MTF, SNR & CNR
- MTF (modulation transfer function) measures sharpness: how faithfully a system reproduces fine detail as the details get smaller and closer together.
- SNR (signal-to-noise ratio) measures how loud the real signal is compared to the random graininess sitting on top of it.
- CNR (contrast-to-noise ratio) is the one that actually decides whether you can see a lesion: it's the difference between two tissues, divided by the noise.
- These three trade off against each other and against dose. You almost never get more of one for free.
Here's the uncomfortable truth about image quality: "this scan looks great" is a vibe, not a measurement. Physicists, being physicists, refused to live with a vibe, so they built three numbers to pin it down — MTF, SNR, and CNR. Learn what each one is actually asking, and suddenly half of imaging physics stops feeling like alphabet soup.
Let me walk you through all three with the same running analogy: you're trying to read a sign across a foggy parking lot.
MTF: how sharp is the system?
The modulation transfer function (MTF) is a fancy way of asking, "as the detail gets finer, how much of the contrast survives?"
Imagine a fence painted in black-and-white stripes. When the stripes are wide and lazy, your eye sees crisp black, crisp white — full contrast. Now squeeze the stripes closer and closer together. At some point the black and white start to blur into a uniform gray mush, and eventually you can't tell there were stripes at all. MTF is the curve that plots exactly how much of that black-to-white contrast your imaging system preserves at each stripe-spacing.
The "stripe-spacing" axis is spatial frequency (think line pairs per millimeter — how many black-white couples you can cram into a millimeter). At low spatial frequency (big, bold features), MTF is near 1.0: the system reproduces them faithfully. As frequency climbs, MTF droops toward zero, because every real system blurs a little. The frequency where the system finally gives up is its limiting resolution.
A single number can't capture sharpness, because sharpness depends on how fine the detail is. MTF gives you the whole curve — the full report card across every level of detail.
This is the rigorous backbone of spatial resolution: MTF is how we actually measure the thing we casually call "sharpness."
SNR: is the signal louder than the static?
Back to the foggy parking lot. Signal-to-noise ratio (SNR) asks: how bright and clear is the sign (signal) compared to the swirling fog and TV-static graininess fuzzing it up (noise)?
In imaging, "signal" is the meaningful measurement at a pixel; noise is the random speckle layered on top, much of it from the statistics of counting a finite number of X-ray photons (or echoes, or whatever the modality counts). Fewer photons means a noisier, grainier picture. SNR is simply the signal divided by that random fluctuation — a high SNR is a clean image, a low SNR is one that looks like it was photographed during a snowstorm.
The classic lever is dose: collecting more photons raises SNR. But there's no free lunch — the relationship is governed by Poisson counting statistics, so chasing a big jump in SNR costs a disproportionately large jump in dose.
More signal isn't automatically better. A blindingly bright, perfectly uniform gray rectangle has enormous SNR and tells you absolutely nothing. SNR measures cleanliness, not usefulness. For usefulness, keep reading.
CNR: can you actually see the lesion?
Here's the metric that earns its keep in the reading room. You don't diagnose by staring at a single tissue — you diagnose by spotting where one tissue differs from its neighbor. A liver lesion only exists on the image because it looks different from the liver around it.
Contrast-to-noise ratio (CNR) captures exactly that: the difference in signal between two regions (the contrast), divided by the noise. In parking-lot terms, it's not "how bright is the sign" — it's "how much does the lettering stand out from the background of the sign, given how much fog is in the way."
This is why CNR, not SNR, decides whether a subtle finding is detectable. You can have a beautifully clean, high-SNR image and still miss a lesion that barely differs from its surroundings. And you can have a grainy image where a high-contrast lesion still screams off the screen.
| Metric | Plain-English question | What raises it |
|---|---|---|
| MTF | How much fine-detail contrast survives? | Sharper system, smaller focal spot/detector element |
| SNR | How clean is the picture? | More photons (more dose), more averaging |
| CNR | Can I tell these two tissues apart? | Bigger tissue difference (e.g. contrast agent), less noise |
This is the real reason we give iodinated and gadolinium-based contrast agents: not to make the image "brighter," but to manufacture contrast between a lesion and normal tissue — driving CNR up so the finding becomes obvious.
The tug-of-war (why you can't max all three)
These metrics don't live in harmony — they fight, and dose is usually the referee.
Want razor-sharp images (high MTF)? Smaller pixels and thinner slices help — but each tiny detector element catches fewer photons, so noise climbs and SNR drops. Want to claw that SNR back? Crank the dose, or smooth the image — but smoothing is just blur, which guts your MTF. Round and round it goes.
A "smoother, cleaner-looking" reconstruction is not automatically a better one. Heavy smoothing or noise-reduction can boost the apparent SNR while quietly erasing fine detail (lower MTF) and blunting subtle lesions. Pretty is not the same as diagnostic.
The practical version of all this lives in resolution, noise, and contrast, and the numbers underneath the pixels are the Hounsfield units you window.
So if you remember one thing: MTF is sharpness, SNR is cleanliness, and CNR is conspicuity — and that last one, conspicuity, is the only metric that directly answers the question you actually care about: can I see the thing?