Virtual staining and segmentation of nuclei and membrane from quantitative phase

These models are used to analyze the data reported in the following preprint:
Mantis: high-throughput 4D imaging and analysis of the molecular and physical architecture of cells

The virtual staining models predict nuclei and cell membranes from quantitative phase images. The quantitative phase images are computed from the defocused transmitted light microscopy stacks using recOrder.

The predicted nuclei and membrane can be segmented with a variety of instance segmentation models. We share CellPose models for the segmentation of nuclei and cell membranes.

Virtual Staining

Make a new python environment and install the version of VisCy (v0.1.0a0) used for the mantis preprint.
For a detailed installation guide, see the GitHub repository.

# This will also install Cellpose
pip install "viscy[metrics] @ git+https://github.com/mehta-lab/VisCy.git@v0.1.0a0"

Run inference of cell nuclei and plasma membrane by modifying the input and output Zarr store paths in ./viscy_model/predict.yml.
See the usage guide in the repository.

Segmentation

To segment nuclei with our fine-tuned Cellpose model, use the following Python snippet:

from cellpose.models import CellposeModel
model = CellposeModel(model_type="./cellpose_model/CP_20220902_NuclFL")
segments = model.eval(...)

Refer to the Cellpose documentation for API/GUI usage.