CT Image Quality QC: MTF, NPS, and Detectability

CT image quality cannot be reduced to a single number. A defensible image quality program measures spatial resolution, noise magnitude, noise texture, and low-contrast detectability, then compares them against baselines and accreditation criteria. On scanners that use iterative or deep-learning reconstruction, the contrast-to-noise ratio alone can mislead, so modern testing increasingly relies on the modulation transfer function (MTF), the noise power spectrum (NPS), and a task-based detectability index.¹²³

Computed tomography image quality QC verifies that anatomy is reproduced accurately, that small low-contrast structures remain visible, and that performance has not drifted since the last test. Because reconstruction algorithms now reshape both resolution and noise, the most useful metrics describe the imaging chain as a system rather than as isolated features. This guide explains the core metrics, how they are measured on the ACR CT accreditation phantom, what they mean clinically, and how DRPS supports facilities through CT physics testing and accreditation support across Florida, Maryland, Virginia, Washington DC, California, and beyond.

Introduction

Image quality in CT is the combination of how well the scanner resolves fine detail, how much noise it produces, what that noise looks like, and how reliably it lets a reader detect a real lesion. Each of these properties can change independently with kVp, mAs, reconstruction kernel, slice thickness, and reconstruction algorithm. A scanner can have excellent high-contrast resolution and still fail at detecting a subtle low-contrast lesion if the noise texture is unfavorable.¹²

Historically, CT image quality was summarized by a handful of measurements: CT number accuracy, uniformity, noise (the standard deviation in a water phantom), high-contrast resolution from a line-pair pattern, and a contrast-to-noise ratio for low-contrast objects. These remain part of every accreditation submission and annual survey. But filtered back-projection (FBP), the algorithm those metrics were built around, is increasingly replaced by iterative reconstruction (IR) and deep-learning image reconstruction (DLIR). Those methods are nonlinear: resolution depends on object contrast and dose, and noise is no longer simply "more or less" but has a texture that can help or hurt detection.²³⁴

This is why the American Association of Physicists in Medicine published Report No. 233, which formalizes task-based performance evaluation of CT systems using the task transfer function (TTF), the noise power spectrum (NPS), and a detectability index. The same report's methods underlie modern image quality software and pave the way toward image quality registries and standardized clinical benchmarks.¹

Topic Explanation

The four properties that define CT image quality

A complete CT image quality assessment characterizes four interacting properties:

• Spatial resolution — the ability to reproduce sharp edges and small high-contrast structures, described by the MTF or, for nonlinear systems, the TTF.
• Noise magnitude — the random fluctuation in CT numbers, typically the standard deviation of HU in a uniform region.
• Noise texture — how that noise is distributed across spatial frequencies, described by the NPS.
• Low-contrast detectability — the ability to distinguish a structure only slightly different in attenuation from its surroundings, summarized by the CNR or, more completely, a detectability index.

The first three combine to determine the fourth. Two protocols can deliver the same noise standard deviation yet very different detectability because their resolution and noise texture differ.²³

For broader context on how acquisition choices feed into these properties, see our companion guides on CT protocol optimization and reconstruction kernels, and on the dose side, CTDIvol and DLP dose metrics.

Why a single resolution or noise number is no longer sufficient

On a linear FBP system, resolution is independent of contrast and dose, so one MTF curve describes the scanner. Iterative and deep-learning reconstructions break that assumption. Their resolution falls as object contrast or dose decreases, and their noise becomes correlated and "blotchy." A line-pair phantom measured at high contrast can therefore overstate the resolution available for a low-contrast liver lesion. Likewise, the noise standard deviation can drop while the texture shifts in a way that actually reduces lesion conspicuity.²³⁴

The practical consequence is that image quality testing must measure resolution and noise as a system under conditions that resemble the clinical task. That is the central idea of task-based assessment.

Key Technical Principles

Spatial resolution: MTF and TTF

The modulation transfer function describes the fraction of an object's contrast at each spatial frequency that survives the imaging process. Formally, it is the normalized magnitude of the optical transfer function:

$$\mathrm{MTF}(f)=\frac{\left|\mathrm{OTF}(f)\right|}{\left|\mathrm{OTF}(0)\right|}$$

An MTF of 1.0 means full contrast transfer; the frequency at which the MTF falls to 50% (denoted $f_{50}$) and to 10% ($f_{10}$) are common summary points. A scanner with $f_{50}=0.45$ cycles/mm and $f_{10}=0.75$ cycles/mm has better fine-detail performance than one with $f_{50}=0.35$ cycles/mm.

For nonlinear reconstructions, the task transfer function is measured instead, usually from the edge of a contrast insert (a circular-edge or "edge-spread" technique). The TTF is the contrast- and dose-specific analog of the MTF and is the descriptor recommended for IR and DLIR systems.¹²

Noise magnitude and the noise power spectrum

Noise magnitude is the standard deviation $\sigma$ of CT numbers in a uniform region. The NPS goes further and describes how the noise variance is distributed over spatial frequency. For a two-dimensional region, a working definition is:

$$\mathrm{NPS}(f_x,f_y)=\frac{\Delta x\Delta y}{N_x{N}_y}⟨{\left|\mathrm{DFT}\left[I(x,y)-\bar{I}\right]\right|}^2⟩$$

where $\Delta x,\Delta y$ are pixel dimensions, $N_x,N_y$ the region size in pixels, $I(x,y)$ the image, $\bar{I}$ its mean, and $⟨\cdot ⟩$ an ensemble average over many noise realizations. The integral of the NPS over all frequencies equals the noise variance, ${\sigma }^2$. The shape of the NPS — where its peak sits — is the noise texture. FBP produces a mid-frequency peak, while iterative reconstruction shifts the peak toward lower frequencies, giving the characteristic smooth-but-blotchy look.¹³⁴

Low-contrast detectability: from CNR to the detectability index

The contrast-to-noise ratio is the simplest detectability surrogate:

$$\mathrm{CNR}=\frac{S_{\text{object}}-S_{\text{background}}}{{\sigma }_{\text{background}}}$$

where $S$ values are mean CT numbers. The ACR CT accreditation program applies a CNR-based criterion to the low-contrast module for the adult abdomen protocol.⁵⁶

A more complete descriptor is the task-based detectability index $d^{\prime }$, which combines the task (the size and contrast of what you are trying to see), the TTF, and the NPS. A non-prewhitening model observer gives:

$$d^{\prime 2}=\frac{{\left[\int W_{\text{task}}(f)\mathrm{TTF}(f)df\right]}^2}{\int W_{\text{task}}^2(f){\mathrm{TTF}}^2(f)\mathrm{NPS}(f)df}$$

where $W_{\text{task}}(f)$ is the task function describing the object to be detected. A higher $d^{\prime }$ means a more detectable lesion. This single number honestly accounts for resolution, noise magnitude, and noise texture at the same time, which is why it tracks human-observer performance better than CNR for nonlinear reconstructions.¹²³

Comparison of the core image quality metrics

The numeric values appropriate for any given scanner depend on the make, model, protocol, and reconstruction; baselines should be established at acceptance and tracked over time rather than copied from another site.¹⁵

A worked low-contrast example

Suppose an adult abdomen scan of the ACR phantom low-contrast module gives a mean of 6 HU inside the low-contrast cylinders relative to the surrounding background, and the background noise is $\sigma =5$ HU. Then:

$$\mathrm{CNR}=\frac{6\mathrm{HU}}{5\mathrm{HU}}=1.2$$

This exceeds the ACR adult-abdomen CNR criterion of 1.0, so the protocol passes on this metric.⁵⁶ Now suppose a deep-learning reconstruction lowers the noise to $\sigma =4$ HU; CNR rises to 1.5. But if the same reconstruction also reduces the TTF at the relevant frequencies and shifts the NPS toward lower frequencies, the task-based $d^{\prime }$ may improve less than the CNR suggests — or, for very small objects, hardly at all. This is precisely the gap between CNR and detectability that task-based metrics are designed to close.²³⁴

Clinical Impact

Detecting subtle pathology

Low-contrast detectability is the property most directly tied to diagnostic confidence for tasks such as identifying a hypodense liver metastasis, a small hypervascular hepatocellular carcinoma, or subtle gray–white differentiation in the brain. A protocol that passes high-contrast resolution can still fail clinically if its low-contrast performance is poor at the dose used. Phantom studies repeatedly show that reconstruction choice changes detectability of simulated lesions even when contrast is held constant.²³

Reconstruction algorithm selection

According to PubMed, a task-based comparison of model-based iterative reconstruction (MBIR) against adaptive statistical iterative reconstruction and FBP on the ACR model 464 phantom found MBIR improved the detectability index and indicated substantial dose-reduction potential for matched detectability (Samei & Richard, 2015, DOI).¹ More recent work shows deep-learning reconstruction can preserve or improve low-contrast detectability relative to iterative reconstruction at routine dose, again measured with model observers rather than CNR alone (Fan et al., 2023, DOI; Pauthe et al., 2025, DOI).²³ These findings only emerge with task-based testing; CNR can hide or exaggerate them.

Consistency across the fleet

For multi-scanner practices, image quality testing also enforces consistency. A radiologist reading from several scanners benefits when noise texture and resolution are matched, so a finding looks the same regardless of which scanner produced it. Tracking MTF/TTF, NPS, and detectability over time catches drift from tube aging, detector changes, or software updates before it affects reads. The ACR phantom can even anchor a routine QA program across CT and cone-beam CT systems when baselines are set carefully (Hobson et al., 2014, DOI).⁷

Practical Optimization Tips

A practical CT image quality program follows a consistent workflow.

1. Establish baselines at acceptance

Measure CT number accuracy, uniformity, noise, MTF/TTF, NPS, slice thickness, and low-contrast performance on each scanner and clinically used reconstruction at acceptance testing. These baselines, not generic values, define "normal" for that scanner.

2. Test under clinically relevant conditions

Measure resolution and noise at the dose levels, kernels, and reconstruction settings actually used clinically. For IR and DLIR, evaluate TTF from a contrast edge and report it with the contrast and dose used, because the result is condition-specific.¹²

3. Use noise texture, not just noise magnitude

Track the NPS shape, not only $\sigma$. A reconstruction that lowers $\sigma$ but pushes the NPS peak to low frequencies may degrade the perception of small structures even though the noise "number" improved.³⁴

4. Prefer task-based detectability for algorithm decisions

When comparing reconstructions or planning dose reduction, use a detectability index that combines TTF and NPS for a defined task. CNR is acceptable for routine accreditation but is unreliable for ranking nonlinear algorithms.¹²³

5. Separate technologist QC from physicist testing

Technologist QC (for example, daily water CT number and noise) catches gross failures quickly. Comprehensive testing — full MTF/TTF, NPS, detectability, slice thickness, and accuracy — belongs to the annual physicist survey and to acceptance testing.

Common pitfalls to avoid

• Comparing scanners on a single number. Resolution and noise are not interchangeable; report them together.
• Trusting CNR for IR/DLIR. Use task-based metrics when reconstructions are nonlinear.
• Measuring TTF at high contrast only. Report the contrast and dose; the result does not generalize.
• Ignoring noise texture. A lower standard deviation does not guarantee better detectability.
• Copying another site's baselines. Baselines are scanner- and protocol-specific.
• Skipping post-update re-testing. Reconstruction software updates can change resolution and noise texture and should trigger re-verification.

Regulatory Considerations

CT image quality testing in the United States is governed by a combination of accreditation requirements, equipment performance standards, and state regulations, with a qualified medical physicist performing the comprehensive evaluation. Facilities seeking Medicare reimbursement for CT generally must be accredited, and the ACR CT Accreditation Program specifies phantom-based image quality submissions plus an annual physicist survey.⁵

Key frameworks to reference:

• ACR CT Accreditation Program — defines phantom image quality criteria (including the low-contrast CNR criterion), dose limits, and the requirement for an annual medical physicist evaluation.⁵⁶
• AAPM Report No. 233 (TG-233) — methodology for task-based performance evaluation using TTF, NPS, and detectability, intended for clinical medical physicists.¹
• ICRU Report 87 — international guidance on radiation dose and image-quality assessment in CT, including resolution and noise descriptors.⁸
• IEC 61223-3-5 — acceptance test methods for the imaging performance of CT X-ray equipment.⁹
• 21 CFR 1020.33 — the U.S. federal performance standard for CT equipment (FDA), including image quality and dose information reporting.¹⁰

CT X-ray systems are regulated by the FDA and by state radiation-control programs rather than by the NRC, which governs radioactive material. Of the states DRPS serves, Florida, Maryland, Virginia, California, Nevada, Pennsylvania, New York, and New Jersey administer their own X-ray inspection and registration programs, while Washington, DC and Delaware also regulate X-ray machines at the state/district level. A facility should confirm its state's specific testing frequency and reporting requirements; for the accreditation perspective, see ACR accreditation physics requirements.

Frequently Asked Questions (FAQs)

What is CT image quality QC?

CT image quality QC is the set of measurements that verify a scanner reproduces anatomy accurately and consistently. It typically includes CT number accuracy, image uniformity, spatial resolution (MTF or task transfer function), image noise and noise texture (noise power spectrum), slice thickness, and low-contrast detectability, evaluated against established baselines and accreditation criteria.

What is the difference between MTF and the task transfer function (TTF)?

The modulation transfer function (MTF) describes how faithfully a system transfers spatial frequencies for a linear, contrast-independent system. The task transfer function (TTF) is the contrast- and dose-dependent resolution measured on nonlinear reconstructions such as iterative or deep-learning reconstruction, usually from a contrast edge in a phantom. On modern scanners, TTF is the more realistic descriptor of clinical resolution.

Why is contrast-to-noise ratio (CNR) no longer enough on its own?

CNR was designed around linear, filtered back-projection images. Iterative and deep-learning reconstructions change noise texture and make resolution depend on contrast and dose, so two images with the same CNR can have very different detectability. Task-based metrics that combine resolution (TTF), noise magnitude, and noise texture (NPS) into a detectability index give a more reliable comparison.

What is the noise power spectrum (NPS) and why does texture matter?

The noise power spectrum describes how image noise is distributed across spatial frequencies, capturing not just how much noise there is but its texture or graininess. Iterative reconstruction often shifts the NPS toward lower frequencies, producing a blotchy appearance that can hide small low-contrast lesions even when the total noise (standard deviation) is reduced.

What does the ACR CT accreditation phantom measure for image quality?

The ACR CT accreditation phantom (model 464) contains four modules that test CT number accuracy, high-contrast (spatial) resolution, low-contrast resolution, and image uniformity, plus in-plane distance accuracy and slice thickness. The low-contrast module uses cylinders at about 6 HU contrast, and the program applies a contrast-to-noise ratio criterion for the adult abdomen protocol.

How often should CT image quality be tested?

Technologists perform frequent (often daily) quality control such as water CT number and noise checks, while a qualified medical physicist performs comprehensive annual performance testing and acceptance testing for new equipment. Accreditation programs and many state regulations also require periodic image quality submissions and an annual physicist survey.

Can image quality testing reduce patient dose?

Yes. By quantifying the true detectability achievable at a given dose, task-based testing lets physicists identify protocols where dose can be lowered without losing diagnostic performance, or where image quality is inadequate and needs correction. Several studies show advanced reconstruction can preserve detectability at substantially lower dose when verified objectively.

Key Takeaways

• Image quality is multidimensional. Resolution (MTF/TTF), noise magnitude, noise texture (NPS), and low-contrast detectability must be measured together.
• Nonlinear reconstruction breaks old assumptions. IR and DLIR make resolution depend on contrast and dose and reshape noise texture, so a single MTF or noise number no longer describes the scanner.
• CNR is necessary but not sufficient. It remains an accreditation metric, but task-based detectability ($d^{\prime }$) is the better tool for comparing reconstructions and planning dose reduction.
• Baselines are scanner-specific. Establish them at acceptance and track them; do not copy values between sites.
• Re-test after software updates. Reconstruction changes can alter resolution and noise texture and must be re-verified.
• Testing supports dose reduction. Objective image quality measurement is what lets a facility safely lower dose while protecting diagnostic performance.

Conclusion

CT image quality QC has evolved from a short checklist of single-number metrics into a task-based discipline. The reason is straightforward: the reconstruction algorithms now in routine clinical use are nonlinear, so resolution and noise depend on the imaging task itself. Measuring the MTF or task transfer function, the noise power spectrum, and a detectability index — alongside the established CT number, uniformity, slice thickness, and CNR checks — gives a far more honest picture of how a scanner will perform for real patients.

For facilities, the payoff is concrete: more reliable detection of subtle pathology, consistent appearance across a scanner fleet, defensible accreditation submissions, and the confidence to lower dose where the data support it. A qualified medical physicist translates these measurements into protocol decisions and a documented quality program that holds up to inspection and accreditation review.

How DRPS Can Help

Diagnostic Radiation Physics Services performs acceptance testing, annual physics surveys, and image quality optimization for CT systems from all major manufacturers. Our CT physics testing includes MTF/TTF, NPS, low-contrast detectability, CT number accuracy, uniformity, and slice thickness, with baselines tracked over time. We also provide accreditation support for the ACR CT program and broader medical physics consulting for protocol review and dose optimization.

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

A strong image quality program is not just about passing accreditation — it is about making sure every scan gives the radiologist the detail they need at the lowest reasonable dose.

Related Resources

• CT protocol optimization
• CT number (HU) calibration QC
• Reconstruction kernels and image quality
• CTDIvol and DLP dose metrics
• ACR accreditation physics requirements
• CT physics testing
• Accreditation support

References

1. American Association of Physicists in Medicine. AAPM Report No. 233: Performance Evaluation of Computed Tomography Systems. 2019. aapm.org
2. Samei E, Richard S. Assessment of the dose reduction potential of a model-based iterative reconstruction algorithm using a task-based performance metrology. Med Phys. 2015;42(1):314-323. doi:10.1118/1.4903899. PubMed
3. Fan M, Thayib T, McCollough C, Yu L. Accurate and efficient measurement of channelized Hotelling observer-based low-contrast detectability on the ACR CT accreditation phantom. Med Phys. 2023;50(2):737-749. doi:10.1002/mp.16068. PubMed
4. Kim B, Han M, Shim H, Baek J. A performance comparison of convolutional neural network-based image denoising methods: The effect of loss functions on low-dose CT images. Med Phys. 2019;46(9):3906-3923. doi:10.1002/mp.13713. PubMed
5. American College of Radiology. CT Accreditation Program Requirements. acr.org
6. Pauthe A, Milliner M, Pasquier H, et al. Impact of deep learning reconstructions on image quality and liver lesion detectability in dual-energy CT: An anthropomorphic phantom study. Med Phys. 2025;52(4):2257-2268. doi:10.1002/mp.17651. PubMed
7. Hobson MA, Soisson ET, Davis SD, Parker W. Using the ACR CT accreditation phantom for routine image quality assurance on both CT and CBCT imaging systems in a radiotherapy environment. J Appl Clin Med Phys. 2014;15(4):4835. doi:10.1120/jacmp.v15i4.4835. PubMed
8. International Commission on Radiation Units and Measurements. ICRU Report 87: Radiation Dose and Image-Quality Assessment in Computed Tomography. 2012. icru.org
9. International Electrotechnical Commission. IEC 61223-3-5: Evaluation and routine testing in medical imaging departments — Acceptance and constancy tests — Imaging performance of computed tomography X-ray equipment. iec.ch
10. U.S. Food and Drug Administration. 21 CFR 1020.33: Computed tomography (CT) equipment. ecfr.gov