Camera Count

This is worth doing on paper before it is done with a purchase order, because the arithmetic is unkind and entirely predictable.

Time at a field of view divides into work that parallelizes and work that does not. Channel exposure and readout parallelize; that is exactly what extra cameras buy. Moving the stage, waiting for it to settle, and autofocusing do not: they happen once per field regardless of how many sensors are watching.

Take a plate scan where 60% of the per-field time is stage and focus and 40% is acquiring four channels in sequence. Add three more cameras so all four channels are simultaneous, and the acquisition fraction drops from 0.40 to 0.10. Total per-field time goes from 1.00 to 0.70, so the scan runs 1.43× faster, not 4×. Two cameras give 0.60 + 0.20 = 0.80, which is 1.25×. The second camera delivered most of what was available and the third and fourth split what was left.

The ceiling is fixed by the part that does not parallelize: even with infinite cameras this scan cannot beat 1/0.60 = 1.67×. If that is not the improvement you were promised, the honest next step is to measure your own split before buying anything, because the same argument that limits the gain also tells you where the gain actually is.

The calculation above points somewhere useful: if 60% of your time is stage and focus, that is where a factor of two is hiding, and it is usually cheaper to reach than a second camera.

Things that commonly dominate, in rough order of how often they turn out to be the answer. Autofocus per field, when a focus map measured once across the plate would serve; a slide is flat enough to interpolate over, and focusing every field is often pure repetition. Stage settle time set conservatively at installation and never revisited. Filter wheel moves ordered by biology rather than by wheel position. Writing data to a network drive rather than a local disk, which turns readout into a bottleneck at high frame rates.

The measurement takes one scan: log a timestamp at each transition and total them by category. Most people are surprised, and the surprise is almost always that acquisition was never the problem.

A camera is not only a purchase. It is a calibration that has to be kept current.

Every sensor needs its own dark frame and its own flat-field correction, and both drift: dark current with temperature, flat-field with anything that changes the illumination or accumulates on a surface. Every sensor has its own map of hot and dead pixels, its own read noise, and its own conversion gain, so a count in channel three does not mean the same number of electrons as the same count in channel one.

For qualitative imaging that is tolerable. For anything ratiometric it is the measurement: comparing intensity between channels acquired on different sensors is comparing two instruments, and it is only valid to the extent that both have been characterized against the same standard recently.

A single camera behind a filter wheel has exactly one of each of those things to maintain, and every channel shares them. That is a genuine advantage, and it is the reason a slower instrument is sometimes the more trustworthy one.

Two constraints shape what configurations are actually buildable. Splitting a beam is naturally binary, so two, four and eight cameras are the optically simple cases; three, five, six and seven require asymmetric splitting, with arms of unequal path length and a harder alignment. And every camera images the full field in its own band, so sensor choice and camera count trade against each other directly.

That second trade is where the interesting answers live, because camera price does not scale with usefulness. A configuration of five modest sensors at moderate magnification can cost on the order of $33,000, against roughly $90,000 for three premium sensors at high magnification. The five-camera build covers about 1,160 mm² per second against roughly 600 mm² per second for the three-camera build, because it never stops to change a filter.

Cheaper and nearly twice as fast is a combination that should make you suspicious, so name what was given up: lower magnification means coarser sampling, and five arms mean five registrations to maintain rather than three. If the science needs the resolution, the expensive configuration is the correct one and the comparison is irrelevant.

The general lesson is that camera count, sensor grade and magnification are one decision rather than three, and pricing them separately is how a specification ends up both expensive and slow.