Mammography CNR and SDNR Quality Control

CNR and SDNR are the numbers that tell a medical physicist whether a mammography unit is still delivering the image quality it was accepted with. Contrast-to-noise ratio (CNR) and signal-difference-to-noise ratio (SDNR) collapse subject contrast, detector noise, and radiation dose into a single tracked value — and because they move in predictable ways when a detector drifts, an AEC miscalibrates, or a target/filter combination changes, they are the physicist's early-warning system long before a radiologist notices a softer image. ^{1, 7}

This article explains what CNR and SDNR actually measure, how they are computed and tracked in the ACR Digital Mammography quality control program, how they trade against mean glandular dose, and how DRPS uses them during the annual MQSA medical physicist survey to keep imaging both diagnostic and dose-efficient.

Introduction

Mammography sits at an uncomfortable intersection: it must resolve sub-millimeter microcalcifications and low-contrast masses in dense glandular tissue, yet it screens healthy women repeatedly over decades, so dose discipline is not optional. Every mammography quality control program therefore needs a metric that captures both halves of that tradeoff at once. A pure resolution number ignores noise; a pure dose number ignores whether the image is actually diagnostic. CNR and SDNR are the metrics that bind signal, noise, and — indirectly — dose together. ^{1, 5}

Under the Mammography Quality Standards Act (MQSA), every facility in the United States must undergo an annual survey by a qualified medical physicist as a condition of its FDA certification. ² The technical backbone of that survey, for the overwhelming majority of digital units, is the ACR Digital Mammography Quality Control Manual, which prescribes the tests, tools, and action limits the physicist follows. ^{1, 3} Among those tests, the quantitative image-quality check — a CNR-style measurement of signal difference over background noise — is the one that most directly answers the clinical question: is this system still producing images as good as the day it passed acceptance?

DRPS performs this analysis as part of its mammography physics and MQSA services and broader diagnostic radiography physics support across Florida, Maryland, Virginia, Washington DC, California, and Nevada.

Topic Explanation

What CNR and SDNR mean

Contrast-to-noise ratio and signal-difference-to-noise ratio are two names for the same idea: how far a feature's signal rises above the random noise of the image. Both are computed as a signal difference divided by a noise term:

$$\text{CNR}=\frac{{\bar{S}}_{\text{signal}}-{\bar{S}}_{\text{background}}}{{\sigma }_{\text{background}}}$$

Here ${\bar{S}}_{\text{signal}}$ is the mean pixel value measured inside a defined contrast object, ${\bar{S}}_{\text{background}}$ is the mean pixel value in an adjacent uniform region, and ${\sigma }_{\text{background}}$ is the standard deviation of the pixel values in that background region — the image noise. ^{1, 7}

In everyday medical-physics usage, the label SDNR is applied when the "signal" is a specific, well-defined contrast detail — for example, a thin aluminum square or foil placed on a slab of polymethyl methacrylate (PMMA) — while CNR is used more loosely for any contrast-object-over-noise measurement. ^{7, 8} The mathematics is identical; the distinction is one of convention and of how tightly the contrast object is specified. The ACR Digital Mammography QC Manual builds its physicist-level image-quality test on this signal-difference-over-noise construction. ¹

Why this single number is so powerful

The value of CNR/SDNR is that it fails in interpretable directions:

• If the detector degrades (dead pixels binned out, gain drift, a failing flat-field calibration), noise rises, so CNR falls.
• If the AEC miscalibrates and cuts the exposure, fewer photons reach the detector, noise rises, and CNR falls — even though nothing is wrong with the detector itself.
• If the wrong target/filter or kV is selected for a given thickness, subject contrast changes, and CNR shifts in a way that maps directly onto the beam quality. ⁵

Because these failure modes each move the number, a physicist who tracks CNR over time is effectively monitoring the entire imaging chain — beam, geometry, detector, and processing — with one measurement. That is why it anchors the quantitative half of the ACR program. ^{1, 3}

For context on the resolution side of the ledger and how it complements a noise-based metric, see our companion pieces on the ACR digital mammography phantom QC and on detective quantum efficiency in digital radiography.

Key Technical Principles

The quantum-limited link between CNR and dose

Mammography detectors operate in a largely quantum-limited regime: the dominant source of image noise is the statistical (Poisson) fluctuation in the number of detected x-ray photons. The standard deviation of a Poisson process scales as the square root of the count, so if $N$ photons are detected, the relative noise is $\propto 1/\sqrt{N}$, and since $N$ is proportional to detector air kerma (dose $D$):

$${\sigma }_{\text{background}}\propto \frac{1}{\sqrt{D}}⟹\text{CNR}\propto \sqrt{D}$$

This is the single most important relationship in the whole discussion. It means:

$$\frac{{\text{CNR}}_2}{{\text{CNR}}_1}=\sqrt{\frac{D_2}{D_1}}$$

A worked example. Suppose a physicist measures, on a 4 cm PMMA phantom with a small aluminum contrast detail, a mean signal of 1000, a mean background of 950, and a background standard deviation of 12.5 (in linearized pixel values). Then:

$$\text{CNR}=\frac{1000-950}{12.5}=4.0$$

If the reading radiologists and the physicist agree that a CNR of about 5 is the target for reliable low-contrast detection (the Rose criterion, discussed below), how much more dose is required? Using the square-root law:

$$\frac{D_2}{D_1}={\left(\frac{{\text{CNR}}_2}{{\text{CNR}}_1}\right)}^2={\left(\frac{5.0}{4.0}\right)}^2=1.56$$

So reaching CNR = 5.0 costs about 56% more detector dose — a concrete illustration of why CNR must never be read without also looking at mean glandular dose (MGD). Chasing a higher number is trivially easy; doing it without runaway dose is the actual engineering problem. ^{5, 6}

The Rose criterion: how much CNR is "enough"

The classic Rose model of visual detection holds that a feature must produce a signal-to-noise (or contrast-to-noise) ratio of roughly 3–5 to be reliably distinguished from noise by a human observer. ¹¹ A value near 5 is the commonly cited threshold for confident detection. This does not appear as a hard pass/fail limit in the ACR manual — the manual uses system-specific baselines — but it gives the physicist and the radiologist a physical anchor for judging whether a measured CNR is comfortably above the perceptual floor or sitting dangerously close to it.

Beam quality: the target/filter tradeoff

Subject contrast in mammography depends strongly on beam quality — the combination of anode target material, added filtration, and tube potential (kV). Lower-energy spectra (for example Mo/Mo at 25–28 kV) produce higher subject contrast but deposit more dose in thick breasts; higher-energy spectra (for example Rh/Rh at higher kV) penetrate better and lower MGD but reduce contrast, which must be recovered with more detector signal. ⁵

Young and colleagues quantified this tradeoff directly: for a 75 mm PMMA-equivalent breast, moving from 28 kV Mo/Mo to 34 kV Rh/Rh reduced subject contrast by about 25%, which required roughly 90% more detector dose to hold CNR constant — yet because the higher-energy beam is so much more penetrating, the net mean glandular dose still fell by about 32%. ⁵ This is the counterintuitive heart of mammography optimization: the higher-energy, lower-contrast beam can be the lower-dose choice once you account for the whole geometry.

Values are representative of the physics of beam-quality selection and are drawn from published optimization studies; the AEC on a modern unit makes these choices automatically, and part of the physicist's job is to confirm it makes them well. ^{5, 6}

What the AEC is really optimizing

Modern mammography automatic exposure control (AEC) systems do not simply target a fixed detector dose. The better designs target a constant CNR (or SDNR) across breast thickness, raising detector dose as needed at greater thicknesses to hold image quality flat. ^{5, 6} A fixed-detector-dose AEC, by contrast, lets CNR sag for thick or dense breasts — exactly the patients where a missed lesion is most costly. This is why CNR is not only a QC metric but the design objective of well-engineered AEC, and why measuring CNR as a function of PMMA thickness is one of the most revealing tests in the survey. For the mechanics of AEC itself, see our detailed post on the mammography AEC and phototimer QC.

Clinical Impact

Catching drift before the radiologist does

The clinical payoff of CNR/SDNR tracking is early detection of gradual degradation. A detector does not usually fail all at once; gain and offset calibrations drift, AEC sensors age, and x-ray output slowly changes. Any one of these can erode image quality by a few percent per quarter — invisible to the eye but obvious in a trended CNR chart. Because the ACR program compares each measurement to a system-specific baseline and flags deviations beyond a set percentage, a downward CNR trend triggers investigation while images are still diagnostic. ^{1, 3}

Preventing the "quiet dose creep"

The opposite failure is equally important. If a service engineer adjusts the AEC to "make the images look better," CNR may rise — but at the cost of higher MGD to every patient screened, potentially thousands per year on a busy unit. Tracking CNR alongside MGD is what distinguishes genuine improvement from silent dose inflation. A CNR that climbs while MGD climbs in lockstep is not an improvement; it is a dose problem wearing a quality-improvement costume. ^{5, 6}

Comparability across the fleet and across accreditation

For multi-unit practices, CNR/SDNR provides a common currency. Detectors from different vendors — amorphous selenium direct-conversion, scintillator-based indirect detectors, and photon-counting systems — cannot be compared by raw pixel values, but a properly normalized CNR-versus-thickness curve lets a physicist benchmark them on the same axis. ^{4, 8, 9} That comparability matters for ACR accreditation, for equitable image quality across a health system's sites, and for making informed replacement decisions. Recent detector-generation comparisons continue to report CNR and SNR as the headline image-quality metrics precisely because they travel well across hardware. ⁹

Practical Optimization Tips

1. Always pair CNR with dose

Never record or trend a CNR value without the corresponding mean glandular dose and the exposure factors (kV, mAs, target/filter, phantom thickness) that produced it. A CNR number without its dose context is uninterpretable. ^{5, 6}

2. Control the measurement geometry

Use the phantom thickness, contrast object, and region-of-interest sizes specified in your QC program, and place the ROIs consistently run to run. Chen and colleagues showed that even the size of the contrast object and ROI changes the measured signal-difference-to-noise value by up to ~25% because of how it samples the scatter distribution — so an inconsistent ROI can manufacture a "failure" that is really a measurement artifact. ⁷

3. Measure CNR across a range of thicknesses

A single-thickness CNR check confirms one operating point; a CNR-versus-PMMA-thickness curve reveals whether the AEC holds image quality across the patient population. Thick-breast underperformance is the most common and most clinically important AEC flaw. ^{5, 6}

4. Linearize before you compute

CNR must be computed on pixel values that are linear with detector air kerma (the "for-processing" or raw data), not on the log-processed "for-presentation" image. Computing noise on processed data corrupts the number. Confirm you are using the correct image pipeline for the test. ¹

5. Trend, don't spot-check

The power of the metric is in the trend line. Keep a running chart against the acceptance baseline, annotate it with service events, and treat any step change coincident with a service visit as guilty until proven innocent.

6. Separate detector noise from beam problems

If CNR drops, check whether the exposure (mAs, detector air kerma) also dropped. Falling CNR with falling dose points to the AEC or x-ray output; falling CNR at constant dose points to the detector itself. This single fork resolves most investigations quickly.

Regulatory Considerations

Mammography is the most tightly regulated imaging modality in the United States, and the CNR/SDNR test sits inside that framework rather than beside it. Under MQSA, codified at 21 CFR Part 900, a facility may not lawfully perform mammography without FDA certification, and that certification is contingent on annual accreditation-body review and an annual survey by a qualified medical physicist. ² The ACR is the dominant accreditation body, and its Digital Mammography QC Manual defines the tests — including the quantitative image-quality metric — that the physicist must perform and the technologist must support. ^{1, 3}

Key frameworks the physicist works within:

• FDA MQSA, 21 CFR Part 900 — the federal quality standard for mammography, including the requirement for an annual medical physicist survey and the equipment and personnel qualifications behind it. ²
• ACR Digital Mammography QC Manual (2018) — the current, publicly available manual defining the QC program for digital mammography and DBT, including physicist and technologist test frequencies and action limits. ¹
• ACR Mammography Accreditation Program — the accreditation pathway most facilities use to satisfy the MQSA accreditation requirement. ³
• IEC 62220-1-2:2007 — the international standard for measuring detective quantum efficiency (DQE) of mammographic detectors, the fundamental laboratory metric that underlies field CNR performance. ⁴

A jurisdictional note DRPS clients should keep straight: mammography x-ray units are FDA-regulated under MQSA and are additionally registered and inspected by the state radiation-control program (or, for radioactive material only, by the NRC or Agreement State — which does not apply to x-ray mammography). Of the states DRPS serves, Florida, Maryland, Virginia, California, and Nevada administer their own x-ray machine programs as the state authority, while the FDA-MQSA layer is uniform nationwide. The annual medical physicist survey is the point where the federal, accreditation, and state expectations converge on a single report. DRPS delivers that report through its mammography physics and MQSA service and coordinates it with broader accreditation support and medical physics consulting. ^{2, 3}

Frequently Asked Questions (FAQs)

What is CNR in mammography?

Contrast-to-noise ratio (CNR) is the difference in signal between a contrast object and its background, divided by the background noise (standard deviation). It expresses how far a feature stands out above the random pixel fluctuations, combining subject contrast and image noise into one number that scales with radiation dose.

What is the difference between CNR and SDNR in mammography?

The two are computed the same way — a signal difference divided by background noise — and are often used interchangeably. In practice, SDNR (signal-difference-to-noise ratio) is the term used when the signal is a defined contrast detail such as an aluminum square on PMMA, while CNR is the more general label. The ACR Digital Mammography QC Manual uses a CNR-style test as the physicist's quantitative image-quality metric.

How does the medical physicist measure CNR in the ACR QC program?

The physicist images a uniform phantom (typically PMMA of a standard thickness) with a small contrast object under automatic exposure control, then measures the mean pixel value inside the object, the mean pixel value in an adjacent background region, and the standard deviation of the background. CNR is the signal difference divided by that standard deviation, compared against a system-specific baseline and action limit.

Why does CNR depend on radiation dose?

Mammography detectors are largely quantum-limited, so image noise is governed by the number of detected x-ray photons. Noise falls as the square root of dose, which means CNR rises as the square root of dose. Doubling the detector air kerma increases CNR by about 1.4 times, so any CNR test is only meaningful when the exposure technique is also controlled.

What CNR value is good enough for mammography?

There is no single universal number. The ACR program establishes a system-specific baseline at acceptance and requires the physicist to flag drift beyond a defined percentage. As a detectability rule of thumb, the Rose criterion holds that a feature needs a CNR of roughly 5 to be reliably seen, but the operative test is whether the measured value stays within the action limits set for that unit.

Does a higher CNR always mean a better mammogram?

No. CNR can always be raised by adding dose, so the goal is an adequate CNR at the lowest reasonable mean glandular dose, not the highest possible CNR. Optimization means choosing the target/filter, kV, and AEC setting that reach the needed CNR for a given breast thickness while keeping dose as low as reasonably achievable.

How often is mammography CNR tested?

Under the ACR Digital Mammography QC Manual, the medical physicist evaluates the quantitative image-quality metrics at least annually during the equipment evaluation and after major service, while the technologist performs more frequent phantom and system checks. MQSA requires the annual medical physicist survey as a condition of the facility's certification.

Key Takeaways

• CNR and SDNR are the same construction — a signal difference divided by background noise — and together form the quantitative image-quality backbone of the ACR Digital Mammography QC program. ^{1, 7}
• CNR scales as the square root of dose. Doubling detector air kerma raises CNR by about 1.4×; reaching a target CNR from a lower value can cost a large dose increment, so CNR must always be read next to mean glandular dose. ⁵
• Beam quality drives the tradeoff. Higher-energy target/filter combinations lower contrast but can lower MGD substantially — for a thick breast, ~25% less contrast yet ~32% less dose. ⁵
• Well-designed AEC targets constant CNR across thickness, not constant detector dose; measuring CNR versus PMMA thickness exposes thick-breast underperformance. ^{5, 6}
• Consistent geometry is everything. Contrast-object and ROI size can swing the measured value by up to ~25%, so measurement discipline separates real drift from artifact. ⁷
• The test lives inside MQSA. The annual medical physicist survey, the ACR manual, and FDA certification are one connected system, not three. ^{1, 2, 3}

Conclusion

CNR and SDNR endure as the central mammography image-quality metrics because they answer the two questions that matter simultaneously: is the image good enough to read, and is it being produced at a defensible dose? A single number that rises with detected photons and falls with noise ties the entire imaging chain to a trend line a physicist can watch. The discipline is not in computing the ratio — that is arithmetic — but in controlling the geometry, pairing it with dose, trending it against baseline, and interpreting its direction when it moves.

Facilities that treat CNR as a live monitoring signal, rather than a box to check once a year, catch detector drift and AEC miscalibration early, hold the line on patient dose, and walk into accreditation with defensible, trended data. That is the difference between a QC program that documents problems and one that prevents them.

How DRPS Can Help

Diagnostic Radiation Physics Services performs the annual MQSA medical physicist survey and ongoing QC support that keep mammography units accredited, diagnostic, and dose-efficient. This includes CNR/SDNR measurement and trending, AEC performance evaluation across breast thickness, mean glandular dose assessment, detector and beam-quality testing, phantom image-quality review, and correction-planning when a metric drifts out of limits. DRPS delivers this through its mammography physics and MQSA service, with accreditation support and medical physics consulting as needed.

DRPS supports facilities across our service locations, including Florida, Maryland, Virginia, Washington DC, California, Nevada, New York, Pennsylvania, New Jersey, and Delaware.

The goal is simple: adequate contrast-to-noise for every breast thickness, at the lowest reasonable dose, documented well enough to defend.

Related Resources

• ACR digital mammography phantom QC
• Mammography AEC and phototimer QC
• Mean glandular dose in mammography
• Mammography quality control under MQSA
• Detective quantum efficiency in digital radiography
• Digital breast tomosynthesis QC
• Mammography physics and MQSA services
• Medical physicist consulting

References

1. American College of Radiology. ACR Digital Mammography Quality Control Manual. 2018. acr.org
2. U.S. Food and Drug Administration. Mammography Quality Standards Act (MQSA); 21 CFR Part 900. ecfr.gov
3. American College of Radiology. Mammography Accreditation Program Requirements. acr.org
4. International Electrotechnical Commission. IEC 62220-1-2:2007 — Medical electrical equipment: Characteristics of digital X-ray imaging devices — Part 1-2: Determination of the detective quantum efficiency — Detectors used in mammography. iec.ch
5. Young KC, Oduko JM, Bosmans H, Nijs K, Martinez L. Optimal beam quality selection in digital mammography. Br J Radiol. 2006;79(948):981-990. doi:10.1259/bjr/55334425. PubMed
6. Jakubiak RR, Gamba HR, Neves EB, Peixoto JE. Image quality, threshold contrast and mean glandular dose in CR mammography. Phys Med Biol. 2013;58(18):6565-6583. doi:10.1088/0031-9155/58/18/6565. PubMed
7. Chen H, Danielsson M, Xu C, Cederström B. On image quality metrics and the usefulness of grids in digital mammography. J Med Imaging (Bellingham). 2015;2(1):013501. doi:10.1117/1.JMI.2.1.013501. PubMed
8. Maldera A, De Marco P, Colombo PE, Origgi D, Torresin A. Digital breast tomosynthesis: Dose and image quality assessment. Phys Med. 2017;33:56-67. doi:10.1016/j.ejmp.2016.12.004. PubMed
9. Pautasso JJ, Van Speybroeck CDE, Michielsen K, Sechopoulos I. Comparative image quality and dosimetric performance of two generations of dedicated breast CT systems. Med Phys. 2025;52(4):2191-2200. doi:10.1002/mp.17623. PubMed
10. American College of Radiology / American Association of Physicists in Medicine. ACR–AAPM Technical Standard for Diagnostic Medical Physics Performance Monitoring of Mammographic Equipment. acr.org
11. Rose A. The sensitivity performance of the human eye on an absolute scale. J Opt Soc Am. 1948;38(2):196-208. doi:10.1364/JOSA.38.000196. doi.org