Your own microscopy · runs in this tab · exploratory
Browser lab: analyse images without uploading them
Drop a nuclear stain and a punctate marker (for example 53BP1 or gamma-H2AX), as one multi-channel file or as two files per field, or a whole batch. Files are read and analysed inside your browser; nothing is sent anywhere.
The analysis engine (engine/engine.js) did not load. You can still add and inspect files, but analysis is unavailable.
1. Add images
PNG, TIFF / BigTIFF / OME-TIFF (8, 12, 16 or 32-bit, multi-page or multi-channel), JPEG (lossy, flagged) or a ZIP of those. Up to 200 MB per file and 16 megapixels per channel.
Drop files or a folder here
No images at hand? Add a sample well from the public IDR study idr0010 (Doil et al., Cell 2009, doi:10.1016/j.cell.2008.12.041, licence CC BY 4.0; screen 1351). Each sample is a Hoechst (nuclei) and a 53BP1 image, 8-bit greyscale as rendered by the IDR. The files are served from this site and read here like any picture, so this page still makes no network request of its own; the browser re-wraps the pixels as a PNG, so the provenance SHA-256 is that of the re-wrapped file, not of the original in samples/. Image-to-annotation pairing of these wells is inferred and unconfirmed (see the Foci Lab home page).
No files yet.
| File | Type | Size | Pixels | Channels | Status | Remove |
|---|
2. Pair files and assign channels
Files whose names contain a nuclei word and a foci word are paired when the rest of the name matches. You can edit the words (comma-separated), pick a preset, or assign every channel by hand below. A single multi-channel file becomes one set by itself.
Label conditions in bulk (optional; e.g. every set whose name contains “ctrl” is “control”)
3. Parameters and engine
Checking the analysis engine…
v1 counts touching nuclei by area. v2 separates them (distance-transform watershed, CPU only; on the BBBC039 test split nucleus F1 at IoU 0.5 is 0.91 vs 0.50 and the count error drops from -17% to -10%, but that is one cell line and stain). Only the nucleus step differs; the foci counting step is identical and remains unvalidated, so a better nucleus count is not evidence of more accurate foci counts.
Each worker holds one image pair in memory while it runs; use 1 for very large images.
4. Run
5. Results
Batch table
| Image set | Condition | Replicate | Status | Nuclei | Foci | Foci / nucleus | % with foci | Backend | Open |
|---|
The provenance file records the engine version, backend, parameters, normalisation, every file name with its SHA-256 (computed here in your browser) and a timestamp, so a result can be reproduced or audited.
By condition
| Condition | Images | Nuclei | Pooled foci / nucleus | Mean of images ± SD | Replicates | Mean of replicates ± SD |
|---|
Objects are nuclei or touching clusters as found by the engine; see the warnings above for anything it flagged.
Per-nucleus table
| ID | Area (px) | Centroid x | Centroid y | Foci |
|---|
Parameters used, normalisation and what this cannot tell you
- Exploratory classical counter, not validated against manual counts.
- 53BP1 / gamma-H2AX foci are DNA-break marks in general, not telomere-specific.
- Fields from one well or dish are technical replicates, not independent experiments.