9.1 Image Quality Metrics — Resolution, Noise & Uniformity
Key Takeaways
- Spatial resolution (measured in line pairs per centimeter) and contrast resolution trade off against each other — a sharp bone kernel maximizes edge detail but increases noise and reduces low-contrast detectability.
- Noise is proportional to 1/√(mAs): quadrupling mAs is required to cut noise in half, not simply doubling it.
- Temporal resolution is the time needed to acquire one image's data; half-scan reconstruction and dual-source CT both shorten it for cardiac and other moving-structure imaging.
- Uniformity means a homogeneous phantom must read the same CT number at the center and the periphery; a uniformity failure on daily QC is a systemic problem, not an isolated bad image.
- The ARRT content specifications name all four metrics explicitly under Image Quality (2.B.1-2.B.4) inside the 22-question Image Evaluation and Archiving subcategory.
Why This Topic Matters
Image Quality is its own lettered item (2.B) inside the Image Evaluation and Archiving subcategory, which carries 22 of the 165 scored questions in the 52-question, 31.5%-weighted Image Production domain. Nine specific sub-items live under Image Quality alone, and four of them — spatial resolution, contrast resolution, temporal resolution, and noise and uniformity — are the physical properties that define whether a CT image is diagnostically useful. Every parameter-selection question on the exam (slice thickness, kernel, mAs, gantry speed) is secretly a question about which of these four properties you are trading away to gain another. A technologist who can only recite definitions, without knowing the trade-offs, will miss the scenario-based questions that dominate this content area.
Spatial Resolution
Spatial resolution is the system's ability to depict two small, closely spaced, high-contrast objects as clearly separate rather than blurred together. It is measured in line pairs per centimeter (lp/cm); clinical CT typically achieves roughly 15-20 lp/cm, far below plain radiography's >100 lp/cm, because CT trades some spatial sharpness for its superior ability to show subtle density differences (see contrast resolution below). Spatial resolution is controlled by:
- Detector element size — smaller elements sample finer detail.
- Focal spot size — a smaller focal spot reduces geometric blur (penumbra).
- Matrix size and pixel size — a 512×512 matrix over the same display field of view (DFOV) produces smaller pixels and finer in-plane detail than a coarser matrix.
- Reconstruction kernel — a sharp/bone kernel maximizes edge detail; a soft/smooth kernel blurs edges.
- Slice thickness — thinner slices improve resolution along the z-axis (longitudinal/through-plane resolution) and reduce partial volume averaging.
Contrast Resolution
Contrast resolution (also called low-contrast resolution) is the ability to distinguish between two tissues that differ only slightly in physical density or attenuation — for example, a liver metastasis only 5-10 Hounsfield units (HU) denser than surrounding liver parenchyma. This is CT's defining diagnostic strength over conventional radiography: cross-sectional acquisition eliminates the superimposition of overlying structures, and CT detectors have a wide dynamic range that plain film cannot match. Contrast resolution is quantified with low-contrast detectability phantoms, which measure the smallest low-contrast object visible at a given dose. It improves with higher mAs (more photons, less noise), thicker slices (more photons collected per voxel), a smooth/soft-tissue kernel, and modern iterative reconstruction, which reduces noise without a proportional dose increase.
The classic exam trap is the inverse relationship between spatial and contrast resolution: a sharp bone kernel that maximizes spatial resolution and edge detail simultaneously increases image noise and reduces low-contrast detectability. There is no parameter choice that improves both simultaneously — every selection is a trade-off, and the exam expects you to identify which one a given protocol is prioritizing.
Temporal Resolution
Temporal resolution is the time required to acquire the data needed to reconstruct a single image — effectively CT's "shutter speed." It matters most for continuously moving structures, and cardiac CT is the textbook application. Temporal resolution is set by:
- Gantry rotation time — modern scanners rotate in roughly 0.25-0.35 seconds per 360°.
- Half-scan (partial-scan) reconstruction — using only about 180° plus the fan angle of projection data, rather than a full 360°, roughly halves the effective temporal resolution compared to a full rotation.
- Dual-source CT — two x-ray tube/detector pairs offset by about 90° collect two data arcs simultaneously, pushing temporal resolution down to roughly a quarter of the gantry rotation time (as low as ~66-75 ms on some systems).
- Multisegment (multisector) reconstruction — combines partial data collected across two or more consecutive heartbeats to further improve effective temporal resolution, but requires a stable, regular heart rate to work correctly.
Noise and Uniformity
Noise is the random, unwanted fluctuation of CT numbers within an otherwise homogeneous object, seen visually as graininess or mottle. It is quantified as the standard deviation of CT numbers measured in a region of interest (ROI) placed in a uniform water phantom. The root physical cause is quantum mottle — statistical fluctuation in the number of photons reaching the detector, since fewer photons produce a proportionally larger relative fluctuation.
The key mathematical relationship, and a favorite exam calculation: noise is proportional to 1/√(mAs). Doubling mAs from 100 to 200 reduces noise by only about 29% (1/√2), not 50%. To cut noise in half, mAs must be quadrupled (for example, 100 mAs to 400 mAs). Noise is also affected by kVp (higher kVp penetrates more, reducing noise but also reducing contrast), slice thickness (thinner slices sample fewer photons per voxel and are noisier), reconstruction kernel (sharp kernels amplify noise), patient size (larger patients attenuate more photons, raising noise unless technique compensates), and iterative reconstruction (denoises the image without an added dose penalty — a major advantage over filtered backprojection alone).
Uniformity requires that a homogeneous material like water produce the same CT number regardless of its position in the image — the center of a water phantom should read the same HU as the periphery. It is checked as part of daily QC. Nonuniformity most often signals beam hardening (producing a "cupping" pattern where the center reads artifactually lower than the periphery) or a subtler detector calibration drift. A uniformity failure on daily QC is a systemic problem affecting every subsequent patient study on that scanner, not an isolated bad image — clinical scanning should stop until it is resolved.
| Metric | What It Measures | Key Influencing Factors | Trades Off Against |
|---|---|---|---|
| Spatial resolution | Separating two small high-contrast objects | Detector/pixel size, focal spot, kernel, slice thickness | Contrast resolution, noise |
| Contrast resolution | Distinguishing subtle density differences | mAs, slice thickness, kernel, iterative reconstruction | Spatial resolution |
| Temporal resolution | Time to acquire one image's data | Gantry rotation time, half-scan, dual-source, multisegment | Noise (faster scans use less data) |
| Noise/uniformity | Random CT-number fluctuation; equal readings across a uniform object | mAs (1/√mAs), kVp, slice thickness, patient size | Dose |
Exam Scenario
A radiologist requests thinner reconstructions (1 mm instead of the original 5 mm) to better characterize a small pulmonary nodule adjacent to the diaphragm. Thinner slices reduce partial volume averaging and improve spatial (z-axis) resolution, letting the nodule's true margins and density be seen clearly. But because fewer photons contribute to each thinner voxel, image noise rises. The correct technologist response is not to simply accept noisier images: apply iterative reconstruction to control the added noise without increasing dose, or, if imaging is being repeated, modestly increase mAs — always balancing the diagnostic benefit against the ALARA (as low as reasonably achievable) principle.
A technologist increases mAs from 100 to 400 while keeping all other parameters constant. Based on the noise-versus-mAs relationship, what happens to image noise?
A technologist selects a sharp bone reconstruction kernel to better delineate temporal bone detail. What is the most likely trade-off?
During a coronary CT angiogram, which technique most directly improves temporal resolution?
Daily QC on a water phantom shows the center of the image reading noticeably lower in CT number than the periphery. What does this most likely indicate?