MicroNucML: A machine learning approach for micronuclei segmentation and the refinement of nuclei-micronuclei relationships

Benchmark dataset for micronuclei detection and segmentation. Contains tiles from training and testing (H2B-GFP.zip), as well as tiles used to evaluate the model's performance on images with different colour and quality (H2B-mCherry.zip).

Images were obtained at 20x magnification (0.625 µm/pixel resolution) from an Incucyte, a live image microscope. Cells imaged include MCF10A and RPE-1 cells exposed to a variety of DNA damaging agents, including radiation. Micronuclei were labelled using LabelStudio, then refined with SAM2. Quality checking was done manually after.
- H2B-GFP.zip : tiles labelled off of pseudo-green images. Includes tiles used in training and testing.
- H2B-mCherry.zip: tiles labelled off of original grey-scaled images, and pseudo-red images.

Preprint: https://www.biorxiv.org/content/10.1101/2025.09.20.677550v1.full.pdf
Github: https://github.com/kumarlab-compomics/MicroNuclei_Detection