Astrocyte Detection (BBBC042)

An astrocyte detection dataset of GFAP-stained rat brain tissue, mirrored from the Broad Bioimage Benchmark Collection (BBBC042v1) and used in Suleymanova et al., Scientific Reports 8:12878 (2018) to train a deep convolutional detector for reactive astrocytes. Every astrocyte in the official position files is converted to a browsable bounding box.

Contents

Images	1,117 brightfield TIFs, 990×708, 8-bit RGB
Annotations	14,989 astrocyte bounding boxes
Classes	1 (astrocyte)
Dataset	train
Split	None published; single pooled set as released
Imaging	Brightfield microscopy, GFAP immunohistochemical staining, rat brain tissue sections

The archive on the source site advertises 1,200 images, but only 1,117 ship with a matching position file in positions.zip; this project mirrors exactly that 1,117-image annotated subset, one-to-one between images/ and positions/.

Using it

Browse any image to see the per-astrocyte boxes. 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/astrocyte-detection-bbbc042

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

Conversion notes

Each positions/<id>.txt file holds one astrocyte per line in the form Cell <idx> 0 0 x1 y1 x2 y2 0 0 0 0 0 0 0. The x/y order of the four coordinate columns was not documented on the source page, so it was determined empirically: scanning the max value of every column across all 14,989 lines, the columns immediately following the two 0 0 placeholders top out at 940 and 989 — only possible if they're x (bounded by the 990px image width) rather than y (bounded by 708px). The next two columns top out at 658 and 707, consistent with y (≤708). So the columns decode as x1 y1 x2 y2, and every box satisfies 0 ≤ x1 < x2 ≤ 990 and 0 ≤ y1 < y2 ≤ 708 with no exceptions. Boxes were imported as COCO bbox ([x, y, width, height]), no segmentations. No boxes were dropped.

License and citation

CC0 1.0 (public domain). Per the source page: "To the extent possible under law, Ilida Suleymanova has waived all copyright and related or neighboring rights to BBBC042v1." Attribution is still good practice. If you use this data, please cite:

Suleymanova, I., Balassa, T., Tripathi, S. et al. A deep convolutional neural network approach for astrocyte detection. Scientific Reports 8, 12878 (2018). https://doi.org/10.1038/s41598-018-31284-x

Source: BBBC042v1, Broad Bioimage Benchmark Collection.

Related

• More microscopy datasets on DataTorch
• All public datasets