Cellpose: deep-learning assistance for microscopy cell segmentation

Research applicability

Useful for researchers preparing cell masks for quantitative microscopy. This entry describes the original 2021 study, not a claim that every current Cellpose model has the same evaluation.

Getting started

Read the version-specific upstream installation guide; pilot on representative images with manual annotations before batch use. Pin the chosen release and record model/settings, image preprocessing and correction decisions.

Maintenance

GitHub API checked 2026-10-15: repository not archived; last pushed_at 2026-06-14. This is a metadata snapshot, not a maintenance or security audit.

License

The retrieved LICENSE is BSD-3-Clause, as also identified by GitHub. The README retains a GPL-v3 badge, so check the chosen release rather than relying on badges. The README separately identifies training/annotated data as CC-BY-NC; code licensing does not remove data/model restrictions.

Workflow & human oversight

A bounded image-analysis task: the model proposes cell masks. Researchers select images, inspect and correct segmentation, and validate downstream biological conclusions. This is non-agentic task assistance, not autonomous research.

Workflow level is not an evidence score; higher is not better evidence.

Problem

Biological image analysis needs cell, membrane and nucleus segmentation; specialist models can require large task-specific training datasets.

Approach

The original study trained a generalist segmentation model on varied cell images and extended its 2D model to 3D without 3D-labeled training data. Human labeling and curation remain part of the research workflow.

Outcomes (as reported by the originating team)

The authors report segmentation across varied image types and a 3D extension. The abstract reports training data containing over 70,000 segmented objects. This is an author-reported method evaluation, not a measured reduction in total research time.

Limitations

Source review used the public publisher abstract, code/data-availability sections and current repository documentation, not the paywalled full article. No independent replication was performed. Domain shift, image quality and downstream measurement validity still require checking. Current Cellpose-SAM/DINO releases differ from the original study.

Reproducibility

The publisher links code and a manually segmented dataset; the repository documents GUI/API use and example notebooks. No repository code, notebook or model was executed for this directory.

Adoption requirements

Microscopy images, domain expertise for mask quality control, a suitable Python environment and sufficient memory/compute; review code, model and dataset licenses independently.