---
license: cc-by-nc-nd-4.0
task_categories:
- image-segmentation
- object-detection
- zero-shot-image-classification
language:
- en
pretty_name: Histomorphological Cell Atlas
size_categories:
- 1M-10M
viewable: false
viewer: false
---

# Histomorphological Atlas of Single Cells across the Human Life Span

A billion-scale single-cell atlas of human tissue from whole-slide histopathology images across 16 organs. This dataset contains cell type densities (cells per mm²), spatial coordinates, morphological phenotypes, and tissue graphs derived from deep learning analysis of H&E-stained tissue sections from 980 donors (ages 20-70) in the GTEx cohort.

## Dataset Description

We applied deep learning to 14,788 whole-slide H&E images from the GTEx project, detecting and classifying over 3.5 billion single cells. This dataset provides the processed outputs of that pipeline, including per-slide cell type densities, per-cell spatial coordinates, cell type identities, and slide-level tissue graphs.

### Organs Covered

16 organs (20 tissue subtypes): Adrenal Gland, Colon (Sigmoid, Transverse), Esophagus (Gastroesophageal Junction, Mucosa, Muscularis), Liver, Lung, Ovary, Pancreas, Prostate, Skin (Sun-Exposed, Not Sun-Exposed), Small Intestine (Terminal Ileum), Spleen, Stomach, Testis, Thyroid, Uterus, Vagina.

### Contents

| File | Description |
|------|-------------|
| `16_tissues_cell_abundances.parquet` | Cell type densities (cells per mm²) across all 16 tissues, collapsed by cell phenotype |
| *(forthcoming)* Cell polygons | Per-cell spatial polygon coordinates for each slide |
| *(forthcoming)* Cell identities | Per-cell type classification for each slide |
| *(forthcoming)* Slide graphs | Tissue-level spatial graphs for each slide |

### Data Format

**`16_tissues_cell_abundances.parquet`**

Columns:
- `slide_id`: GTEx slide identifier
- `organ`: Organ/tissue name
- `mpp`: Microns per pixel (image resolution)
- `pixels`: Total pixel area analyzed
- `cells_pmm2`: Total cell density (cells per mm²)
- `endothelial_error_pmm2`: Endothelial segmentation error density (cells per mm²)
- Remaining columns: Cell phenotype densities (cells per mm²) — including Hepatocyte/Chromaffin, Connective, Immune, Cuboidal, Muscle, Glandular, Elongated, Glandular (Columnar), Epithelial, Dead, Endothelial, Squamous, Alveolar, Seminiferous, Follicular

### Source Data

Whole-slide images and metadata from the GTEx project:
- [GTEx Histology Portal](https://gtexportal.org/home/histologyPage)
- [GTEx Sampling Site Page](https://gtexportal.org/home/samplingSitePage)

GTEx whole-slide images are publicly available from the GTEx portal. Associated donor metadata and transcriptomic data are available through dbGaP under controlled access.

### Pipeline Summary

Cells were segmented and classified from whole-slide H&E images using CellViT (SAM-H backbone). Cell phenotyping integrates CellViT morphological features and DINOv2 (ViT-S/14) nuclear features, with PLIP providing text-based guidance, to classify each cell into a phenotype class.

## Citation

If you use this dataset, please cite:

```
Abila, E., Zheng, Y., Bago-Horvath, Z. & Rendeiro, A.F. A single-cell view of human
tissue aging reveals architectural decline beyond cellular composition. (2026).
```

## License

This dataset is licensed under CC-BY-NC-ND-4.0.

## Contact

Andre F. Rendeiro — arendeiro@cemm.oeaw.ac.at