Cellpose
Deep learning-based universal cell segmentation — automatic mask generation for fluorescence and brightfield images.
Cellpose is an open-source deep-learning cell segmentation tool developed at the Janelia Research Campus (Mouse Lab). In one line: a universal mask predictor that handles essentially any microscopy image — fluorescence, H&E tissue, live-cell, electron microscopy — without per-experiment retraining.
Classical segmentation (watershed, thresholding) requires re-tuning whenever cell type or stain changes. Cellpose trains a U-Net with vector flow representations so it generalizes across morphologies and modalities without needing prior knowledge. The analogy: 'as GPT handles any text, Cellpose handles any cell image.'
For life-science researchers, the value is (1) zero-shot use on most images, (2) coverage across fluorescent, H&E, live-cell, and EM modalities, (3) a GUI that lets non-coders click their way to masks. Cellpose 2.0+ adds human-in-the-loop training: label 100-200 images and bootstrap a domain-specific model fast.
On data security: fully local execution means clinical or unpublished experimental images never leave the lab. On cost: zero license fees, CPU-only inference is possible (slow), and Cellpose 3.0 ships fast nuclei models plus SAM integration, cementing its place as the best free cell-segmentation tool.
💻 System Requirements
Minimum 8GB, recommended 16GB (32GB+ for processing large whole-slide images)
NVIDIA GPU with 4GB or more recommended (real-time performance with RTX 3060 or higher). CPU-only operation is possible, but processing time will be 10-60 seconds per image.
Cellpose model: 1 model ~50MB, total model package: under 1GB. Training data is separate.
⚡ Installation
pip install cellpose[gui]
# Run the GUI:
python -m cellpose
# Python API:
from cellpose import models
model = models.Cellpose(model_type='cyto3')
masks, flows, styles, diams = model.eval(img, channels=[0,0])
🧬 Bio Use Cases
Automatic nuclei/cell segmentation in fluorescence microscopy
Run the 'cyto3' model on DAPI + membrane-marker 2-channel images to extract thousands of cell masks in seconds (1-2s per 1024x1024 frame on GPU). Output NumPy masks enable area/intensity/adjacency analysis.
Live-cell time-lapse tracking
Combine Cellpose + omnipose for frame-by-frame segmentation, then track with trackpy/btrack. Detect division, migration, and death events. Quantify morphology changes after drug treatment.
H&E and IHC tissue analysis
Count immune cells in H&E or IHC (CD3/CD8) slides using the 'tissuenet' or a custom model. Quantify tumor-infiltrating lymphocytes (TILs) — a biomarker for immunotherapy response.
FAQ
What is Cellpose?
Cellpose is an open-source deep-learning cell segmentation tool developed at the Janelia Research Campus (Mouse Lab). In one line: a universal mask predictor that handles essentially any microscopy image — fluorescence, H&E tissue, live-cell, electron microscopy — without per-experiment retraining. Classical segmentation (watershed, thresholding) requires re-tuning whenever cell type or stain changes. Cellpose trains a U-Net with vector flow representations so it generalizes across morphologies and modalities without needing prior knowledge. The analogy: 'as GPT handles any text, Cellpose handles any cell image.' For life-science researchers, the value is (1) zero-shot use on most images, (2) coverage across fluorescent, H&E, live-cell, and EM modalities, (3) a GUI that lets non-coders click their way to masks. Cellpose 2.0+ adds human-in-the-loop training: label 100-200 images and bootstrap a domain-specific model fast. On data security: fully local execution means clinical or unpublished experimental images never leave the lab. On cost: zero license fees, CPU-only inference is possible (slow), and Cellpose 3.0 ships fast nuclei models plus SAM integration, cementing its place as the best free cell-segmentation tool.
When should I use Cellpose?
Deep learning-based universal cell segmentation — automatic mask generation for fluorescence and brightfield images.
What is a biomedical use case for Cellpose?
Automatic nuclei/cell segmentation in fluorescence microscopy: Run the 'cyto3' model on DAPI + membrane-marker 2-channel images to extract thousands of cell masks in seconds (1-2s per 1024x1024 frame on GPU). Output NumPy masks enable area/intensity/adjacency analysis.
📝 Update Notes
No update notes yet.
🧪 Related Code of Life
No related Code of Life posts yet.