BCSS: Breast Cancer Semantic Segmentation of TCGA H&E regions of interest

Provider
Mohamed Amgad et al., Emory University / PathologyDataScience (re-hosted by MedOtter)
License
CC0-1.0 (source)
Size
12 files
Format
Parquet (in sampled files)
Stored at
Hugging Face (https://huggingface.co/datasets/MedOtter/BCSS)
Access
Free and public. No account with the data provider is needed.
Category
Health & Biomedical
Source
https://doi.org/10.1093/bioinformatics/btz083

About BCSS: Breast Cancer Semantic Segmentation of TCGA H&E regions of interest

BCSS (Breast Cancer Semantic Segmentation) is a histopathology dataset from Amgad et al., Bioinformatics 2019, in which pathologists, residents and medical students annotated regions of interest from TCGA breast cancer whole-slide images through a structured crowdsourcing protocol. The masks cover 22 tissue classes, among them tumor, stroma, lymphocytic infiltrate, necrosis or debris, blood vessels, fat, mucoid material, normal acinus or duct and other immune infiltrate, with an outside-ROI don't-care class. This repository, re-hosted by MedOtter, holds the 151 ROIs as colour-normalised RGB PNG images at 0.25 microns per pixel with indexed masks and columns for image ID, TCGA patient ID and ROI coordinates. There is no official split, so split by patient. As measured 2026-11-14 the data are ten Parquet shards totalling about 4.9 GB; load with datasets (load_dataset('MedOtter/BCSS')) or pyarrow and decode with PIL. Used for tissue segmentation, tumor microenvironment and tumor-infiltrating lymphocyte research, and pathology foundation model evaluation. The authors release the data under CC0 1.0 in the BCSS GitHub repository; cite Amgad et al. 2019.

How to load this dataset in Python

from datasets import load_dataset
ds = load_dataset("MedOtter/BCSS")
print(ds)

Data structure

File types in sampled files: Parquet

Files

Example file: data/train-00000-of-00010.parquet

Analyze this dataset without downloading it

On Scigantic, BCSS: Breast Cancer Semantic Segmentation of TCGA H&E regions of interest opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.

This page is also available as Markdown and as JSON.

Related Health & Biomedical datasets