Epithelial cells form diverse structures from squamous spherical organoids to densely packed pseudostratified tissues. Quantification of cellular properties in these contexts requires high-resolution deep imaging and computational techniques to achieve truthful three-dimensional (3D) structural features. Here, we describe a detailed step-by-step protocol for deep-learning-assisted cell segmentation to achieve accurate quantification of fluorescently labelled individual cells in 3D within live tissues.

We provide

• A jupyter notebook to explore how to obtain an accurate 3D segmentation from your images. It has detailed comments sections entailing: Initial segmentation using Cellpose; Automated tracking using TrackMate; Manual segmentation using napari; and refining the segmentation model in Cellpose.
• A Nextflow workflow for you to run seemingly the protocol as many times as you want. It contains the workflow exploiting Nextflow functionalities. If you want to learn Nextflow, here you can find a beautiful begginer guide.

Clone or download this repository.

Install Nextflow. You can follow this tutorial.

Google colab: Simply follow the steps here with your Google account.

Local version: Create an environment with python 3.10. We recommend using venv:

python3 -m venv cellpose_3d

but you can also use conda:

# Create an environment with python 3.10.15
conda create --name cellpose_3d python=3.10.15

Then, activate the environment.

source cellpose_3d/bin/activate

Install cellpose3 with graphical user interface and jupyter notebook

pip install cellpose[all]==3.1.0 matplotlib==3.7.3 plotly scikit-learn gdown notebook

You will first need to change the input and output directories in the config file of the pipeline named nextflow/nextflow.config. You can also change cellpose segmentation parameters in the same file.

Then, simply run:

nextflow run nextflow/nextflow_pipeline.nf

First, activate your environment:

source cellpose_3d/bin/activate

Run jupyter notebook

jupyter notebook

If you encounter any problems, please file an issue along with a detailed description.

If you use this protocol in your research, please cite the following paper:

@article{Paci2025,
   author = {Giulia Paci and Pablo Vicente-Munuera and Inés Fernandez-Mosquera and Álvaro Miranda and Katherine Lau and Qingyang Zhang and Ricardo Barrientos and Yanlan Mao},
   doi = {10.1038/s44303-025-00099-7},
   issn = {2948-197X},
   issue = {1},
   journal = {npj Imaging},
   month = {9},
   pages = {40},
   title = {Single cell resolution 3D imaging and segmentation within intact live tissues},
   volume = {3},
   url = {https://www.nature.com/articles/s44303-025-00099-7},
   year = {2025}
}

and software:

@software{Vicente-Munuera_3D_Protocol,
  author       = {Pablo Vicente-Munuera and
                  Hsu, Wilton and
                  Paci, Giulia and
                  Mao, Yanlan},
  title        = {Pablo1990/3D-deep-segmentation-protocol},
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.15469937},
  url          = {https://doi.org/10.5281/zenodo.15469937},
  swhid        = {swh:1:dir:f057288ad80c902ae809e442560901e584ccd3d5
                   ;origin=https://doi.org/10.5281/zenodo.15469937;vi
                   sit=swh:1:snp:6d437220fc5d86720d83672b636564b7cc17
                   1e6b;anchor=swh:1:rel:4f42c0e32443a72f174c1c2166ba
                   68ff0b2e4e23;path=Pablo1990-3D-deep-segmentation-
                   protocol-f9da31a
                  },
}