree, otherwise unknown (255). Link tables keep all source IDs. The CODEX image stacks are not included. Files: crop index and instance-link Parquet tables, classes.json, class distributions and license texts; about 8.5 GB (measured 2026-11-17). Images and masks are stored as PNG or TIFF bytes inside Parquet; read them with polars or pandas plus Pillow, or with Hugging Face datasets. Curated and packaged on Hugging Face by the SPORA project (csmp-hub), which keeps the original license and citation files. Images and SPORA predictions are CC0 1.0; the original cell labels on Mendeley Data (t4k5kj3cxr) are CC BY 4.0, so attribution is required. Cite Haist et al. and the Mendeley record doi 10.17632/t4k5kj3cxr.1.","displayTitle":"Haist HNSCC: head and neck cancer and lymph node H&E crops with CODEX cell-type labels from 79 patients","doi":"10.17632/t4k5kj3cxr.1","huggingFaceUrl":"https://huggingface.co/datasets/csmp-hub/haist-hnscc","id":"bfd96d2f-8b58-5f77-b372-2489d8b03486","license":"CC-BY-4.0","licenseEvidenceUrl":"https://data.mendeley.com/datasets/t4k5kj3cxr/1","objectCountEstimate":20,"probeStatus":"ok","referenceUrl":"https://pmc.ncbi.nlm.nih.gov/articles/PMC13289805/","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"csmp-hub/haist-hnscc\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"csmp-hub/haist-hnscc\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=csmp-hub, license:cc-by-4.0, size_categories:10K<n<100K, downloads=135"],"format":"mixed","generatedAt":"2026-11-17T19:40:18.744Z","layout":{"example":"class-distribution.csv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"code,name,source_cells,predicted_instances\r\n1,B,228190,142810\r\n2,CAF,64599,47461\r\n3,CD4T,300728,196795\r\n4,CD8T,146245,97452\r\n5,DC,55516,31318\r\n6,FDC,8700,4875\r\n7,LEC,17454,10503\r\n8,NK,11908,7983\r\n9,TFH,4152,2761\r\n10,Treg,101258,69817\r\n11,VSMC_myofibroblast,13856,8732\r\n12,gdT,7230,4908\r\n13,granulocyte,37746,23508\r\n14,macrophages,104542,70300\r\n15,mast,7649,5270\r\n16,nerve,868,701\r\n17,plasma,65347,46370\r\n18,stromal,128956,80638\r\n19,tumor,159464,137534\r\n20,tumor_CA9,38766,32797\r\n21,vasculature,46332,28208\r\n255,U","encoding":"utf8","key":"class-distribution.csv","truncatedAt":512}],"sampleObjects":["LICENSE-CC-BY-4.0.txt","LICENSE-CC0.txt","class-distribution.csv","class-distribution.png","classes.json"],"sidecars":[{"bytes":"Haist HNSCC H&E crops and source-linked cell labels\n\nSource images and SPORA VirTues predictions: CC0 1.0 Universal.\nSee LICENSE-CC0.txt and https://www.ebi.ac.uk/biostudies/bioimages/studies/S-BIAD1638\n\nOriginal cell measurements and labels: CC BY 4.0.\nMaximilian Haist and John Hickey (2026), Lymph node colonization induces tissue remodeling via immunosuppressive fibroblast-myeloid cell niches supporting metastatic tolerance, version 1.\nhttps://doi.org/10.17632/t4k5kj3cxr.1\nSee LICENSE-CC-BY-4.0.txt and original-release.json.\n\nThe labelled collection is distributed under CC BY 4.0; the CC0 status of its source images and predictions is unchanged.\nChanges: selected the 389 SPORA HNSCC regions; extracted overlapping H&E crops; sampled supplied 1um predictions on the native image grid; linked original cell centres to predictions; added agreement/majority class masks, quality flags, pixel availability and counts.\nSource cells and predicted regions have independent IDs. No new segmentation or registration was performed.\nSPORA: https://spora.bioimagearchive.org/projects/spora\n","encoding":"utf8","key":"LICENSE","name":"LICENSE","size":1088,"truncated":false},{"bytes":"---\nlicense: cc-by-4.0\ntask_categories:\n- image-segmentation\nconfigs:\n- config_name: default\n  data_files:\n  - split: train\n    path: train.parquet\n  - split: test\n    path: test.parquet\n---\n\n# Haist HNSCC\n\nH&E images of head-and-neck squamous cell carcinoma and lymph nodes, paired with\ncell-type labels from the study's CODEX measurements. The selected cohort has\n389 tissue regions from 79 patients and 21 source cell types, including tumour,\nimmune, vascular and stromal populations.\n\nThe images and SPORA predictions are released under CC0 1.0. The original cell\nlabels are [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/); retain attribution when using or sharing this labelled\ncollection. Please credit Haist and colleagues, the original dataset and SPORA.\nOriginal contributors: Maximilian Haist, John Hickey.\n\n## Images and annotations\n\nThere are 13,615 overlapping 512 × 512 RGB crops at 0.37744 µm per pixel.\nH&E values are unchanged. The original 59-channel CODEX images support the source\nlabels; these H&E crop files do not include the CODEX image stacks.\n\nThe cell regions are **VirTues predictions supplied by SPORA**, rather than the\noriginal study's segmentation masks. Source cells are linked by their centres\nto these predictions. A region gets a cell type when its linked source labels\nagree. Missing matches and conflicting labels remain unknown (255). More than\none source cell can link to a predicted region; the link tables retain those\nIDs so that merged regions can be identified. The majority alternative is kept\nin `type_proposed` for comparison.\n\nThe predictions were supplied at 1 µm per pixel and sampled onto the H&E grid.\nTheir outlines therefore have coarser detail than the image. Original prediction\nIDs and source cell UIDs are kept separately.\n\n![Examples: H&E, boundaries, instances and cell types](examples.png)\n\n## Splits and quality\n\nThe original patient split is preserved: 10,605 training crops\nfrom 59 patients and 3,010 test crops from 20 patients.\nThere is no validation split. Crops from one patient stay together; overlapping\ncrops must not be randomly divided into new train and test sets.\n\nTwelve crops were visually reviewed: eleven usable and one excluded. Rejected\nareas also flag overlapping crops through `skip_training`. Other crops remain\n`unreviewed`; a nearby approval does not approve them. A usable review does not\nverify every individual cell or label.\n\n`he_valid` marks available H&E pixels; border-connected zero padding is marked\n0. `mask_valid` marks the available prediction area, including small differences\nat image edges. Keep both masks when computing losses: use only pixels where\nboth are 1, and ignore 255 for cell-type loss. The image and source cell masks\nare retained even where supervision should be ignored.\n\n## Files\n\n`train.parquet` and `test.parquet` contain `id`, `image`, `inst`, `type`,\n`type_proposed`, `he_valid`, `mask_valid`, patient/region IDs, crop coordinates\nand quality flags. Images and uint8 masks are PNG; instance masks are uint32\nTIFF. Instance 0 and type 0 are background. `sem` is empty.\n\n`instance-counts.csv` lists each crop's instance count and counts for every class.\nOverlapping crops repeat cells. The totals below count each predicted source\ninstance once per region. `class-distribution.csv` also records the source-table\ncell counts. These counts differ because source cells and predicted regions\ndo not have a one-to-one correspondence.\n\n| Code | Source class | Source cells | Typed predicted instances |\n| --- | --- | ---: | ---: |\n| 1 | B | 228,190 | 142,810 |\n| 2 | CAF | 64,599 | 47,461 |\n| 3 | CD4T | 300,728 | 196,795 |\n| 4 | CD8T | 146,245 | 97,452 |\n| 5 | DC | 55,516 | 31,318 |\n| 6 | FDC | 8,700 | 4,875 |\n| 7 | LEC | 17,454 | 10,503 |\n| 8 | NK | 11,908 | 7,983 |\n| 9 | TFH | 4,152 | 2,761 |\n| 10 | Treg | 101,258 | 69,817 |\n| 11 | VSMC_myofibroblast | 13,856 | 8,732 |\n| 12 | gdT | 7,230 | 4,908 |\n| 13 | granulocyte | 37,746 | 23,508 |\n| 14 | macrophages | 104,542 | 70,300 |\n| 15 | mast | 7,649 | 5,270 |\n| 16 | nerve | 868 | 701 |\n| 17 | plasma | 65,347 | 46","encoding":"utf8","key":"README.md","name":"README.md","size":6340,"truncated":true},{"bytes":"{\n  \"dataset\": \"Haist HNSCC\",\n  \"license\": \"CC-BY-4.0; source images/predictions CC0-1.0\",\n  \"sources\": {\n    \"data_root\": \"https://livingobjects.ebi.ac.uk/bioimaging-01-pub/spora-submissions/haist2026lymph/\",\n    \"labels\": \"https://doi.org/10.17632/t4k5kj3cxr.1\",\n    \"images\": \"https://www.ebi.ac.uk/biostudies/bioimages/studies/S-BIAD1638\",\n    \"spora\": \"https://spora.bioimagearchive.org/projects/spora\"\n  },\n  \"splits\": {\n    \"train\": {\n      \"crops\": 10605,\n      \"patients\": 59\n    },\n    \"test\": {\n      \"crops\": 3010,\n      \"patients\": 20\n    }\n  },\n  \"source_cells\": 1549506,\n  \"predicted_instances\": 1255888,\n  \"quality\": {\n    \"unreviewed\": 13603,\n    \"skip_training\": 6,\n    \"human_approved\": 11,\n    \"usable\": 11,\n    \"exclude\": 1\n  },\n  \"source_annotation_sha256\": \"8e9c07bcfd80ca0e7ce053c269f5055adaef4a967a28def2957c8af8b0e00b5d\",\n  \"sha256\": {\n    \"instance-links.parquet\": \"a8eaaeb9e034b9088f36fe083d45ae00ac4b9b9ed2dd827e0eaa5ad29015f6d4\",\n    \"LICENSE-CC0.txt\": \"6d489af6292662d9e36d34ce49423784984a5f6e41d7b58f49b01264df59fa03\",\n    \"instance-counts.csv\": \"2b4745e95a393d898d439b8e0ba7e5cd35a48b621c91f7992ea13e8481440f1a\",\n    \"pairing-summary.json\": \"a8e13918ab1c7f507409d3d76596d1eb3707ca498fe6ffbfb09d68e94db52270\",\n    \"train.parquet\": \"6110460457af551937abc8a38bd365746cb6426f532050dbd49601df7e6d4902\",\n    \"test.parquet\": \"e67cf8f5cd19bbf118fc629d820fec70986a7c7797c702e41c70bbf4214172a4\",\n    \"class-distribution.png\": \"19091ba141401123eb3722aee47670ea8c6d841b066bb9a34cd1b342c1849b0a\",\n    \"LICENSE-CC-BY-4.0.txt\": \"9ba9550ad48438d0836ddab3da480b3b69ffa0aac7b7878b5a0039e7ab429411\",\n    \"LICENSE\": \"131a72ea517490125f5f4cab5f9425dbe457cae4d0e90d2154a9dec6664839a9\",\n    \"classes.json\": \"874204c21b868e898cbb3c1de594e8d286db123e066a1eabbad07bf9baaf8bb0\",\n    \"source-cells.parquet\": \"0a56dc1bb5a8663c610b214edd1ddf664058c413ea48de43afd7cbe61910e81e\",\n    \"crop-index.parquet\": \"7c230e40a53eb41210ad6936bdffe486432078042eae949afd5d095f10b5f72a\",\n    \"examples.png\": \"9039c3e0569a19c44e905f4adb84ff38491d9188591b081ab2385c3d2dcb4198\",\n    \"README.md\": \"2167016c3f27fc46602c8a8a251e7c84bc9f46ee5c815f87bd4ede3509bd4fab\",\n    \"source-regions.parquet\": \"aec3d65544d0b0407c8f78231b12873129e7e1ba84cfe2cf3d4ba88c14bf6d3e\",\n    \"original-release.json\": \"83d9fd5fa66fb7c4b9027456412e56e678d746df5cbeb0ff39bf561a2c652ff6\",\n    \"validation-summary.json\": \"e6bf5a5c744e5c638f4e2606141856cfd97eb86efdd57de3eaf936c41d7eefa9\",\n    \"class-distribution.csv\": \"f639b46f3a4102439b50dec181ddaf798922ae8034163e8fd6598ae525c2cda0\"\n  }\n}\n","encoding":"utf8","key":"manifest.json","name":"manifest.json","size":2545,"truncated":false}],"sourceObjectKey":"class-distribution.csv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/csmp-hub/haist-hnscc/class-distribution.csv\")\ndf.head()","starterSamplePath":"class-distribution.csv","totalObjects":20},"summary":"H&E images of head and neck squamous cell carcinoma (HNSCC) primary tumors and lymph nodes from the study of Haist, Hickey et al., in which 59-channel CODEX imaging showed that…","title":"Haist HNSCC: head and neck cancer and lymph node H&E crops with CODEX cell-type labels from 79 patients","updatedAt":"2026-11-17T23:05:35.577Z","visibility":"public","websiteUrl":"https://www.ebi.ac.uk/biostudies/bioimages/studies/S-BIAD1638"}}