Basic workflow

1. Add one photo to inspect it, or several (or a whole folder) for a batch report. You can also drop them anywhere in the window. In a batch, click a row to open that photo, then use the arrows above the photo to step through the rest.
2. Pick a mode (Total count, Viability, or Fluorescence) and click Auto-tune settings — with the AI engine it works out the cell size and a sensible minimum size for this photo by itself.
3. Fix what it gets wrong: click a marker to remove or restore it, or click empty space to add a cell it missed. If cells are merged or over-split, adjust Cell diameter (smaller splits more) or Mask strictness (lower keeps more cells). In Viability mode, use Mark by circling to drag a loop around dead (red) or alive (green) cells, or Shift+click a single cell.
4. Download the annotated image — or in batch mode, the CSV report and a ZIP of every annotated photo. Every photo you open is also kept in History (with its live / dead counts when Viability was on), and Download CSV there saves all of them as a spreadsheet. The app keeps History on your device until you clear it; on the web page it is cleared each time the page is opened, so download it before you leave. A batch report always includes count, live, dead and viability.

Two engines

• AI model (default) — a trained neural network (Cellpose “cyto3”) that runs on your own device. It outlines each cell individually, separates touching cells, and finds faint or out-of-focus ones that edge detection misses. The web page needs internet the first time, to download the AI model and its runtime, and keeps them on your device afterwards; the desktop app has them built in. About 1–2 s per photo with a capable graphics chip (WebGPU); roughly 10–60 s without one.
• Classic — the original edge-detection + watershed method. Instant and fully offline, but it merges dense clumps and misses faint cells. The app falls back to it automatically if the AI engine can’t start.

Strengths

• Runs entirely on your device — nothing is uploaded to a server, no account needed.
• Handles brightfield/phase-contrast counting, dye-based viability (live vs. dead), and fluorescence.
• The AI engine copes with uneven lighting, faint cells and touching cells far better than the classic method did on the test photos it was checked against.
• Traces each cell’s actual outline, and every correction you make is remembered per photo in History.

Weaknesses — and what to do about them

• The AI model was trained on many cell types but not on your cultures. It is not perfect: very dense, heavily overlapping clumps can still be under-split, and some very faint cells are still missed. Try a smaller cell diameter and lower strictness, then fix the rest by clicking.
• The cell diameter is estimated automatically; for unusual photos (very sparse, very large or very tiny cells) set it by hand and untick “detect automatically”.
• Dead cells. In Viability mode choose how dead cells are found. Look (unstained, beta) is an AI model that flags cells that look bubbly, blebbing or ruptured, with debris leaking out around the outline. We tested it on photos of this cell line, but it can still make mistakes (it tends to over-flag blurry photos and can miss very sharp ones), so correct it by circling cells. Dye stain is the older rule for trypan-blue photos (uniformly dark, no bright core). Whichever you use, fix mistakes with Mark by circling → Dead (drag a loop; everything inside a red loop counts as one dead cell, a loop around nothing the detector found counts as one missed dead cell), a green loop to mark cells alive, or Shift+click one cell. Marks are saved with the photo in History. Auto-tune only sets the staining threshold when it finds two clearly separate groups.
• Manual corrections apply only to the photo you’re currently viewing — batch mode doesn’t carry them over to other photos.
• Not validated against a manually-counted ground-truth dataset. Treat counts as a fast estimate and spot-check against a manual tally, especially before using them in anything you’ll report. If you publish results from the AI engine, please cite Cellpose (Stringer et al., Nature Methods 2021).

Cell Counter is a free, non-commercial research tool. Your photos are analysed on your own device and are never uploaded.

AI engine: Cellpose

The AI engine runs the Cellpose ‘cyto3’ cell-segmentation model by Carsen Stringer, Marius Pachitariu and colleagues (HHMI Janelia Research Campus). It was converted to run in your browser, and the step that turns its output into cell outlines was re-written in JavaScript. If you publish results from it, please cite:

• Stringer, C., Wang, T., Michaelos, M. & Pachitariu, M. Cellpose: a generalist algorithm for cellular segmentation. Nature Methods 18, 100–106 (2021).
• Pachitariu, M. & Stringer, C. Cellpose 2.0: how to train your own model. Nature Methods 19, 1634–1641 (2022).
• Stringer, C. & Pachitariu, M. Cellpose3: one-click image restoration for improved cellular segmentation. Nature Methods 22, 592–599 (2025).

Cellpose code: BSD 3-Clause licence, © 2020 Howard Hughes Medical Institute (full text below). The Cellpose authors state that their models are trained on data licensed CC-BY-NC (non-commercial), which is why Cell Counter is free and non-commercial. Cell Counter is independent and is not endorsed by HHMI, Janelia or the Cellpose authors.

BSD 3-Clause License

Copyright © 2020 Howard Hughes Medical Institute

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
Neither the name of HHMI nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

Other software

• ONNX Runtime Web, © Microsoft, MIT licence. Runs the model in your browser (loaded from cdn.jsdelivr.net).
• JSZip, © Stuart Knightley and contributors, used under the MIT licence. Builds the batch download (loaded from cdnjs.cloudflare.com).
• Space Grotesk and IBM Plex Mono fonts, SIL Open Font Licence 1.1 (loaded from Google Fonts).

Dead-cell “Look” detector

A small model made for this tool from cells labelled dead or alive by hand in 18 photos of a doxorubicin series on MDA-MB-231 cells.

Please note

For research use only. This is not a medical device and must not be used for diagnosis or treatment decisions. Counts are estimates: spot-check them against a manual count before relying on them.

Privacy

Your photos and History stay on your device. The app keeps the newest 40 photos in History until you remove them; on the web page, History is cleared each time the page is opened. The web page downloads its fonts and code libraries from Google Fonts, cdnjs.cloudflare.com and cdn.jsdelivr.net the first time you open it, so those services can see your IP address; after that it normally uses the copy saved on your device. The desktop app has them built in and does not use the network at all.