# Haist HNSCC: head and neck cancer and lymph node H&E crops with CODEX cell-type labels from 79 patients

## Facts

- Provider: Maximilian Haist, John Hickey et al.; curated by SPORA
- License: CC-BY-4.0 (https://data.mendeley.com/datasets/t4k5kj3cxr/1)
- Size: 20 files
- Format: CSV, PNG, JSON (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/csmp-hub/haist-hnscc
- Access: public, readable without an account
- DOI: https://doi.org/10.17632/t4k5kj3cxr.1
- Category: Health & Biomedical
- Source: https://pmc.ncbi.nlm.nih.gov/articles/PMC13289805/
- Page: https://scigantic.com/archive/bfd96d2f-8b58-5f77-b372-2489d8b03486
- Keywords: Health & Biomedical, histopathology, image segmentation, CSV, PNG, JSON, Hugging Face

## About

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.

## Load the data

```python
import pandas as pd
df = pd.read_csv("hf://datasets/csmp-hub/haist-hnscc/class-distribution.csv")
df.head()
```

## Data structure

File types in sampled files: CSV, PNG, JSON

Example file: `class-distribution.csv`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/bfd96d2f-8b58-5f77-b372-2489d8b03486
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/bfd96d2f-8b58-5f77-b372-2489d8b03486
- Over MCP: `get_schema_card` with id `bfd96d2f-8b58-5f77-b372-2489d8b03486` at https://api.scigantic.com/mcp

## Related topics

- [JSON datasets](https://scigantic.com/topics/json)
- [CSV datasets](https://scigantic.com/topics/csv)
- [Image segmentation datasets](https://scigantic.com/topics/image-segmentation)
- [Health & Biomedical datasets](https://scigantic.com/topics/health-and-biomedical)
- [PNG datasets](https://scigantic.com/topics/png)
- [Histopathology datasets](https://scigantic.com/topics/histopathology)
