# NuCLS: Nucleus Classification, Localization and Segmentation in TCGA breast cancer H&E (corrected single-rater set, 59k nuclei)

## Facts

- Provider: Mohamed Amgad, Lee Cooper et al. (Northwestern University); mirrored by MedOtter
- License: CC0-1.0 (https://sites.google.com/view/nucls)
- Size: 5 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/MedOtter/NuCLS
- Access: public, readable without an account
- Category: Health & Biomedical
- Source: https://doi.org/10.1093/gigascience/giac037
- Page: https://scigantic.com/archive/827d0917-1786-5f6c-9714-89f51c81dd99
- Keywords: Health & Biomedical, histopathology, image segmentation, Parquet, Hugging Face

## About

NuCLS (Nucleus Classification, Localization and Segmentation) is a crowdsourced and pathologist-reviewed annotation set of cell nuclei in breast cancer histology. This repository holds the corrected single-rater subset: 1,744 H&E fields of view cropped at 40x (about 0.2 microns per pixel) from 124 TCGA-BRCA whole-slide images, with 59,485 annotated nuclei. Nuclei are labeled in a 13-class scheme that collapses to 7 or 4 super-classes (tumor, fibroblast, lymphocyte, plasma cell, macrophage, mitotic figure, vascular endothelium, myoepithelium, apoptotic body, neutrophil, ductal epithelium, eosinophil). About one third of nuclei have true polygon outlines and two thirds are bounding boxes rasterized as filled rectangles; the per-FOV annotation CSV keeps the type so the two can be told apart. Masks are 3-channel PNGs encoding class, region of interest and instance id. Parquet under data/, about 400 MB (measured 2026-11-21); read with Hugging Face datasets or pandas plus Pillow. The uncorrected and multi-rater tiers remain at the official site. Use for nucleus detection, classification and tumor-infiltrating lymphocyte scoring. The authors release NuCLS under CC0 1.0; cite Amgad et al., GigaScience 2022 (doi 10.1093/gigascience/giac037).

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("MedOtter/NuCLS")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `data/train-00000-of-00003.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/827d0917-1786-5f6c-9714-89f51c81dd99
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/827d0917-1786-5f6c-9714-89f51c81dd99
- Over MCP: `get_schema_card` with id `827d0917-1786-5f6c-9714-89f51c81dd99` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Image segmentation datasets](https://scigantic.com/topics/image-segmentation)
- [Health & Biomedical datasets](https://scigantic.com/topics/health-and-biomedical)
- [Histopathology datasets](https://scigantic.com/topics/histopathology)
