Pixel Size, Matrix & Zoom in Nuclear Medicine

Pixel size in nuclear medicine is a sampling decision, not a cosmetic one. Chosen too large, the image undersamples the camera's resolution and throws away detail the system could have shown; chosen too small, it scatters a fixed number of detected counts across too many pixels and buries the image in statistical noise. Getting it right means matching the pixel to the system's spatial resolution and to the counts you actually have. ^{1, 2}

This guide derives the familiar FWHM/3 rule from the Nyquist sampling theorem, shows exactly how matrix size and acquisition zoom set the pixel size, and explains the count-density tradeoff that decides the right choice for planar and SPECT studies. DRPS covers these acquisition parameters in its PET/CT and nuclear medicine physics acceptance testing and protocol-review services across Florida, Maryland, Virginia, Washington DC, California, Nevada, Pennsylvania, New York, New Jersey, and Delaware.

Introduction

A gamma camera does not record a continuous image. It records discrete counts in a grid of pixels, and the physicist choosing the acquisition decides how fine that grid is. That single choice — expressed through matrix size and zoom — sets the pixel size, and the pixel size sits at the intersection of two competing pressures. ^{1, 3}

On one side is resolution: a pixel much larger than the system's ability to resolve detail will blur that detail away, and the image will look coarse regardless of how good the collimator and detector are. On the other side is noise: because every nuclear medicine study is count-limited, a pixel much smaller than necessary spreads the same counts thinner, raising the statistical noise in each pixel. ^{1, 2}

The art of choosing an acquisition matrix is finding the pixel size that is small enough to preserve the system's real resolution but no smaller, so that counts are not wasted. The physics that governs this choice is sampling theory, and the working rule it produces — pixel size no larger than about one third of the system FWHM — appears in essentially every nuclear medicine quality-assurance reference. ^{1, 2, 3}

Topic Explanation

How pixel size is set

Pixel size is the imaged field of view divided by the number of pixels across it. Two acquisition settings control that number: the matrix size (for example 64, 128, or 256 pixels on a side) and the acquisition zoom, which magnifies the acquired region so that the same matrix spans a smaller field of view:

D = \frac{FOV}{Z \times N}

Here $D$ is the pixel size, $FOV$ is the widest dimension of the useful field of view, $Z$ is the zoom factor, and $N$ is the matrix dimension. A large modern detector might have a useful field of view of about 400 mm. Acquired on a 128 by 128 matrix with no zoom, the pixel size is: ^{1, 3}

D = \frac{400 mm}{1 \times 128} \approx 3.1 mm

The same camera on a 64 by 64 matrix gives a 6.25 mm pixel; on a 256 by 256 matrix it gives a 1.56 mm pixel. Apply a zoom of 2 to the 128 matrix and the pixel drops to about 1.56 mm, but the field of view now covers only half the detector. This is why matrix and zoom must be chosen together and in light of the anatomy. For small-organ work this connects directly to studies like thyroid uptake measurement and pediatric protocols.

The role of system resolution

The pixel size only matters relative to how sharply the system can image in the first place. Overall system spatial resolution combines the intrinsic resolution of the detector with the collimator resolution, which dominates and which degrades with distance from the collimator face. System resolution is conventionally reported as a full width at half maximum (FWHM) of the imaged profile of a point or line source. For a low-energy high-resolution collimator at a typical 10 cm source-to-collimator distance, system FWHM is on the order of 7 to 10 mm. ^{1, 2, 6}

The pixel-size decision is therefore a question about this FWHM: how finely must the grid sample a blur of a given width so that the grid itself does not become the limiting factor?

Key Technical Principles

Nyquist sampling and the FWHM/3 rule

The Nyquist sampling theorem states that to faithfully represent a signal containing spatial frequencies up to some maximum $f_{m a x}$ , the sampling must occur at least twice per cycle of that highest frequency. Equivalently, for a pixel size $D$ , the highest spatial frequency the image can represent — the Nyquist frequency — is: ^{1, 3}

f_{N} = \frac{1}{2 D}

Any true detail finer than $f_{N}$ is not merely lost; it can be aliased, folding back into the image as false lower-frequency structure. To avoid aliasing and preserve resolution, the pixel must be small enough that the system's limiting spatial frequency falls at or below $f_{N}$ .

The system's limiting frequency is tied to its FWHM. A Gaussian-like response with full width at half maximum FWHM carries meaningful signal out to a spatial frequency of roughly $1/ FWHM$ . Setting the Nyquist frequency at least equal to that requires:

\frac{1}{2 D} \geq \frac{1}{FWHM} \Rightarrow D \leq \frac{FWHM}{2}

Pure Nyquist sampling therefore demands a pixel no larger than half the FWHM. In practice the nuclear medicine community adopts the more conservative convention of about one third of the FWHM, which places roughly three samples across the resolution element and comfortably protects against aliasing and partial-pixel effects: ^{1, 2}

D \leq \frac{FWHM}{3}

For a system with a 9 mm FWHM at the imaging distance, the pixel should be about 3 mm or smaller — which the 128 matrix (3.1 mm) on our example camera essentially meets, while the 64 matrix (6.25 mm) clearly undersamples.

The count-density tradeoff

If smaller pixels always sampled better, we would always choose the largest matrix. We do not, because nuclear medicine is count-limited: the total number of detected counts is fixed by the administered activity, the uptake, and the acquisition time. Those counts are divided among the pixels, so the counts per pixel scale with the pixel area:

counts per pixel \propto D^{2}

Halving the pixel size (doubling the matrix dimension) quadruples the number of pixels and cuts the counts per pixel to about one quarter. Because the statistical (Poisson) noise in a pixel with $n$ counts has a fractional size of $1/ n$ , quartering the counts per pixel roughly doubles the fractional noise:

\frac{σ}{n} = \frac{1}{n}

So shrinking the pixel below the FWHM/3 point buys no additional real resolution — the collimator and detector already set that limit — while it measurably worsens noise. The matched choice sits right at the sampling requirement: small enough to preserve resolution, large enough to keep counts per pixel high. ^{1, 2, 7}

The table shows the tradeoff for the example 400 mm camera with a 9 mm system FWHM.

Matrix (no zoom)	Pixel size	Nyquist frequency	Pixels in image	Relative counts/pixel	Sampling verdict
64 × 64	6.25 mm	0.08 cycles/mm	4,096	16×	Undersampled (pixel over FWHM/3)
128 × 128	3.13 mm	0.16 cycles/mm	16,384	4×	Matched (pixel near FWHM/3)
256 × 256	1.56 mm	0.32 cycles/mm	65,536	1×	Oversampled (noise, no resolution gain)

Why SPECT and planar differ

Planar images concentrate all counts into one two-dimensional frame, so counts per pixel are relatively plentiful and fine matrices (256 or 512) are routine. SPECT divides the counts among many projection angles and then reconstructs a three-dimensional volume, so counts per voxel are far scarcer; 64 or 128 matrices are the norm to keep per-voxel statistics usable. The same FWHM/3 logic applies, but the count-density penalty bites harder in tomography, which is why matrix selection is one of the parameters evaluated in a SPECT/CT quality-control program. Studies of voxel size in reconstructed SPECT confirm that spatial resolution improves with smaller voxels only up to the sampling limit, after which the dominant effect is noise. ^{5, 7}

Clinical Impact

The pixel-size choice shows up in every image a department produces:

Small-organ and pediatric imaging benefit from zoom and finer matrices because the anatomy is small and detail matters; a thyroid or pediatric renal study is a classic place to magnify.
Whole-body and high-count planar imaging can use fine matrices because counts are plentiful.
SPECT myocardial perfusion, bone, and brain studies are generally acquired on 64 or 128 matrices, matched to the collimator resolution and the available counts, so the reconstruction is not dominated by per-voxel noise. This underlies acquisition choices in cardiac SPECT MPI quality control.
Quantitative studies — split renal function, standardized uptake, dopamine-transporter binding ratios — depend on an accurate pixel-size calibration, because measured dimensions and recovery coefficients are computed from the pixel grid. A phantom study of a dopamine-transporter quantification tool showed its output correlated strongly with the tomographic spatial resolution that acquisition parameters, including pixel size, determine. ⁷

A mismatched pixel size rarely makes an image unreadable, which is exactly why it goes unnoticed: a 64-matrix bone SPECT still produces a picture, but it is quietly throwing away resolution the collimator worked to provide. The physicist's job is to make the acquisition matrix match the physics rather than habit. ^{1, 2}

Practical Optimization Tips

1. Know your system FWHM first

Measure or look up the system spatial resolution (intrinsic plus collimator) at the clinically relevant source-to-collimator distance, following NEMA NU-1 methods. The FWHM at that distance — not the intrinsic resolution alone — sets the sampling target. ⁴

2. Apply the FWHM/3 rule

Choose the smallest matrix (largest pixel) whose pixel size is still at or below one third of that FWHM. That protects resolution while keeping counts per pixel as high as possible.

3. Use zoom deliberately for small anatomy

For the thyroid, pediatric, or other small-field studies, use zoom to magnify and sample finely — but confirm the field of view still contains all the anatomy and that the pixel has not shrunk far below the useful limit.

4. Protect SPECT counts

In tomography, resist the temptation to over-matrix. A 128 matrix that satisfies FWHM/3 is preferable to a 256 matrix that merely adds noise. Match matrix, time per projection, and number of projections together.

5. Verify the pixel-size calibration

During acceptance testing, image two point or line sources a known distance apart and confirm the measured separation agrees with the physical separation, verifying the stored pixel-size calibration that all quantitative results depend on. ^{1, 4}

Common pitfalls to avoid

Defaulting to the vendor preset without checking the FWHM. The right matrix depends on collimator and distance, not on habit.
Over-matrixing SPECT. A finer matrix with the same counts is a noisier reconstruction, not a sharper one.
Zooming without checking the field of view. Magnification can crop anatomy out of the image.
Trusting measured dimensions on an uncalibrated system. A wrong pixel size silently corrupts organ sizing and quantitation.

Regulatory Considerations

Pixel size and matrix selection are acquisition-quality parameters rather than items of federal radiation-safety law, but they are governed by performance standards and quality-assurance guidance and are checked during acceptance testing and accreditation. The medical physicist documents the system resolution, the pixel-size calibration, and the acquisition matrices used for each protocol.

NEMA NU-1-2018 (Performance Measurements of Gamma Cameras) defines the standardized methods for measuring intrinsic and system spatial resolution and the acquisition conditions that make those measurements comparable across systems. ⁴
IAEA Human Health Series No. 6 (Quality Assurance for SPECT Systems, 2009) and the IAEA Quality Control Atlas for Scintillation Camera Systems (2003) give the practical acceptance and routine test procedures, including the sampling and pixel-size considerations for SPECT. ^{2, 8}
IAEA Nuclear Medicine Physics: A Handbook for Teachers and Students (2014) provides the underlying sampling theory, the Nyquist relationship, and the FWHM-based pixel-size rule. ¹

The radioactive material used to produce these images is governed separately by the NRC under 10 CFR Parts 20 and 35, or by an Agreement State program; of the states DRPS serves, most are NRC Agreement States while Washington DC and Delaware are regulated directly by the NRC for radioactive material. Acquisition-parameter review is part of the physics support that accompanies accreditation support and routine medical physicist consulting. ⁴

Frequently Asked Questions (FAQs)

What sets the pixel size in a nuclear medicine image?

Pixel size is the field of view divided by the product of the acquisition zoom and the matrix dimension. For a 400 mm field of view acquired on a 128 by 128 matrix with no zoom, the pixel is about 3.1 mm. Increasing the matrix dimension or the zoom makes each pixel smaller; decreasing them makes each pixel larger.

What is the FWHM/3 rule?

The pixel size should be no larger than about one third of the system spatial resolution expressed as full width at half maximum (FWHM). Sampling theory requires at least two samples across the finest detail the system can resolve; using roughly three pixels per FWHM is the conservative convention that preserves resolution without undersampling. For a system with a 9 mm FWHM at the imaging distance, the pixel should be about 3 mm or smaller.

Why not just use the smallest possible pixel?

Because the total number of detected counts is fixed by the administered activity and acquisition time. Halving the pixel size quadruples the number of pixels and cuts the counts per pixel to about one quarter, roughly doubling the statistical noise in each pixel. Beyond the point where the pixel is already about one third of the FWHM, shrinking it further adds noise without adding real resolution.

What is the Nyquist frequency in imaging?

The Nyquist frequency is the highest spatial frequency an image can faithfully represent for a given pixel size, equal to one divided by twice the pixel size. Detail finer than the Nyquist limit is not just lost — it can be aliased, appearing as false lower-frequency structure. Choosing a pixel small enough that the system resolution limit falls at or below the Nyquist frequency prevents aliasing.

What matrix sizes are typical in planar and SPECT imaging?

Planar images are commonly acquired on 256 by 256 or 512 by 512 matrices because counts are relatively plentiful and fine detail is wanted. SPECT studies are commonly acquired on 64 by 64 or 128 by 128 matrices because the counts are divided among many projection angles and across three dimensions, so smaller matrices preserve counts per pixel. The correct choice depends on the system FWHM and the available counts.

How does acquisition zoom interact with matrix size?

Zoom magnifies the acquired region so the same matrix covers a smaller field of view, which reduces pixel size. Zoom is useful for small organs such as the thyroid or for pediatric imaging, where a magnified, finely sampled image is wanted. It must be used deliberately: too much zoom shrinks the pixel below the useful sampling limit and worsens noise, and it can crop anatomy out of the field of view.

Is pixel size a quality-control concern?

Yes. The accuracy of the stored pixel size is verified during acceptance testing by imaging two point or line sources a known distance apart and confirming the measured separation, because an incorrect pixel-size calibration corrupts measured organ dimensions, uniformity analysis, and any quantitative result such as a split-function or standardized-uptake measurement.

Key Takeaways

Pixel size is a sampling choice. It is the field of view divided by matrix dimension times zoom, and it must be matched to the system resolution and the available counts.
The FWHM/3 rule comes from Nyquist. Pure Nyquist sampling needs a pixel no larger than FWHM/2; the conservative FWHM/3 convention puts about three samples per resolution element and avoids aliasing.
Smaller is not always better. Counts per pixel scale with the pixel area, so over-matrixing a count-limited study just doubles noise without adding resolution.
SPECT is stricter than planar. Tomography divides counts among angles and voxels, so 64 or 128 matrices are normal while planar can afford 256 or 512.
Calibrate the pixel. Quantitative results depend on an accurate pixel-size calibration, verified with sources a known distance apart.

Conclusion

The matrix and zoom settings on an acquisition screen look like preferences, but they encode a specific physics decision about how finely to sample a blurred image made of a fixed, scarce number of counts. The Nyquist theorem sets the floor — sample at least twice across the finest resolvable detail — and the count-density penalty sets the ceiling, because every unnecessary pixel dilutes the statistics.

Matching the pixel to about one third of the system FWHM resolves the tension: it preserves every bit of resolution the collimator and detector provide while keeping counts per pixel as high as the study allows. A department that sets its matrices by that rule, rather than by vendor default or habit, gets images that are as sharp as the hardware permits and no noisier than they have to be.

How DRPS Can Help

Diagnostic Radiation Physics Services evaluates gamma camera and SPECT acquisition parameters — system resolution, pixel-size calibration, and the matrix and zoom choices behind each protocol — as part of its PET/CT and nuclear medicine physics acceptance testing and medical physicist consulting. A DRPS review ties NEMA-based resolution measurements to the department's clinical matrices so that planar and SPECT studies sample correctly, and connects that work to accreditation support.

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

Related Resources

References

International Atomic Energy Agency. Nuclear Medicine Physics: A Handbook for Teachers and Students. Vienna: IAEA; 2014. iaea.org
International Atomic Energy Agency. IAEA Human Health Series No. 6: Quality Assurance for SPECT Systems. Vienna: IAEA; 2009. iaea.org
Groch MW, Erwin WD. SPECT in the year 2000: basic principles. J Nucl Med Technol. 2000;28(4):233-244. PubMed
National Electrical Manufacturers Association. NEMA Standards Publication NU 1-2018: Performance Measurements of Gamma Cameras. Rosslyn, VA: NEMA; 2018. nema.org
Kappadath SC. Effects of voxel size and iterative reconstruction parameters on the spatial resolution of 99mTc SPECT/CT. J Appl Clin Med Phys. 2011;12(4):3459. doi:10.1120/jacmp.v12i4.3459. PubMed
Erlandsson K, Kacperski K, van Gramberg D, Hutton BF. Performance evaluation of D-SPECT: a novel SPECT system for nuclear cardiology. Phys Med Biol. 2009;54(9):2635-2649. doi:10.1088/0031-9155/54/9/003. PubMed
Poli GL, Bianchi C, Guerra UP. Use of the BasGan algorithm for [123I]FP-CIT SPECT quantification: a phantom study. Q J Nucl Med Mol Imaging. 2013;57(4):391-400. PubMed
International Atomic Energy Agency. IAEA Quality Control Atlas for Scintillation Camera Systems. Vienna: IAEA; 2003. iaea.org
Kohli V, King MA, Glick SJ, Pan TS. Comparison of frequency-distance relationship and Gaussian-diffusion-based methods of compensation for distance-dependent spatial resolution in SPECT imaging. Phys Med Biol. 1998;43(4):1025-1037. doi:10.1088/0031-9155/43/4/029. PubMed