Skip to content

CellMigrationLab/NucleiSky

Repository files navigation

NucleiSky Banner

NucleiSky ✨🔬

Constellation-based point-set registration for microscopy in 2D and 3D Align calibrated partial views to whole-slide images or thick tissue volumes using landmark geometry.

PyPI License Downloads DOI Open 2D App in Colab Open 3D App in Colab


✨ What is NucleiSky?

Microscopy alignment gets messy when images are rotated, scaled, or captured on completely different platforms. NucleiSky solves this by treating your nuclei like stars in a constellation.

Instead of relying on pixel intensity similarity, NucleiSky matches the geometry of the cells. Whether your data is a 2D field of view or a 3D tissue stack, NucleiSky finds the calibrated similarity transform that best overlaps two landmark constellations.

👩‍🔬 For the Biologist: The "Telescope" Setup

Imagine looking through a small telescope at a random patch of the night sky. Even without seeing the whole galaxy, you can figure out exactly where you are by matching your small star pattern against a full sky map. NucleiSky does the same for your tissue: it anchors high-magnification ROIs into whole-slide scans by estimating rotation, scale, and shift.

👨‍💻 For the Developer: The Constellation Engine

NucleiSky is a robust point-set registration pipeline built for real-world microscopy noise. It provides modular feature extraction, dynamically scaled RANSAC, geometric hashing, and tetrahedral pyramid matchers — ready to slot into automated spatial analysis pipelines.


⚡ Quickstart

Choose your setup and align your images in minutes:

🌟 Try it in the Browser (Fastest)

No installation required. Run our interactive apps directly in Google Colab:

🖥️ No-Code Desktop GUI

Prefer a local app? Use our standalone desktop installer.

🐍 Python Workflows (Local Installation)

To install the package for your own scripts:

pip install nucleisky

Use pip install "nucleisky[all]" when you also need optional segmentation, OME-Zarr, SimpleITK, and notebook dependencies. Core matching runs on CPU; GPU support only affects optional deep-learning segmentation backends such as Cellpose and InstanSeg.

2D Pipeline (ROI → Whole Slide)

  1. 2D Data Preparation
  2. 2D End-to-End Workflow
  3. Quality Control

3D Pipeline (Subvolume → Thick Tissue)

  1. 3D Data Preparation
  2. 3D End-to-End Workflow
  3. Quality Control

(Need a leaner installation? See our Installation Guide for modular backend options).


🔑 Key Capabilities

  • 2D & 3D registration: Shared constellation logic for flat slides and volumetric stacks. 3D workflows should be reviewed with manual QC before downstream use.
  • Scale-aware: Matches images captured at different pixel/voxel sizes when calibration metadata is correct.
  • Rotation-aware: Supports unrestricted rotations unless you configure an angle bound.
  • Modality-Agnostic: If you can segment the nuclei (or spots/cells), NucleiSky can match them.
  • Export-Ready: Saves transforms, warped hyperstacks, and QC overlays for downstream analysis.

⚠️ The Two Golden Rules (Make or Break)

  1. Pixel and Voxel size matter immensely. NucleiSky measures physical distances in micrometres. If your scale metadata (µm/px) is missing or wrong, the search bounds break, and alignment fails. Always verify your metadata! See Data Preparation.
  2. Segmentation quality drives match quality. NucleiSky registers point sets. Missing, massively merged, or noisy nuclei obscure the constellation and make matching much harder. See Segmentation.

📚 Documentation Index

Start Here

Quality Control & Troubleshooting

Developer Deep Dives


🤝 Contributing

We welcome PRs, feature ideas, and bug reports! For the fastest triage, please use our GitHub issue forms.

If you’re using NucleiSky in a new workflow, drop an issue and tell us:

  • What you’re aligning (restains, multi-round, ROI-to-reference, etc.)
  • What success looks like in your pipeline
  • What would make adoption easier for your lab

Need troubleshooting help? Please include a QC overlay from your save_dir/original/ folder when you open a bug report!


📢 Citation

If you find this tool useful in your research, please cite:

NucleiSky enables cross-scale multimodal registration of microscopy data using nuclei constellations.

Iván Hidalgo Cenalmor, Adán Olguín-Olguín, Carolina Prieto, Johannes Kumra Ahnlide, Pontus Nordenfelt, Ricardo Henriques, Mario Del Rosario, Guillaume Jacquemet bioRxiv 2026.06.29.735028;

doi: https://doi.org/10.64898/2026.06.29.735028

@article {Cenalmor2026,
	author = {Cenalmor, Iv{\'a}n Hidalgo and Olgu{\'\i}n-Olgu{\'\i}n, Ad{\'a}n and Prieto, Carolina and Ahnlide, Johannes Kumra and Nordenfelt, Pontus and Henriques, Ricardo and Del Rosario, Mario and Jacquemet, Guillaume},
	title = {NucleiSky enables cross-scale multimodal registration of microscopy data using nuclei constellations},
	elocation-id = {2026.06.29.735028},
	year = {2026},
	doi = {10.64898/2026.06.29.735028},
	publisher = {Cold Spring Harbor Laboratory},	
	URL = {https://www.biorxiv.org/content/early/2026/06/29/2026.06.29.735028},
	eprint = {https://www.biorxiv.org/content/early/2026/06/29/2026.06.29.735028.full.pdf},
	journal = {bioRxiv}
}

About

Image registration and ROI matching for microscopy images using nuclei constellations Resources Readme

Resources

License

Contributing

Stars

Watchers

Forks

Packages

 
 
 

Contributors