Nuclei Segmentation (BBBC039)

Fluorescence nuclei segmentation from a high-throughput chemical screen, mirrored from the Broad Bioimage Benchmark Collection (BBBC039v1) with every nucleus converted to a browsable polygon annotation. A natural companion to the DSB2018 dataset: where DSB2018 tests generalization across wildly different imaging setups, BBBC039 tests robustness within one modality across 200 different bioactive-compound treatments (BBBC039 is in fact one of the sources DSB2018 drew from).

Contents

Images	200 fields of U2OS cells (520x696, 16-bit TIFF)
Annotations	19,379 nucleus polygons
Classes	1 (nucleus)
Dataset	images
Split	Official train (100) / validation (50) / test (50), per the source metadata
Imaging	Single-channel fluorescence, Hoechst-stained U2OS cell nuclei

Three fields legitimately contain no nuclei and carry zero annotations.

Using it

Browse any image to see the per-nucleus polygons. To download everything (free account required):

• ZIP: use the Download button on this project.
• Python SDK: pip install datatorch, then:

datatorch login
datatorch pull datatorch/nuclei-segmentation-bbbc039

Annotations export in COCO format through the project's export schema.

Conversion notes

The original release ships per-pixel class masks (background / nucleus interior / boundary between touching nuclei), not instance masks. Instances were recovered by connected-component labeling of the interior class using 4-connectivity, with the boundary class separating touching nuclei, then traced to polygons. This reproducible decode yields 19,379 instances; the source page's approximate "~23,000 nuclei" figure could not be reproduced from the official masks, so the verified count is reported here. Image filenames match the original release, so results remain comparable with published work.

License and citation

CC0 1.0 (public domain). The contributors waived copyright; attribution is still good practice. If you use this data, please cite:

Caicedo et al. 2018, image set BBBC039v1, available from the Broad Bioimage Benchmark Collection [Ljosa et al., Nature Methods, 2012].