A second reading
If your automated cell counter can export its raw images, CellCount will count the same pixels independently and mark every cell it found on the frame, so you can see where the two readings differ. On 16 images it had never seen, concordance with a leading counter was 0.91.
When the instrument says one number and the chamber says another, the usual advice is to trust the instrument, and the usual advice comes from the company that sold it. A second reading is different from an opinion: it is the same pixels, counted again by a reader that had nothing to do with the first, with every decision drawn on the image so the two can be compared one cell at a time.
CellCount runs the same algorithm on a counter’s exported frames as on a chamber under a microscope, so a facility that runs both stops arguing about whose number is right and starts looking at the cells where the numbers differ.
From the session on your counter, export the captured images themselves, not the report and not the images with the instrument’s markings drawn on them. Only a raw frame can be counted again by another reader.
Sign in to CellCount, open the counting page and drop the export folder on it. Each frame becomes an image of one sample; the scale is read from the file when the frame has no ruled grid.
Open each frame. Every cell CellCount found is marked on it, live and dead in their own colours, with the count, the concentration and the viability beside it. Put the instrument’s own numbers for the same frame next to them.
Most disagreements are a clump one reader split and the other did not, debris one reader counted, or a size gate. Add, delete or flip the calls you disagree with; nothing is final until you accept the count.
We took sixteen images straight from a leading automated cell counter and let CellCount count the same pixels. CellCount had never seen an image from that instrument. Concordance on cell concentration was 0.91 (Lin’s coefficient), the live-or-dead call agreed on 95% of the 6,765 cells both readers found, and the mean difference in concentration was 0.002 million cells per mL with 95% limits of agreement of about plus or minus 0.29 million.
The plots, the method and the sources are on the validation page. The method, the data and every number are kept with the study and are available on request.
A clump that one reader split into cells and the other reported as one aggregate. Debris that one reader counted as a cell. A cell-size gate set differently on the two. A dilution factor typed wrong on one side. Each of these is visible on the marked image, and each is a decision a person can make and record rather than a mystery between two black boxes.
The instrument is a second measurement of the same object, not the truth. The benchmark is one preparation of one cell type on one instrument, and what it shows is agreement over that range. Whether the two readings agree on your cells is what your own export will show, which is the point of running it.