Cellpose Stardist Bioimage

Cellpose + StarDist Bioimage Segmentation

Use this skill when the task is not just "run Cellpose", but choose, compare, or validate modern bioimage segmentation workflows.

Model Choice

• Use cellpose-cell-segmentation for generalist cell/nucleus segmentation across diverse microscopy images.
• Use StarDist when objects are approximately star-convex, especially nuclei in fluorescence or histology images.
• Use classical thresholding/watershed only for simple, high-contrast images or as a baseline.
• Use supervised fine-tuning when acquisition conditions differ strongly from pretrained examples.

Segmentation Workflow

1. Inspect representative images across batches, channels, magnifications, and staining conditions.
2. Decide target objects:
   • nuclei
   • whole cells
   • tissue regions
   • organelles or colonies
3. Normalize channels and intensity consistently.
4. Run a baseline model on a small representative subset.
5. Validate masks with overlays, object-size distributions, and boundary errors.
6. Tune diameter, flow/probability thresholds, tiling, and stitching for 3D or large images.
7. Export masks in a downstream-friendly format:
   • labeled TIFF
   • OME-Zarr labels
   • tables with object measurements

Quality Checks

• Check undersegmentation and oversegmentation separately.
• Evaluate per batch or imaging run; global quality can hide failed plates/slides.
• Preserve pixel size and channel metadata.
• Report model name/version, parameters, and any manual correction or fine-tuning.