Spatial Resolution, Modulation Transfer Function & Matrix Parameters

---

Fundamental Concepts of Spatial Resolution

Computed tomography transforms raw x-ray attenuation readings into a three-dimensional digital volume composed of discrete volumetric picture elements (voxels). The spatial fidelity with which this volume mirrors physical anatomy is defined by the system's spatial resolution.

Spatial resolution—historically termed high-contrast resolution—is the measure of an imaging system's ability to distinguish two separate, small, high-density structures situated in close spatial proximity against an adjacent background. High-contrast evaluation assumes an extreme attenuation difference between the target objects and their surroundings, typically exceeding $100\text{ HU}$ to several thousand Hounsfield Units. Typical clinical examples include differentiating fine osseous trabeculae in the petrous temporal bone, resolving calcified atherosclerotic plaques within a contrast-opacified coronary artery, and identifying micro-nodules or bronchovascular margins in thin-section pulmonary imaging.

Units of Measurement & Spatial Frequency

Spatial resolution in CT is expressed as a spatial frequency in units of line pairs per centimeter ($\text{lp/cm}$) or line pairs per millimeter ($\text{lp/mm}$). A line pair consists of one high-density, radio-opaque bar and one adjacent radiolucent space of identical width:

$$1\text{ lp/mm} = 10\text{ lp/cm}$$

For equal-width bars and spaces, bar width is 1/(2f), where f is line-pair frequency in lp/mm. At 1.5 lp/mm, the bar width is approximately 0.333 mm. This is the width of the test pattern's bar, not a universal minimum lesion diameter. Lesion detectability also depends on contrast, noise, geometry and task.

Determinants of In-Plane ($xy$) Spatial Resolution

In-plane spatial resolution describes resolving power within the transverse axial cross-section ($x$-axis and $y$-axis). It is dictated by five interrelated geometric and computational factors:

1. Focal spot and geometric blur

A finite focal spot creates geometric blur. A smaller focal spot can improve high-contrast detail under suitable conditions, while a larger spot can support greater loading. The observed effect depends on magnification and the system geometry. Selecting a small spot does not remove detector-aperture, reconstruction or motion limitations, and some exposure settings may require a larger supported spot.

2. Detector Element Aperture & Sampling Pitch

The physical dimensions and active collection aperture of the solid-state scintillator crystals determine the finest spatial variation the detector array can measure:

• Detector Aperture Width: A smaller physical crystal width captures attenuation profiles over a narrower spatial increment, reducing aperture blurring.
• Sampling Theorem & The Nyquist Limit: In-plane spatial resolution is fundamentally bounded by the spatial sampling frequency ($f_s$), which is determined by the center-to-center detector spacing ($d$) projected at the gantry isocenter. According to the Nyquist-Shannon sampling theorem, the highest spatial frequency that can be accurately reconstructed without aliasing—termed the Nyquist frequency ($f_{\text{Nyquist}}$)—is exactly half the sampling frequency:

$$f_{\text{Nyquist}} = \frac{f_s}{2} = \frac{1}{2 \times d}$$

If anatomical structures exhibit spatial variations exceeding $f_{\text{Nyquist}}$, the scanner cannot resolve them and instead misinterprets them as spurious low-frequency patterns known as aliasing artifacts (fine radiating streaks or moiré patterns).

3. Flying focal spots improve supported sampling

Dynamic focal-spot deflection can interleave measurements from different source positions in supported systems. It improves sampling without requiring the same increase in physical detector elements. The benefit depends on the geometry and mode; it does not automatically double every measure of physical resolution or eliminate all aliasing. Distinguish increased sample density from the finite aperture response and reconstruction bandwidth.

4. Reconstruction Convolution Kernels

During filtered back projection (FBP) or iterative reconstruction, raw projection data are mathematically convolved with a user-selected reconstruction kernel (filter) that dictates how spatial frequencies are weighted:

• Sharp / High-Pass Kernels (e.g., Bone, Edge, Lung Filters): Enhance high spatial frequencies, sharpening structural interfaces such as cortical bone margins, temporal bone air cells, and fine pulmonary interstitium. The trade-off is substantial amplification of high-frequency quantum noise (graininess).
• Smooth / Low-Pass Kernels (e.g., Soft Tissue, Standard, Brain Filters): Attenuate high spatial frequencies to suppress image noise and maximize contrast detectability in parenchymal tissues (liver, spleen, cerebral cortex). The trade-off is structural edge blurring and reduced spatial resolution.

5. Matrix Size, Display Field of View (DFOV), and Pixel Mathematics

A cross-sectional CT image is arranged as a digital matrix of rows and columns containing $N \times N$ pixels. The physical dimension of each square pixel along the $x$ and $y$ axes is determined by the Display Field of View (DFOV) and the reconstruction matrix:

$$\text{Pixel Size (mm)} = \frac{\text{DFOV (mm)}}{\text{Matrix Size (pixels)}}$$

• Matrix Dimensions: Standard CT utilizes a $512 \times 512$ matrix ($262,144\text{ pixels}$). Modern high-definition scanners offer $1024 \times 1024$ ($1,048,576\text{ pixels}$) or $2048 \times 2048$ matrices.
• Display Field of View (DFOV): The diameter of the circular reconstruction field selected by the technologist to be displayed across the matrix. This is distinct from the Scan Field of View (SFOV), which represents the actual physical area within the gantry bore where raw projection data are collected ($40\text{ to }50\text{ cm}$).
• Geometric Constraint: Pixel size constrains sampling, but it is not a complete measure of the physical spatial response. Smaller pixels cannot overcome blur or missing projection information.

Worked Example: Pixel Size Calculation

A technologist performs a high-resolution CT of the temporal bones. The radiologist requests a targeted reconstruction with a $12\text{ cm}$ ($120\text{ mm}$) DFOV across a standard $512 \times 512$ matrix. What is the resulting pixel size, and how does it compare to a routine $36\text{ cm}$ ($360\text{ mm}$) abdominal reconstruction?

$$\text{Pixel Size}_{\text{Temporal Bone}} = \frac{120\text{ mm}}{512} \approx 0.234\text{ mm} \quad (0.23\text{ mm})$$ $$\text{Pixel Size}_{\text{Abdomen}} = \frac{360\text{ mm}}{512} \approx 0.703\text{ mm} \quad (0.70\text{ mm})$$

Targeting the reconstruction to a $12\text{ cm}$ DFOV decreases pixel dimension from $0.70\text{ mm}$ to $0.23\text{ mm}$, yielding a threefold improvement in spatial sampling; resolving the ossicles still depends on the system response (malleus, incus, stapes).

---

Longitudinal ($z$-Axis) Spatial Resolution & Volumetric Isotropy

Longitudinal spatial resolution describes the scanner's resolving power along the patient's long axis ($z$-axis, head-to-toe direction). Unlike projection radiography, which is purely planar, multi-detector CT acquires a three-dimensional volume.

The Slice Sensitivity Profile (SSP)

In an ideal scanner, an acquired slice would be a perfect rectangle of uniform thickness along the $z$-axis. In practice, x-ray beam divergence, focal spot geometry, pre-patient collimator penumbra, and helical interpolation cause the slice profile to flare at its margins. The actual shape of the slice response along the $z$-axis is known as the Slice Sensitivity Profile (SSP).

Full Width at Half Maximum (FWHM)

The effective slice thickness of a CT system is quantified by the Full Width at Half Maximum (FWHM) of the SSP curve. FWHM is defined as the distance along the $z$-axis between the two points on the SSP curve where the sensitivity equals exactly 50% ($0.5$) of its maximum peak value:

• When the SSP is narrow and rectangular, the FWHM matches the nominal collimated slice thickness, delivering sharp $z$-axis resolution.
• When the SSP broadens with wide sloping tails, the effective slice thickness increases beyond the nominal value. Surrounding tissues outside the target plane contribute to the image, causing partial volume averaging along the $z$-axis that obscures subtle anatomical margins.

Isotropic Voxels & 3D Multiplanar Reformations

A voxel is the 3D volume element whose base is the in-plane pixel ($\Delta x \times \Delta y$) and whose depth is the reconstructed slice thickness ($\Delta z$):

$$\text{Voxel Dimensions} = \Delta x \times \Delta y \times \Delta z$$

When the reconstructed slice thickness equals the in-plane pixel dimension ($\Delta x = \Delta y = \Delta z$), the voxel forms a perfect geometric cube and is termed isotropic:

• Clinical Value of Isotropic Imaging: When voxels are isotropic (e.g., $0.5\text{ mm} \times 0.5\text{ mm} \times 0.5\text{ mm}$), multiplanar reformations (coronal, sagittal, oblique) and 3D volume renderings provide fine geometric sampling in alternative planes, while actual spatial resolution remains dependent on the acquisition and reconstruction, reducing the stepped "staircase" artifact characteristic of thick-slice anisotropic imaging.

---