Cellpose Options

This page documents the legacy built-in Cellpose family (segmentation_family: "Cellpose"), which uses Cellpose v.1.0.2 [citation].

Note

For the newer Cellpose-derived plugins, see:

Cellpose2 Options — Cellpose2 plugin with multi-channel support and cellprob_threshold

CellposeSAM Options — CellposeSAM plugin combining Cellpose 4 with SAM-based segmentation

The legacy Cellpose plugin source is also available in the vpt-plugin-cellpose repository. That repository README includes additional implementation notes and source-install guidance for users who need to work from a plugin checkout.

In addition to supporting the default model weights (e.g. cyto2, nuclei), vpt supports custom weights that may be more appropriate for a specific tissue type.

Input Data

The number of channels Cellpose requires is based on the model selected. For the nuclei model, only a nuclear channel is necessary (typically DAPI). For the cyto2 model, both a nuclear and entity fill channel are required.

Cellpose Model Properties

The legacy Cellpose task schema accepts the following keys in segmentation_properties:

model: Required. Selects the built-in Cellpose weights to load. Supported values are "cyto2" for whole-cell segmentation and "nuclei" for nucleus-focused segmentation.
model_dimensions: Required. Accepts "2D" or "3D". "2D" applies the 2D Cellpose model to each z-plane in the z_layers list separately, after which VPT merges the masks into a 3D output. "3D" passes the selected z-planes to Cellpose as a volumetric stack and runs the native 3D model.
custom_weights: Optional. Path to a local custom Cellpose weights file. When provided, the plugin loads this file instead of the model zoo weights named by model.
version: Optional. Free-form provenance string stored in the task JSON for record keeping. It does not select a different model implementation in the legacy plugin.

Usage:

"segmentation_properties": {
  "model": "cyto2",
  "model_dimensions": "2D",
  "custom_weights": null,
  "version": "latest"
},

Cellpose Model Parameters

The legacy Cellpose task schema accepts the following keys in segmentation_parameters:

nuclear_channel: Required. Name of the input image channel containing nuclear signal, typically DAPI. This channel should also be present in task_input_data. For "cyto2" it supplies Cellpose's nuclear guidance channel; for "nuclei" it is often the main informative channel.
entity_fill_channel: Required. Name of the input image channel containing cell body or cell boundary signal. For whole-cell segmentation this is usually a cytoplasmic or membrane-associated stain such as PolyT.
diameter: Required. Expected object diameter in pixels. Cellpose uses this value to internally rescale the image before inference. Values that are too small or too large can lead to oversegmentation or undersegmentation.
flow_threshold: Required. Flow consistency threshold passed directly to Cellpose. Higher values allow more candidate masks to pass filtering and generally increase recall; lower values apply stricter quality filtering.
mask_threshold: Required. Cell probability threshold used by Cellpose v1. Lower values accept weaker signal and generally produce larger or more numerous masks. Higher values suppress weak detections.
minimum_mask_size: Required. Minimum mask area in pixels. Masks smaller than this value are removed before VPT converts them to polygons.

Detailed descriptions of each parameter are available in the Cellpose documentation

Usage:

"segmentation_parameters": {
    "nuclear_channel": "DAPI",
    "entity_fill_channel": "PolyT",
    "diameter": 70,
    "flow_threshold": 0.95,
    "mask_threshold": -5.5,
    "minimum_mask_size": 500
}

citation: Stringer, C., Wang, T., Michaelos, M., & Pachitariu, M. (2021). Cellpose: a generalist algorithm for cellular segmentation. Nature methods, 18(1), 100-106.