Cellpose Segmentation

Cellpose is a deep learning model for cell and nucleus segmentation. Synapse includes a lightweight ONNX-based Cellpose integration that runs without PyTorch.

Nodes

Cellpose Segment

Single-image interactive segmentation.

Direction	Port
Input	image
Output	mask
Output	label_image
Output	overlay

Properties:

• Model — nuclei, cyto, cyto2, cyto3
• Channel — select which channel contains the signal
• Diameter — approximate cell diameter in pixels (0 = auto-detect)

Cellpose Batch

Batch segmentation of a folder of images.

Direction	Port
Input	folder
Output	table
Output	label_images

Properties: Same as Cellpose Segment, plus folder and pattern selection.

Model Selection Guide

Model	Best For
nuclei	Round, well-separated nuclei (DAPI, Hoechst)
cyto	Cytoplasm with nuclear marker in second channel
cyto2	Improved cytoplasm model
cyto3	Latest cytoplasm model, best general-purpose

Tips

First-time use

Models are automatically downloaded from HuggingFace on first use. An internet connection is required for the initial download only.

Diameter

If your cells vary significantly in size, set diameter to 0 for auto-detection. For consistent cell sizes, setting an explicit diameter improves accuracy.