Frame Rate

Maximum frame rate is quoted under the conditions that maximize it: full sensor, lowest bit depth, fastest readout mode. A sensor advertised at 498 frames per second reaches that in an 8-bit speed mode. Switch to the 16-bit mode you would actually use for quantitative work and the same sensor may deliver a small fraction of it.

Region of interest is the other lever. Most sensors read faster over fewer rows, so cropping to a quarter of the height can multiply the rate severalfold — which is genuinely useful, and is also how the headline number is often obtained. Read the conditions beside the figure; they are always printed, and almost never the conditions of your experiment.

Frame interval is exposure plus readout, and for most fluorescence work exposure dominates so completely that the sensor's rate never enters. At a 200 ms exposure you get at most 5 frames per second, and a camera capable of 100 or 500 makes no difference whatsoever — it spends the entire time waiting for you.

Frame rate becomes the binding constraint only when exposures are short: fast calcium transients, high-speed particle tracking, anything where you are already photon-rich enough to expose in milliseconds. Before paying for speed, work out the exposure your signal actually requires. If it is 100 ms or longer, the specification is decorative.

Most sCMOS sensors expose row by row rather than all at once. Each row's exposure window is offset slightly from its neighbour's, so a frame is not a single instant but a sweep across the sensor, typically taking about 10 ms end to end.

For a static specimen this is invisible. For anything moving quickly relative to that sweep it is a geometric distortion: an object crossing the field is recorded at different positions by different rows, so a round object images as a slanted one. It also matters for synchronized illumination, since rows are not all sensitive at the same moment — which is why light-sheet systems use a rolling light sheet matched to the shutter.

Global shutter modes expose every row simultaneously and remove the artifact, at a cost in read noise or in frame rate. Worth choosing deliberately when motion is part of the measurement.