Contents
What it is
The sperm chromatin dispersion test reads DNA fragmentation from how the chromatin spreads around each sperm head after treatment. A large halo means intact DNA, a small or missing halo means fragmented DNA. This dataset turns that reading into instance segmentation: every sperm in the field is outlined and classed, so a model can find each cell and call it in one pass.
| Fact | Value |
|---|---|
| Record | Zenodo 21628868, version 1, published 2026-07-27 |
| Images | 2,000 bright-field PNG, 1920 x 1080, from 40 coded slides |
| Annotated | 1,994 images, 3,743 instances (6 images have no sperm to annotate) |
| Classes | Fragmented (0), Non-fragmented (1) |
| Formats | YOLOv8-seg and COCO polygons |
| Split | by slide: 1,610 train, 200 validation, 190 test |
| Extras | a YOLOv8s-seg benchmark model and prediction code |
| Size | about 2.1 GB, one zip |
| License | CC BY 4.0 |
All counts in this table are the record's own. The microscope, stain and camera are not named on the record, and the authors advise validating any model on images from other labs or setups before relying on it.
What is in the published zip
The record describes 2,000 images, but its only version holds 1,102 image files. Slides L01 to L17 and L32 are missing, along with 36 images of L18 and 2 of L19, all in the training split. The annotation files still list every image, so 894 annotated images have outlines but no picture.
Other things a seed of the zip found, in case you train on it directly:
- Four pairs of byte-identical images, two in train and two in test.
- 15 training images with COCO outlines but no YOLO label file.
- 70 polygons that run slightly past the 1920 x 1080 frame.
- The class names are in Spanish and differ between the two COCO files ("No fragmentado" and "No Fragmentado").
- The README names an annotations/coco folder, and the archive calls it annotations/coco_json.
Every PNG decodes, the check found no class conflicts, and the COCO files match the record's counts.
Browse it in DataTorch
Dataset · datatorch/sperm-chromatin-dispersion-scd · 1,099 files
Sperm Chromatin Dispersion (SCD), partial copy
Sperm Chromatin Dispersion (SCD) assay bright-field microscopy images with polygon instance segmentation of sperm cells, classified as Fragmented or Non-fragmented (DNA fragmentation; original Spanish labels "Fragmentado" and "No fragmentado"/"No Fragmentado"). PARTIAL COPY: the published Zenodo archive contains only 1,102 of the record's 2,000 images; slides L01 to L17 and L32 (plus 36 images of L18 and 2 of L19), 898 train images in all, are missing from it. This copy holds 1,098 images (1,920x1,080 PNG; train 710, val 200, test 188, after removing 4 byte-identical duplicates) with 2,014 polygons (701 Fragmented, 1,313 Non-fragmented), and will be completed if the authors re-upload the missing images. License: CC BY 4.0. Source: https://zenodo.org/records/21628868. Please cite: Ornelas Nande, E. J., Adel Domínguez, M. A., Mora Topete, A., Magdaleno Márquez, L. E., Aranguren, I., Valdivia-G, A., & Sanchez, O. D. (2026). Sperm chromatin dispersion microscopy image dataset for instance segmentation [Data set]. Zenodo. https://doi.org/10.5281/zenodo.21628868
Open →
The hosted copy has the 1,098 distinct images in the zip, with 2,014 polygons (701 Fragmented, 1,313 Non-fragmented) and the record's split kept as train 710, validation 200 and test 188. The class names are given in English, with the original Spanish noted in the project description. The list of what is missing is in the project too, as errata.txt. If the authors upload the missing slides, the copy will be completed.
Open any image to see each outline with its class, zoom into the halos, and correct or add outlines with your team. Exports come out as COCO.
What people use it for
Training and benchmarking instance segmentation models that count sperm and call DNA fragmentation from SCD slides, in place of scoring by eye under the microscope. The record ships a YOLOv8s-seg baseline to compare against.
Source and citation
Source: https://zenodo.org/records/21628868 (read 2026-11-09). A Data Descriptor with the same title is under review at Scientific Data. Cite: Ornelas Nande, E. J., Adel Domínguez, M. A., Mora Topete, A., Magdaleno Márquez, L. E., Aranguren, I., Valdivia-G, A., and Sanchez, O. D. (2026). Sperm chromatin dispersion microscopy image dataset for instance segmentation [Data set]. Zenodo. https://doi.org/10.5281/zenodo.21628868
Label your own microscopy slides
Outline cells, review each other's calls and export COCO, in the browser.
Frequently asked questions
What does the SCD dataset label?
Each sperm head with its halo is outlined as a polygon and given one of two classes, Fragmented (class 0) or Non-fragmented (class 1), the reading of the sperm chromatin dispersion test for DNA fragmentation.
How many images are in the SCD dataset?
The record describes 2,000 PNG images from 40 coded slides (L01 to L40), 1,994 of them with at least one annotated sperm, 3,743 instances in all. The published zip holds 1,102 of the image files. The hosted copy has the 1,098 distinct ones.
Why is the hosted copy partial?
The only version of the Zenodo zip is missing slides L01 to L17 and L32, plus 36 images of L18 and 2 of L19, all from the training split. The copy holds what the zip holds and will be completed if the authors upload the missing images.
How is it split?
By slide, to reduce leakage between sets. The record gives 1,610 train, 200 validation and 190 test images. The hosted copy has 710, 200 and 188, after the missing files and four exact duplicates.
What is the license of the SCD dataset?
Creative Commons Attribution 4.0 International (CC BY 4.0). Cite the Zenodo record, DOI 10.5281/zenodo.21628868.
Open the hosted copy on DataTorch.
Browse the files in your browser, then download the dataset with a free account.
