Images & uploading
What is an “image set”?
One field of view: one place on the slide, photographed in one or more channels. A set is either a single merge file, or the separate channel files of that field (DAPI, GFP, red, optional merge) dropped together. Four files of one field are one set, not four. Four merges of four fields are four sets.
Is the merge alone enough?
Yes. The merge your software exports is a plain red / green / blue stack of the channels (red = 5-HTP or tdTomato, green = GFP, blue = DAPI), so the app splits it back into its three planes at upload and counts them like separate files. The card says “merge split into 3 channels”. Counts from a merge and from the single channels come out the same.
If you drop both the merge and the single channels of one field, the single channels are used and the merge is kept as a picture to look at.
Do I drop files one at a time or all together?
All together, always: select every file of the batch in one go, or drop the folder. One merge per section is the simplest. If you use single channels, drop the three (or four) files of a field together; they are grouped by name. Merge and single channels of the same field together is fine too: the single channels are used, the merge is kept for viewing. One file at a time also works, it is just slower, because the app regroups after every upload.
When would I still upload single channels?
When your merge is not a clean red / green / blue stack: for example if a channel was coloured magenta or cyan, or brightness was changed per channel before merging. The app warns when two planes of a merge carry the same picture. Single channels always work.
How does it know which file is which channel?
From the last keyword in the filename: dapi, gfp, htp, tom, red, merge (and a few synonyms like hoechst, tdtomato, rfp, mcherry). It then checks that the pixels agree, e.g. that a “gfp” file really has its signal in the green plane, and warns if not. You can change any role on the set card before counting.
How does it group files into sets?
By what is left of the filename once the channel word is removed. naive 1 gfp.tif, naive 1 htp.tif and naive 1 dapi.tif share the key “naive 1” and become one set. Files in different folders never mix, so dropping one folder per field is the easiest way to keep fields apart. Several merges that share a name (four merge.tif from four folders, or merge (1)…(4)) become one set each.
If the grouping is wrong, drag a file from one card to another, or remove a set and let the files fall into “Files not in a set”, then assign them.
Which formats and magnifications?
TIFF, PNG or JPEG, 8-bit RGB as exported by the microscope software. Prefer TIFF or PNG: JPEG compression can add a couple of faint false objects. The pixel size is read from the TIFF metadata (1.54 px/µm = 10x, 3.08 = 20x, 6.15 = 40x); plain PNG / JPEG have none, so write 10x, 20x or 40x in the filename or pick it on the card. The magnification sets the expected cell size and the µm scale of everything.
Does a burnt-in scale bar matter?
No. White overlays like scale bars and their text are masked out before counting and before the tissue area is measured. A very faint anti-aliased edge can occasionally leave one dim blob next to the bar; one click removes it.
How many sections per batch?
Up to 8. A batch is one “run” in the history; name it after the experiment and give each section a condition (naive, DSS d4, …) on its card, the CSV carries both.
Can I add more images after the first ones?
Yes, as long as the batch has not been counted: use Add images / Add a folder next to the Images heading, or drop files anywhere on that column. A file that is already in the batch is not uploaded twice; you get a notice instead. Once a batch is counted, start a new batch for more sections.
What about brightfield H&E photos?
A photo with a white background is recognised as H&E and counted for crypts instead of marker cells: every ring of stained epithelium around a pale lumen is one crypt. Goblet cells are deliberately not counted. H&E photos are always a set of their own.
How it counts
What does it actually count?
Bright blobs of about one cell’s size in the GFP plane and in the red plane, separately. Each accepted blob is one marker-positive cell: GFP+ (green ring) and Red+ (red ring). A GFP cell and a red cell at the same place are one double-positive cell (yellow ring). DAPI is never counted as a marker; it has its own nuclei estimate, see below.
How does the detector work, in plain words?
Per channel: the slow background glow is removed, the image is lightly blurred, and everything brighter than an automatic threshold becomes a candidate. The threshold is chosen from the image itself (Otsu’s method or a robust noise estimate, whichever is higher), measured only on tissue, and capped so that at most 3 % of the field can light up; if the cap kicks in you see a high background flag. The threshold never starts below 36 grey levels: on a dim channel, such as an 8-bit JPEG whose cells are only a little brighter than the tissue glow, the automatic estimate would sit inside that glow and a third of the field would count, so it starts at the floor and you see a dim channel flag. The sensitivity slider still moves it either way. Candidates are then kept only if their size fits a cell (0.2 to 6 cell areas), and big clumps are split into single cells by a watershed. No neural network, no GPU, about two seconds per channel, and the same code runs on your laptop and on the server.
What is the “cell diameter”?
The one number everything scales from: 15 px at 10x, 26 px at 20x, 40 px at 40x (about 8 µm). It sets the blur, the size window, the split distance and the double-positive radius. Change it in Detection settings only if your cells are clearly bigger or smaller than that; recounting keeps your manual edits.
What does “sensitivity” do?
50 is the automatic threshold. Every +25 halves the threshold (finds dimmer cells, risks debris), every −25 doubles it (stricter). The exact threshold used is shown next to the slider. Sensitivity is per channel and per section, and it is written into the CSV so a count can always be reproduced. Once one section of a batch is right, Apply to the other images in this batch under the sliders recounts the other unverified sections of the same magnification with these settings (sensitivity, sizes, diameter, split, nucleus filter); your added and removed cells are kept and verified sections are left alone.
How is a double-positive decided?
A GFP cell and a red cell whose centres are closer than 0.75 cell diameters are paired, one to one, nearest first. Then every cell still without a partner gets a second look in the other plane. A double-positive cell is often bright in one channel and dim in the other, too dim for that channel’s threshold, which is set strictly so that tissue texture never counts on its own. Under a cell the other channel has already found, the only question is whether there is signal there at all, so a lower bar applies: 70 % of the threshold, and never below what empty tissue shows. If the signal passes, the dim partner is added and the cell becomes a double. Such partners say “dim, confirmed under R17” when you hover or click them, the inspector’s audit line counts them, and the CSVs carry them (gfp_confirmed, red_confirmed, confirmed_by). Recounting a channel rebuilds them; a partner you removed stays removed.
The pair is drawn once, as a yellow ring on the GFP cell, with both labels (G12/R9). Cells you add by hand pair the same way, and you can also add a cell directly as “both”.
What does the DAPI filter do?
Optional. When on, a GFP or red cell is only kept if there is a DAPI nucleus under it; cells without one are set aside as “no nucleus” ghosts you can restore. Useful when autofluorescent debris is a problem; off by default.
How are the nuclei counted?
Approximately, by shape. The DAPI plane is filtered at the size of a nucleus (about half a marker cell), which turns every nucleus into a bright spot even where nuclei touch, and one nucleus is counted per spot that stands out of the noise between nuclei. It works on dim DAPI too. In tightly packed crypt epithelium two nuclei can still merge into one, so the number is shown with “≈” and should be read as an estimate, good for comparing sections, not for a census.
How is the tissue area measured?
The brightest of the channels per pixel, smoothed at half a cell, is bright on tissue and black on lumen or empty glass. If the field has a real black background, everything darker than the floor plus a small step is background; gaps are bridged, small holes (crypt lumina) are filled, and the result is the dashed outline you see on the image (key T). The outline is deliberately a little generous at the edge so cells sitting on the lumen border stay inside. A field that is tissue edge to edge counts whole, and the caption says so.
What are the limits?
- Touching cells are split by shape; a tight clump of three may count as two.
- Very dim cells below the automatic threshold are missed; raise the sensitivity and check the ghosts.
- Bright debris of cell size is counted; one click removes it, and the DAPI filter can pre-empt it.
- The nuclei number is an estimate (see above). The marker counts are the exact ones.
- JPEG artefacts can add faint objects; prefer TIFF or PNG.
The numbers
Total cells
Distinct marker-positive cells: GFP+ plus Red+ minus the double-positives, so a double is one cell. For H&E sets it is the crypt count.
GFP+, Red+, Both and the percentages
GFP+ and Red+ include the double-positives (a double cell is GFP+ and Red+). “Both” is the number of pairs. The percentage next to each is its share of the total; the results footer gives the share of GFP+ cells that are co-labelled.
Cells per mm²
Total cells divided by the tissue area inside the dashed outline, never by the whole frame, so the lumen and black background do not dilute it. The caption under the total shows the tissue area and what share of the field it is. If a file has no pixel size (plain PNG / JPEG without a magnification), there is no density until you pick the magnification.
Nuclei ≈ and “% of nuclei”
All DAPI nuclei in the tissue (estimate), and each marker population as a share of them: GFP+ cells ÷ nuclei. Handy to compare sections of different size without relying on area.
Detected → your edits → final calls
The funnel in the inspector is the audit trail: how many calls the detector made per channel (a double counts once per channel), how many you removed or added, and the final number. “Kept” are detector calls you did not touch. It is all in the CSV too.
Mean ± SD
The results footer averages every column over the sections of the batch (n = number of fields). The SD is the sample standard deviation.
Checking & editing
What do the rings and colours mean?
Green = GFP+ only, red = Red+ only, yellow = both, teal = crypt. A ring with a centre dot was added by you. A dashed grey ring is a cell you removed (a ghost: click it to bring it back). Small dashed rings are candidates the detector set aside (wrong size, no nucleus); show them with J and restore any that are real. Labels are G1, G2… and R1, R2…; cells you add are Gm1… / Rm1….
How do I look at one channel on its own?
The View chips under the set name: All is the composite, GFP, 5-HTP, DAPI and Merge show that single image. Shift-click a chip (or Shift+1…4) to toggle it into the composite instead. Each channel opens at a brightness that makes the tissue visible; the sliders in Detection settings change the display only, never the count.
How do I add, remove or reclassify a cell?
Select tool (V): click a ring to open its popover, where you can change its class (GFP / red / both) or remove it. Click empty tissue to add a cell as GFP, red or both. Add tool (A): every click adds a cell of the class chosen in By marker (keys 1 2 3). Remove tool (R): click rings to drop them, click ghosts to restore. Nothing is ever deleted: removed cells stay on record as ghosts.
Undo and redo
Ctrl+Z undoes the last edit, Ctrl+Shift+Z redoes it. A class change or “add as both” is one step. Edits are stored on the server, so they survive a reload.
What happens to my edits when I recount?
Cells you added stay. Removals and restores are re-applied to the nearest new object within half a cell diameter; if a removed cell is not found again, the app tells you how many edits could not be re-applied.
What does “Mark verified” do?
It records that a person signed the counts of that section off, with a timestamp that goes into the CSV. Any later change to the counts clears it, so a verified section is always one nobody touched afterwards. Verify all at the bottom signs off every section of the batch.
Results & export
What is in the summary CSV?
One row per section with: batch and section names, condition, magnification and px/µm; field and tissue area in mm², tissue fraction and cells per mm² of tissue; GFP and red counts split into auto / added / removed / final plus how many were dim cells confirmed under the other marker, double-positives and their percentages; DAPI nuclei and the marker shares of nuclei; whether the section was counted from the merge or from channels; every detection setting (thresholds, sensitivity, diameter, size window, watershed, DAPI filter, pairing radius); the second-opinion counts if any; analysed and verified timestamps.
What is in the cells CSV?
One row per cell (accepted, removed or set aside) with its section, channel, label, class, position in px and µm, area, intensities, whether it was detected or added by hand (and, for a dim cell, which marker confirmed it), its status, and its partner label and distance for double-positives.
Annotated PNG and zip
The annotated PNG is the image exactly as you see it: the channels you have on, at the current brightness, with rings, labels, legend, scale bar and the tissue outline. The zip of a batch contains both CSVs, every annotated PNG, the parameters as JSON and the run metadata.
Troubleshooting
I dropped four pictures and got one image set
They were the four channels of the same field (dapi, gfp, htp, merge share the same name key), which is exactly one set. To count four different sections, drop four merges, or one folder per field.
“Next image set” is not there
The batch has only one section, so there is nothing to step to. It appears as soon as a batch has two or more sets.
I cannot see a channel, the image looks black
The channel is probably just dim: tissue glow in a 5-HTP plane is only a few percent of full brightness. Each channel now opens at an automatic brightness; if it is still dark, use the Display brightness slider in Detection settings, or click that channel’s View chip to see it alone. Brightness never changes the count.
Uploading seems to do nothing
Three usual reasons: the file is already in the batch (you get a notice), the batch was already counted (start a new batch), or the file is not TIFF / PNG / JPEG. Anything else and the toast at the bottom tells you what went wrong.
The counts look wrong
First check the magnification on the card: a wrong one makes every cell the wrong size. Then look at the ghosts (X, J) to see what was set aside, and move the sensitivity in steps of 10 with Recount. If one channel has a high background flag, lower its sensitivity; a dim channel flag means the threshold started at the floor of 36, so raise the sensitivity if real cells are missing. For a handful of cells, clicking is faster than any setting.
The tissue outline is wrong
If a field has no real black background the whole frame counts as tissue and the caption says so. Very dim tissue at the edge of a section can fall outside the outline; the outline is already generous by about half a cell, and the cells themselves are counted regardless of the outline, only the per-mm² number uses it.
The nuclei number looks off
It is an estimate: very dense epithelium undercounts a little, and bright out-of-focus blobs can add a few. Compare sections with it, do not chase single nuclei. The marker counts are exact and editable; the nuclei count is not editable.
Two of my files were called duplicates
Two files of the same channel landed in one set, e.g. gfp.tif and gfp-1 SB.tif. The cleaner one (without a scale bar, shorter name) is used and the other is ignored. If they are really two different fields, give them different names or folders.
Nothing matches that. Try another word, or clear the box.