MTF and Spatial-Resolution Testing

Modulation Transfer Function (MTF): Quantitative Spatial Fidelity

While line pair counts provide a qualitative assessment of spatial resolution, the Modulation Transfer Function (MTF) represents the objective, universal mathematical standard for characterizing the spatial resolution performance of a CT system across the complete spectrum of spatial frequencies.

Modulation and interpretation

MTF describes the transfer of spatial contrast as a function of frequency and is commonly normalized at zero frequency. It can be obtained from the normalized magnitude of a Fourier transform of a suitable line-spread response under the measurement model. Edge-enhancing kernels can give values above one at some frequencies; the curve need not simply decline monotonically from one to zero.

The frequencies at 50% and 10% modulation are useful reporting conventions. MTF10 is not a point at which contrast universally becomes zero, nor is it a universal guaranteed clinical lesion limit. For nonlinear reconstruction, measured spatial response may depend on contrast and dose; the physicist selects an appropriate method rather than assuming that one high-contrast measurement represents every low-contrast task.

Kernels, patterns and reproducibility

Sharper kernels generally preserve or enhance higher spatial frequencies while increasing noise relative to smoother kernels. Compare measurements made with the same acquisition, reconstruction and field of view. A high-resolution bone result cannot be substituted for the performance of a soft-tissue protocol.

A line-pair phantom supports visual assessment of bars and spaces, but observer, window and noise affect the apparent limit. Follow the phantom's specified procedure and applicable program. Do not invent an ACR mandate that every scanner resolve 5 lp/cm with body and 12 lp/cm with bone reconstruction. The accreditation and periodic-test requirements must be read in their actual context.

An applied comparison can reveal the limitation: a sharp kernel may show a higher line-pair group more clearly but make a low-contrast insert harder to see. That result is not contradictory. The two measurements assess different tasks, and neither alone establishes that the clinical protocol is optimized. Record the kernel and exposure so that a later measurement can be compared fairly.

If measured response changes after service, first reproduce centering, phantom orientation, reconstruction and analysis. A changed DFOV can change sampling even with unchanged hardware. A changed algorithm can alter edge behavior and noise texture. Only after checking these inputs should the team interpret a difference as evidence of system drift or degraded performance.

Interpreting a measured response

For a hypothetical comparison, a protocol's MTF reaches 50% at 0.4 lp/mm and 10% at 0.8 lp/mm. Another kernel reaches those levels at 0.6 and 1.1 lp/mm. The second preserves more high-frequency contrast in that measurement, but the values alone do not tell how well a low-contrast liver lesion is detected. Compare noise, contrast dependence and the clinical task before choosing the kernel.

The test object must be sufficiently small or sharp for the method. A finite bead or edge can contribute its own blur, and poor sampling can distort the measured response. The analyst may need an oversampling or slanted-edge method, appropriate background subtraction and normalization. A plotted curve with many points is not proof that the underlying image contained sufficiently fine independent measurements.

For a longitudinal test, a thin bead or other prescribed object can be scanned and reconstructed at closely spaced positions to estimate the slice sensitivity profile. The width at half maximum describes one aspect of that response; tails can still contribute partial-volume effects. Record both the acquisition configuration and reconstructed thickness. Two settings with the same displayed thickness may have different profiles, so a meaningful comparison requires the method and conditions.

Quantitative Evaluation: PSF, LSF, and ESF Analysis

While bar patterns rely on subjective human visualization, medical physicists measure spatial resolution objectively using mathematical impulse functions:

1. Point Spread Function (PSF): The two-dimensional image profile produced by an infinitesimally small, highly attenuating test object—typically a high-density tungsten carbide or gold bead ($< 0.1\text{ mm}$ diameter) suspended in a uniform acrylic block. The degree to which the bead is smeared in the reconstructed image represents the system's PSF.
2. Line Spread Function (LSF): The one-dimensional attenuation profile across an infinitesimally thin wire ($0.05\text{ mm}$ tungsten or platinum wire) positioned perpendicular to the scan plane along the $z$-axis.
3. Edge Spread Function (ESF): The attenuation step response measured across a razor-sharp planar boundary between two materials with contrasting linear attenuation coefficients (e.g., an acrylic-air or PTFE-water edge).

Mathematical Derivation of MTF from Impulse Responses

The Line Spread Function is mathematically related to the Edge Spread Function as its first spatial derivative:

$$\text{LSF}(x) = \frac{d}{dx} \left[ \text{ESF}(x) \right]$$

The Modulation Transfer Function is the normalized modulus of the one-dimensional Fourier Transform ($\mathcal{F}$) of the Line Spread Function:

$$\text{MTF}(f) = \frac{|\mathcal{F}\{\text{LSF}(x)\}|}{|\mathcal{F}\{\text{LSF}(x)\}|_{f=0}}$$

This Fourier analysis decomposes the blurred line profile into its constituent spatial frequency sine-wave components, yielding the continuous MTF curve from $0$ to the limiting cutoff frequency.

---