GoudaCell

Cell segmentation and feature extraction on the Whitehead HPC using Cellpose.

Features

• Segmentation modes: nuclei-only, cells-only, or dual (both)
• Feature extraction: CellProfiler-equivalent morphological and intensity features
• File formats: TIFF, Nikon ND2, DeltaVision (.dv)
• Cellpose 3 & 4: Supports both versions with automatic model selection

Getting Started

1. Set Up Your Environment (one time)

# Clone the repository on fry
git clone https://github.com/cheeseman-lab/goudacell.git
cd goudacell

# Create the environment
conda create -n goudacell -c conda-forge python=3.11 uv pip -y
conda activate goudacell

# Install goudacell (choose ONE):
uv pip install -e ".[cellpose3]"  # For most cells (rounded shapes)
uv pip install -e ".[cellpose4]"  # For complex cell shapes

# Register as a Jupyter kernel
python -m ipykernel install --user --name goudacell --display-name "goudacell"

2. Test Parameters Interactively

# Start Jupyter on a GPU node (run from goudacell directory)
cd /path/to/goudacell
sbatch scripts/jupyter_gpu.sh

# Check the output file for the URL
cat goudacell_jupyter-*.out

Open the notebook at notebooks/segmentation.ipynb and:

1. Set your image directory and file pattern
2. Choose segmentation mode: "nuclei", "cells", or "dual"
3. Adjust parameters (diameter, thresholds) using the sweep cells
4. Run feature extraction (optional)
5. Generate batch config when happy with results

3. Run Batch Segmentation

# Use the config file generated by the notebook
sbatch scripts/run_segmentation.sh /path/to/segmentation_config.yaml

Segmentation Modes

Mode	Output	Use case
nuclei	*_nuclei_mask.tif	Nuclear segmentation only
cells	*_mask.tif	Cell segmentation only
dual	*_nuclei_mask.tif + *_cell_mask.tif	Both nuclei and cells

Feature Extraction

Extract CellProfiler-equivalent features from segmented images (~100+ features per compartment):

• Intensity: mean, std, min, max, median, quartiles, edge intensities
• Shape: area, perimeter, solidity, eccentricity, Zernike/Hu moments
• Texture: Haralick (13), PFTAS (54)
• Distribution: radial intensity distribution
• Correlation: channel correlation, colocalization metrics
• Neighbors: counts, distances, angles
• Foci: count and area per channel (optional)

Which Cellpose Version?

Version	Install with	Use for
Cellpose 3	.[cellpose3]	Round cells (most common)
Cellpose 4	.[cellpose4]	Irregular/complex shapes

File Formats Supported

• TIFF (.tif, .tiff)
• Nikon ND2 (.nd2)
• DeltaVision (.dv)