Fluorescence imaging with a CMOS camera: the sensor numbers

The question comes up on every microscopy bench: will the mono machine-vision camera already in the drawer capture fluorescence, or does the job need a scientific camera? It can, if three sensor numbers clear the bar at your dye's emission wavelength: quantum efficiency, read noise and usable exposure. This guide gives those bars as numbers, works a photon budget end to end, and shows where a colour sensor quietly discards the signal you paid to generate. Filtering the light is the companion guide on excitation and emission filters; this page is the sensor side — what the camera must do once the filtered emission reaches it.

Can a machine-vision CMOS camera capture fluorescence at all?

Yes, for most routine assays: quantum efficiency of 50% or more at the emission wavelength, read noise of 2–4 electrons and exposures of tens to hundreds of milliseconds give a mono CMOS camera a usable signal-to-noise ratio on all but the dimmest stains.

Fluorescence is a photon-starvation problem: a few hundred to a few thousand photons land on each pixel per exposure, and the sensor converts them one at a time. Quantum efficiency sets how many photons become electrons; read noise sets the floor under the count; exposure sets how long the well fills. Modern machine-vision CMOS holds read noise down near 2 electrons and typically 50–70% QE through the visible, which clears the first two bars outright. That is why most mono area scan cameras image routine immunofluorescence and fluorescent-protein stains without any scientific-grade hardware. What remains is the photon budget, worked out below — the one number that moves by orders of magnitude between stains.

Why does a monochrome sensor beat colour at a single emission band?

A monochrome sensor returns roughly three to four times the usable signal of a colour sensor at one emission band, because the Bayer mosaic sends every pixel through a red, green or blue filter and only the matching channel passes the dye's light.

Walk the mosaic. Each 2 × 2 block on a colour sensor carries one blue-filtered pixel, one red-filtered pixel and two green-filtered pixels. At a fluorescein emission near 520 nm the red and blue sites receive almost nothing, the green sites pass only part of the band through their filter, and the camera then interpolates across a half-empty grid, softening real structure. Mono deletes the mosaic, the interpolation and the colour correction in one step, which is why fluorescence benches run mono as standard. If a colour camera is your only option, capture the raw Bayer stream rather than an already-debayered one so the debayering happens once, under your control, and average frames on samples that hold still.

How much quantum efficiency do you need at the emission wavelength?

Aim for 50% or more at your dye's emission peak; modern CMOS sensors peak near 70% around 600 nm, drop to about 30% in the UV and fall to roughly 10% by 1000 nm.

Read the QE curve at the emission wavelength — the light that reaches the sensor — never at the excitation wavelength. Most common stains emit between roughly 375 and 550 nm, squarely in CMOS's strongest region. UV emitters give up about half their potential signal at 30% QE, and beyond 800 nm the curve fades toward 10% at 1000 nm, so near-infrared dyes need proportionally longer exposures. A working rule: below about 30% QE at the emission peak, divide the photon rate by the QE and budget exposure from that number. Narrow-band excitation also keeps the emission bundle tight — the same logic that makes one wavelength win on the illumination side.

How long an exposure does a dim stain need?

Exposure must collect enough electrons for your target SNR: with 3 electrons of read noise, a 20:1 SNR needs about 400 electrons per pixel, which at 60% QE and a sensor flux of 20,000 photons per pixel per second is a 33 ms exposure.

S is signal electrons per pixel, N_photons the photons collected per pixel, R the photon rate per pixel, and t the exposure in seconds. Work the budget backwards from the target. A 20:1 SNR needs about 400 electrons, because in the shot-noise-limited regime SNR ≈ √S, so S = 20² = 400. Check the read-noise term at n_read = 3 electrons: total noise = √(400 + 9) = 20.2, giving SNR = 400 / 20.2 = 19.8 — the target still holds. At 60% QE, typical for a visible emission near 520 nm, those electrons take 400 / 0.60 = 667 photons per pixel. The last input belongs to your optics: an illustrative mid-bright immunofluorescence scene might deliver 20,000 photons per pixel per second at the sensor, so t = 667 / 20,000 = 0.033 s ≈ 33 ms. Halve the flux for a dim stain and t = 667 / 2,000 = 333 ms, where a dark rate of about 2 electrons per second adds only 0.7 electrons of dark charge — still beneath the read-noise floor. That photon rate is the illustrative input: dye concentration, objective NA and filter band set it, and it moves by orders of magnitude, not by percent.

Collect photons before touching gain. Gain multiplies signal and read noise alike, so it cannot rescue a starved exposure — the exposure-before-gain discipline applies on a fluorescence bench exactly as on a brightfield one, and raw exposure and gain units differ between vendors, so log the values with each capture.

Do you need a cooled camera?

For exposures up to about one second, no: dark charge on an uncooled modern CMOS sensor runs a few electrons per second at room temperature and halves with every 5–7 °C drop, so it only starts to matter when exposures run for seconds or frames are summed.

A running camera body also sits a few degrees above room temperature, which pushes dark current up — the exposure length decides, not the badge. Below about one second, dark charge stays well under the 2–4 electron read-noise floor; the 333 ms case above added 0.7 electrons. From several seconds upward, or when you stack many long frames, cooling earns its place. Otherwise sum or average shorter frames, which recovers the same SNR on a fixed-frame-rate camera without cooling hardware. Decision rule: uncooled below 1 s; consider a cooled head once exposures run for many seconds or the sample cannot take more excitation light without bleaching.

What field of view does the relay give your sensor?

The objective's magnification divides the sensor format: a 1-inch sensor's active area of about 12.8 × 9.6 mm shrinks to roughly 1.28 × 0.96 mm of sample behind a 10× objective, and a 5 µm structure needs only 512 pixels across that field.

Check pixels against the field: 1,280 µm ÷ 5 µm = 256 pixels minimum, doubled to 512 under the two-pixels-per-feature rule — far below any 6 MP sensor, so pixel count is rarely the binding constraint; photons are. Sensor pixels run roughly 2–4.5 µm in this range: 4,000 pixels across 12.8 mm is 3.2 µm, or 0.32 µm at the sample — well under the ~1.1 µm diffraction spot of a 0.3 NA objective at 520 nm (0.61 × 0.52 / 0.3). The optics limit resolution; the sensor's job is to catch what they pass, and bigger pixels catch more per pixel: a 4.5 µm pixel collects (4.5 / 2.0)² ≈ 5 times the photons of a 2 µm pixel at the same image illuminance. Size your own field and feature with the calculator below.

For wide fields behind low-magnification objectives, choose the largest sensor format your relay supports:

The MV-CS200-10UM pairs that 1-inch mono sensor with USB3 in a C-mount body.

Which camera fits which fluorescence job?

Match the camera to the exposure budget, not the pixel count: long dark exposures favour mono sensors with larger pixels, while fast kinetics need the USB3 or GigE models that hold their full frame rate at long exposure control ranges.

Static, dim stains are the common case: expose long, sum frames, keep gain low. Any mono part clearing the QE and read-noise bars works; the 6 MP GigE option on a 1/1.8-inch sensor covers it at a gentle data rate, and its rolling shutter is harmless on samples that hold still.

Fine structure with cropping latitude wants more pixels at full speed: the LS-CMR-0166 delivers 4000 × 3000 pixels at 30 fps over USB3, so you can crop the field and keep the sampling you need.

Low-magnification wide fields take the 1-inch MV-CS200-10UM from the section above. All three are rolling-shutter cameras, which costs nothing on fixed cells and slowly drifting tissue; the FAQ covers the moving cases.

What do you check before ordering?

Check three things against your dye before you order: the QE curve value at the emission wavelength, the exposure range the camera allows at your frame rate, and mono output; then confirm the C-mount relay matches the sensor format so the field covers the sample.

• QE at the emission peak — read the curve at the dye's emission, not its excitation; 50% or more for visible dyes.
• Mono output — colour costs you about two thirds of the band.
• Exposure range and manual lock — the exposure-time control lives in the camera's set-up software; confirm it runs to the hundreds of milliseconds your budget needs.
• Sensor format vs relay image circle — a 1-inch sensor behind a relay sized for 2/3-inch formats vignettes at the corners.
• Shutter type — rolling is fine for static samples, global for fast motion.
• Trigger input — needed if you strobe the excitation source per frame.

Filter the camera finder to mono and your sensor format, or start at the camera step of the configurator. If the light side of the chain is still open, the lighting topic gathers the guides, and the contact page is open when you want a second opinion on a photon budget.