Interdependence of Image Quality Parameters & Radiation Dose

Optimize for the task rather than one attractive number

CT image quality involves spatial detail, low-contrast detectability, temporal behavior, noise and artifacts. Dose and acquisition time constrain the available choices. A bone fracture and a subtle liver lesion do not require the same reconstruction, and an image with less visible noise is not automatically more useful. Optimize the complete acquisition and reconstruction for the question and patient.

Some changes involve a tradeoff, while improved hardware or reconstruction can improve performance without the same historical penalty. Avoid the claim that every improvement invariably worsens something else. More importantly, do not assume that a new reconstruction algorithm permits any desired dose reduction without testing whether clinically important findings remain detectable.

Exposure and the inverse-square-root approximation

When voltage, patient size, geometry, thickness and reconstruction remain fixed, dose is approximately proportional to tube current-time product, mAs. In a quantum-noise-limited situation with otherwise unchanged processing, image noise is approximately inversely proportional to the square root of mAs:

$$\frac{\sigma_2}{\sigma_1}\approx\sqrt{\frac{mAs_1}{mAs_2}}$$

Suppose noise is 12 HU at 100 mAs. At 200 mAs, the estimate is 12/√2 = 8.49 HU, a reduction of about 29.3%. At 400 mAs, the estimate is 6 HU. Quadrupling mAs produces four times the dose under the stated assumptions, which is a 300% increase, not a 400% increase. Electronic noise, modulation and nonlinear reconstruction can make this approximation less accurate.

Thickness, interval and kernel

Thicker reconstructed slices generally average more information and reduce noise, but increase partial-volume effects. Thin slices improve longitudinal sampling and support fine reformations. With otherwise comparable linear reconstruction, changing thickness from 5 mm to 1.25 mm can approximately double noise because √(5/1.25) = 2. This is a useful model, not an instruction to automatically quadruple patient exposure whenever thin images are reconstructed.

A single thin acquisition can provide both thin images for reformations and thicker images for low-contrast review. The reconstruction interval controls the spacing between image centers. Overlapping images do not by themselves add x-ray exposure when made from already acquired data. They can improve reformation smoothness, but do not create new independently acquired spatial information.

A sharp kernel emphasizes high-frequency detail and increases noise relative to a smooth kernel. Use it for appropriate high-contrast tasks such as bone or lung detail. A smooth kernel is often better for low-contrast soft tissue. Neither is universally superior, and the degree of noise change is not a fixed percentage for every scanner.

Pixel size and field of view

For a square reconstruction, pixel width equals DFOV divided by matrix dimension. A 320 mm DFOV on a 512 matrix gives 0.625 mm pixels. Reducing DFOV to 160 mm gives 0.3125 mm pixels. This improves sampling of the selected region, but does not guarantee twice the physical resolving power.

CT pixels do not each correspond to a separate detector bucket receiving photons from one voxel. Reconstruction combines many projection measurements, so image noise also depends on bandwidth, correlations and the algorithm. Do not use a universal noise-proportional-to-inverse-pixel-width rule for every display-field change. A magnified display of already reconstructed pixels likewise differs from a new targeted reconstruction from raw data.

Pitch and scan duration

Beam pitch is table travel per rotation divided by total nominal acquisition collimation. Increasing pitch generally covers anatomy faster for a fixed rotation time and collimation. For a fixed mAs per rotation and the applicable helical dose-index definition, CTDIvol = CTDIw/pitch. Increasing pitch from 1.0 to 1.5 then changes the index by a factor of 1/1.5 = 0.667.

That dose comparison is conditional. An automatic system or effective-mAs setting may increase tube output as pitch increases, maintaining image noise and changing the dose result. Modern multidetector reconstructions can also keep reconstructed slice width largely independent of pitch over their validated ranges. Do not apply a universal 15–25% slice broadening to every MDCT scan.

Voltage, patient size and iodine

Lower tube voltage changes the spectrum and can increase iodine contrast. It also reduces penetration and can increase noise, particularly through a large patient. The appropriate voltage depends on patient size, the task, tube capacity and the scanner's validated protocol. There is no universal voltage choice for all children or all large adults, nor a guaranteed percentage of dose reduction from changing 120 to 80 kVp.

Automatic exposure control can alter current as the patient size and projection angle change. Center the patient before the localizer so that the system's size estimate and bowtie geometry are appropriate. An off-center scout can lead to inappropriate output or uneven noise; correct positioning is a dose-and-quality action, not merely a cosmetic one.

Choose the change that addresses the actual failure

If motion is the main problem, simply increasing mAs may add dose without removing the artifact. If a small lesion is obscured by partial volume, a suitable thin reconstruction from available raw data may help without rescanning. If photon starvation is present, a physicist-approved change in exposure or voltage may be needed. Review the cause before changing parameters, and document a protocol modification so that its quality and dose effects can be assessed.

References: AAPM CT terminology and protocols, FDA optimization principles.

Interscan spacing, gaps and overlap

For sequential axial acquisition, distinguish the nominal acquired width from the table increment between exposures. In a simplified single-width example, a 5 mm acquisition repeated every 7 mm leaves a 2 mm nominal gap. A small lesion in that gap may be missed. The same 5 mm width repeated every 3 mm overlaps by 2 mm. At unchanged exposure per acquisition, that overlap increases output per unit scanned length. Actual multirow coverage and dose profiles require the scanner's acquisition definitions.

Reconstructed image spacing is a different choice. Making 5 mm-thick images every 3 mm from existing adequate volume data creates overlapping reconstructions without adding radiation. It does not recover anatomy never adequately sampled during acquisition. Confirm the source coverage, reconstructed thickness and center-to-center interval separately before fixing a missing-detail problem. This distinction prevents both undetected gaps and unnecessary repeat scans.