The words that come up around cell counting, each in a paragraph, with the study behind any figure. Where CellCount does something with a term, that is one sentence at the end, not the definition.
A thick glass slide with a chamber of known depth, usually 0.1 mm, and a ruled grid etched into its floor. A cell suspension is loaded under a cover glass, cells are counted in squares of known area under a microscope, and the count is converted to a concentration. It is the reference method most automated counters are validated against.
The most common hemocytometer grid: a 3 mm by 3 mm area divided into nine 1 mm squares, with the four corner squares split into sixteen and the centre square into twenty-five smaller squares. One 1 mm square over a 0.1 mm depth holds 0.1 microlitre. Bright-line versions etch the rulings so they show as bright lines on a dark field.
A viability stain. Trypan blue cannot cross an intact cell membrane, so live cells exclude it and stay clear, while cells with a compromised membrane take it up and turn blue. Counting clear and blue cells separately gives live and dead counts. The stain is also a common dilution step, since sample and dye are often mixed one to one.
The share of counted cells that are alive, given as a percentage: live divided by live plus dead. With trypan blue it is the share of clear cells among clear and blue. It describes the condition of the population at the moment of counting, so the time between staining and counting matters, because the dye eventually enters live cells too.
Cells per millilitre of the original suspension. From a hemocytometer it is the average count per 1 mm square, multiplied by 10,000 (because a square over a 0.1 mm depth holds 0.1 microlitre), multiplied by the dilution factor. Any error in the square count, the chamber depth or the dilution passes straight into the concentration.
How many times the original suspension was diluted before counting, as a multiplier applied to the concentration measured in the chamber. Mixing one part sample with one part trypan blue is a dilution of 1:1, a factor of two. A wrong dilution factor is one of the commonest reasons two people report different concentrations from the same chamber.
Which of the nine large squares are counted, and in what order. Counting the four corners and the centre is a common convention, and so is counting a single corner square when cells are dense. The pattern should be fixed in a written protocol, because the squares of a chamber do not always hold the same number of cells and changing the pattern changes the answer.
How much counts of the same sample differ between people. Published figures put it at 14.6% between laboratories counting latex beads of a known concentration, and at roughly 20% between two blinded experts counting the same cell cultures. It has a name because it is everywhere, and it is the reason a lab wants one counting standard rather than one per person.
Source: Zuvela E, Matson P. Reprod Biomed Online. 2020;41(4):671-678. doi.org/10.1016/j.rbmo.2020.07.008
How much one person's repeated counts of the same sample differ from each other. On latex beads of a known concentration, the cleanest possible case with no clumps and no live-or-dead call, published repeat counts spread by about ~5% (3.0 to 6.7 percent). Real cells with a stain spread more.
Source: Costa AF, Onofre FBM, Onofre ASC. JBRA Assist Reprod. 2024;28(2):215-223. doi.org/10.5935/1518-0557.20240023
A measure of agreement between two methods measuring the same thing. It multiplies precision (how tightly the pairs correlate) by accuracy (how close they sit to the line of equal readings), so two methods that track each other but sit apart do not score. One is perfect agreement. CellCount against a leading automated counter, on 16 images, reached 0.91.
Source: Lin LI. A concordance correlation coefficient to evaluate reproducibility. Biometrics. 1989;45(1):255-268. doi.org/10.2307/2532051
A plot of the difference between two methods against their mean, one point per sample. The average difference is the bias; the 95% limits of agreement are that average plus and minus 1.96 standard deviations, the band most differences fall within. In CellCount's benchmark the mean difference was 0.002 million cells per mL with limits of about plus or minus 0.29 million.
Source: Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;327(8476):307-310. doi.org/10.1016/S0140-6736(86)90837-8
Images kept out of a model's training and used only to test it, so the test says how the model does on frames it has never seen rather than on frames it memorised. CellCount's shipped model is scored on such a set, where its live-or-dead call agreed with an expert annotator on 98.7% of the 2,682 cells both found.
The decision, for each detected cell, whether it is live or dead. In a trypan blue image that is whether the cell is clear or blue. Accuracy of the call is measured on the cells that both the model and a human annotator found, so it is separate from whether a cell was found at all, which is detection.
How confident the model must be before it marks a cell. A higher sensitivity keeps fainter and less certain detections, which finds more real cells in a dim or crowded image at the cost of more false marks; a lower one keeps only confident detections. CellCount offers High, Balanced and Low, and the setting is part of the shared protocol, so a lab does not drift from image to image.
A minimum and maximum cell size, in micrometres, applied to the detections: anything smaller than the gate (debris, fragments) or larger than it (clumps, bubbles) is excluded from the count. The gate needs a known scale, which a hemocytometer image supplies through its rulings. It is set once in the protocol and shown as a shaded band on the size histogram.
The single size given to a detected object so it can be compared with a gate. CellCount uses the geometric mean of the width and height of the detection box, so a thin fibre 40 by 4 pixels reads about 13 pixels rather than 40, and is excluded by a gate set for round cells. The pixel size is converted to micrometres with the scale measured from the chamber rulings.
The distribution of equivalent diameters across all detected cells in a sample, pooled over its images. It shows whether a population is uniform or mixed, where debris and clumps sit, and where the size gate should go. Pooled per sample rather than per image, because a gate set on one corner square would strand cells in the others.
The unprocessed image a camera or an automated counter captured, as opposed to an image the instrument rendered with its own markings drawn on top. Only a raw frame can be counted again by another reader; an overlay image carries the first reader’s decisions baked into the pixels. When exporting from a counter for a second reading, export the raw frames.
The fixed set of choices that turns an image into a number: the dilution, which squares are counted, the sensitivity and the cell-size gate. Written once and shared by everyone who counts, it is what makes two people, two instruments or two days comparable. In CellCount a protocol is set for the team and applied identically to a chamber image and to a counter’s exported frames.
The record is the complete description of an accepted count: the counts, the grid, the dilution, the protocol, the model version, the person who accepted it and when. The digest is a SHA-256 hash computed over the record’s exact bytes and stored with it, so anyone holding the record later can recompute the hash and confirm nothing has changed since acceptance.
Peripheral blood mononuclear cells: the lymphocytes and monocytes isolated from blood, usually by density-gradient separation. They are small and faint-edged in brightfield, and preparations often carry red blood cells and platelets that a trypan blue count can mistake for cells, which is why a stain, a size gate and a person checking the marked image matter more here than for a cell line.
Primary cells are taken directly from tissue or blood and divide a limited number of times; they vary in size and condition from donor to donor. Immortal cell lines divide indefinitely and are uniform, which is why most counting benchmarks use them. A counting method that only works on a uniform line is not yet a counting method for a lab.
The range of image conditions within which a counting method is reliable: magnification, resolution, focus, exposure and framing. For a chamber image the framing rule is that one full 1 mm square, with all four corners, is in the frame and the rulings are crisp; outside that envelope the grid cannot be measured and the concentration cannot be trusted.