PET Partial Volume Effect & Recovery Coefficients

The partial volume effect (PVE) is the systematic bias that makes small lesions on PET appear less intense than they truly are, because the scanner's finite spatial resolution blurs activity out of small objects and blurs background activity in. The consequence is that the standardized uptake value (SUV) of a small lesion is underestimated — sometimes severely — and unless this is understood and corrected, quantitative PET can mislead staging, response assessment, and dosimetry.

PET is prized for being intrinsically quantitative: in principle it measures activity concentration in becquerels per milliliter anywhere in the body. In practice, that promise holds only for objects large compared with the system resolution. Below that size, the partial volume effect degrades accuracy in a size-dependent, predictable way. This guide explains the physics of PVE, how recovery coefficients quantify and correct it, the role of NEMA NU-2 phantom measurements and EARL harmonization, and how DRPS evaluates these factors as part of PET/CT and nuclear medicine physics support.

Introduction

A PET image is not a perfect map of the underlying activity distribution. It is that distribution blurred by the point-spread function (PSF) of the entire imaging chain — positron range, photon non-collinearity, detector size, depth-of-interaction effects, and reconstruction. The PSF is usually summarized by its full-width-at-half-maximum (FWHM), which for modern clinical scanners is on the order of a few millimeters intrinsically and somewhat larger after reconstruction.¹

When the object being imaged is large relative to that FWHM, the blurring affects only the edges and the measured activity concentration in the center is accurate. When the object is small relative to the FWHM, the blurring smears its activity across a larger region, lowering the peak and mean values that are read back. This is the partial volume effect, and it is the dominant source of quantitative error for small lesions in oncologic PET.¹

The effect is not random noise that averages out — it is a systematic, reproducible bias. That is both the problem and the opportunity: because PVE is predictable as a function of object size and system resolution, it can be measured with phantoms and corrected with recovery coefficients. The rest of this article develops that idea and its clinical and regulatory implications.

Topic Explanation

Two mechanisms behind PVE

The partial volume effect actually combines two distinct phenomena.¹

• The resolution effect (spill-out and spill-in). Finite spatial resolution convolves the true distribution with the PSF. For a small hot lesion, counts "spill out" into neighboring voxels, lowering the apparent concentration inside the lesion. Simultaneously, activity from surrounding tissue "spills in." For a hot lesion in cooler background, net spill-out dominates and the lesion is underestimated.
• The tissue-fraction (sampling) effect. PET images are sampled on a finite voxel grid, and a voxel at the boundary of a structure contains a mixture of tissues. The reconstructed value is an activity-weighted average of whatever falls within the voxel, so partial occupancy of voxels by the object further dilutes the measured concentration.

Both effects scale with the ratio of object size to system resolution. They are why a 7-mm node and a 7-cm mass with identical true SUV will read very differently on the same scan.

Spill-out, spill-in, and contrast

The direction and magnitude of the bias depend on the contrast between the object and its surroundings. A hot lesion in a cold background loses more than it gains and is underestimated. A small low-uptake region inside a hot organ gains more than it loses and is overestimated. This contrast dependence is why recovery coefficients must be characterized not only as a function of size but, ideally, of object-to-background ratio as well.³⁴

For the foundational quantity that PVE distorts — the SUV — see our companion guide on PET SUV quantification, and for why SUVs must be standardized across scanners, see EARL SUV harmonization.

Key Technical Principles

Image formation as convolution

To a good approximation, the measured PET image $I_{\text{meas}}$ is the true activity distribution $T$ convolved with the system point-spread function:

If the PSF is modeled as a 3D Gaussian, its width is described by the FWHM, related to the standard deviation $\sigma$ by:

Convolving a small uniform sphere with this Gaussian lowers and broadens its profile — the mathematical origin of the partial volume effect.

The recovery coefficient

The practical metric for PVE is the recovery coefficient (RC), the ratio of measured to true activity concentration for an object of a given size:

An RC of 1.0 indicates full recovery; values below 1.0 indicate underestimation. RC depends on object size, shape, contrast, the radionuclide, and the reconstruction. Measured RCs for hot spheres in the NEMA image-quality phantom span a wide range — for example, roughly 0.38 to 1.00 in one study without an out-of-field scatter source, widening to about 0.27 to 1.02 when realistic out-of-field activity was included — illustrating both the size dependence and the influence of acquisition conditions.³

The 2–3 × FWHM rule

A useful rule of thumb follows directly from the convolution: an object recovers its true activity concentration only when its smallest dimension is roughly two to three times the system FWHM. Below that, RC falls progressively; far below it, RC can drop well under 0.5.¹

For a clinical scanner with reconstructed resolution of about 5–6 mm FWHM, this means:

• Lesions larger than about 2 cm are recovered nearly fully (RC near 1).
• Lesions around 1–1.5 cm are measurably underestimated.
• Sub-centimeter lesions can be underestimated by half or more.

Representative recovery curve

The table below shows an illustrative recovery curve for hot spheres against a typical clinical scanner. The exact numbers are scanner-, reconstruction-, and contrast-specific and must be measured for each system; they are shown here only to convey the shape of the relationship.¹³⁵

The six sphere diameters listed — 10, 13, 17, 22, 28, and 37 mm — are those of the standard NEMA NU-2 image-quality phantom used to characterize recovery on clinical systems.⁵

Partial volume correction

Once an RC curve exists, the simplest correction divides the measured value by the recovery coefficient for the lesion's size:

Worked example: a 10-mm lesion is measured at ${\mathrm{SUV}}_{\text{max}}=4.0$ on a scanner whose RC for a 10-mm sphere is 0.50. The partial-volume-corrected value is:

The true uptake is roughly double the uncorrected reading — a difference large enough to change a staging or response decision. Naturally, RC-based correction depends on accurate lesion sizing and amplifies noise, so it must be applied carefully.³⁴

More sophisticated methods include resolution-recovery (PSF) reconstruction, which models the point-spread function inside the iterative reconstruction; image-based deconvolution; and anatomy-based correction using co-registered CT or MR to define object boundaries (for example, geometric transfer matrix methods). PSF reconstruction improves contrast recovery but can introduce edge-overshoot ("Gibbs") artifacts that inflate small-lesion SUVs, so it must itself be characterized.⁶⁷

The radionuclide matters

PVE is not purely geometric — the tracer changes the effective resolution. F-18 has a short positron range and typically yields the highest recovery coefficients (smallest PVE). Higher-energy positron emitters such as Ga-68 have longer range and therefore larger blurring and lower RC for the same lesion size.⁸ I-124, used for thyroid lesion dosimetry, is further complicated by prompt-gamma coincidences and a positron branching ratio of only about 23%, both of which degrade quantification and require dedicated corrections.⁷ This is one reason recovery coefficients must be characterized per radionuclide, not assumed.

Clinical Impact

The partial volume effect directly threatens the three things quantitative PET is used for: staging, treatment-response assessment, and dosimetry. If a small lesion's SUV is underestimated, it may be misclassified as benign or as a responder when it is neither.¹²

In response assessment, the danger is subtle. As a lesion shrinks during therapy, its recovery coefficient falls even if its true activity concentration is unchanged, so the measured SUV drops for a purely geometric reason. Without partial volume awareness, this can be misread as a metabolic response. Conversely, a lesion that grows can appear to "increase uptake" simply because its RC rises. Standardized acquisition and harmonized recovery are what keep serial SUVs comparable.²

In hypoxia and other low-contrast tracer imaging, PVE can dominate. A study of FMISO PET in head-and-neck cancer found that PVE caused substantial underestimation of activity in small high-signal subvolumes, and that recovery-coefficient correction increased the measured hypoxic subvolume and SUV significantly, bringing results into better agreement with published hypoxic-fraction data.⁵ In bone-metastasis imaging with F-18 fluoride, multimodal partial volume correction raised recovered SUVs by tens of percent compared with standard reconstruction.⁶

The practical message for clinical programs is that SUV thresholds, response criteria, and dosimetry calculations are only meaningful when the partial volume context is understood. Accurate quantification also depends on the upstream corrections covered in our guides to PET/CT attenuation correction and the resolution benefits of time-of-flight PET.

Practical Optimization Tips

Characterize your scanner

• Measure recovery coefficients on your own system. Use the NEMA NU-2 image-quality phantom with its six spheres to build an RC curve for each clinically used reconstruction. Do not borrow another scanner's curve.⁵
• Repeat per reconstruction and per radionuclide. PSF on/off, iterations, and post-filter all change RC, as does the tracer. Document the settings that the clinical SUVs are tied to.¹⁸
• Watch for Gibbs overshoot. PSF reconstruction can inflate small-lesion SUVs through edge artifacts; verify against the phantom before trusting absolute values.⁶

Acquire and report consistently

• Standardize the protocol. Uptake time, injected activity, blood glucose, and scan duration all influence quantification; follow a fixed protocol so serial scans are comparable. See PET uptake time.
• Use a harmonized reconstruction for SUV reporting. An EARL-compliant reconstruction keeps SUVs comparable across sites even when a higher-resolution reconstruction is used for visual reading.²¹¹
• Report the method. State whether SUVs are partial-volume corrected, and which reconstruction produced them. An uncorrected SUV and a corrected SUV are different numbers.

Apply correction judiciously

• Correct when size is reliable. RC-based correction needs an accurate lesion dimension; on co-registered CT or MR this is feasible, but errors in delineation propagate into the corrected value.³⁴
• Mind the noise. Dividing by a small RC amplifies noise. Use peak or mean metrics designed for robustness where appropriate, and avoid over-correcting noisy sub-centimeter findings.

Regulatory Considerations

Partial volume performance is governed not by a dose regulation but by the accreditation and standardization frameworks that make quantitative PET trustworthy. The key references are:

• NEMA NU 2, Performance Measurements of Positron Emission Tomographs, which defines the image-quality phantom and the contrast-recovery measurements used to characterize partial volume behavior.¹⁰
• The EANM/EARL accreditation program, which sets recovery-coefficient specifications (the EARL standards) so that SUVs are comparable across scanners and sites — the foundation for multicenter trials and SUV-based response criteria.¹¹
• The EANM FDG PET/CT procedure guidelines (version 2.0), which standardize acquisition and reconstruction to support SUV harmonization and reproducibility.²
• The ACR PET accreditation program, which in the United States requires phantom-based image-quality and quantitative performance testing by a qualified medical physicist.

These are quality and standardization requirements rather than radioactive-material rules, but they intersect with the broader regulatory picture. Possession and medical use of PET radiopharmaceuticals fall under NRC 10 CFR Part 35 or the equivalent Agreement State program, while the scanner's quantitative performance is the medical physicist's responsibility under accreditation. DRPS supports both sides through PET/CT and nuclear medicine physics, accreditation support, and medical physics consulting.

Frequently Asked Questions (FAQs)

What is the partial volume effect in PET?

The partial volume effect (PVE) is the loss of apparent activity concentration in objects that are small relative to the spatial resolution of the PET scanner. Because the system blurs the image, counts from a small hot lesion spill out into surrounding voxels while background counts spill in. The measured SUV of a small lesion is therefore lower than the true value, and the smaller the lesion, the larger the underestimation.

What is a recovery coefficient?

A recovery coefficient (RC) is the ratio of the measured activity concentration to the true activity concentration for an object of a given size. An RC of 1.0 means full recovery (no partial volume bias); an RC of 0.5 means the measured value is only half the truth. RC curves are generated by scanning spheres of known size and activity, and they form the basis of recovery-coefficient partial volume correction.

How small does a lesion have to be for the partial volume effect to matter?

Partial volume bias becomes significant when an object is smaller than about two to three times the system's full-width-at-half-maximum (FWHM) spatial resolution. For a clinical PET scanner with roughly 5–6 mm reconstructed resolution, that means lesions smaller than about 1.5–2 cm are increasingly underestimated, and sub-centimeter lesions can be underestimated severely.

Does the partial volume effect always cause underestimation?

For a hot lesion in a colder background, PVE causes underestimation because more activity spills out than spills in. But a small cold or low-uptake structure surrounded by hot background can be overestimated because background activity spills into it. The direction of the bias depends on the contrast between the object and its surroundings.

How is the partial volume effect corrected?

Common approaches include recovery-coefficient correction (dividing the measured value by the RC for the lesion size), resolution-recovery reconstruction using the scanner point-spread function, image-based deconvolution, and anatomy-based methods that use co-registered CT or MR to define object boundaries. Each method has trade-offs in accuracy, noise amplification, and dependence on accurate lesion delineation.

Why does EARL harmonization matter for partial volume effects?

The EANM Research Ltd (EARL) accreditation program specifies recovery-coefficient limits so that scanners of different makes, models, and reconstruction settings produce comparable SUVs. Because partial volume bias depends on resolution, harmonizing recovery coefficients across sites is what allows SUV-based response assessment and multicenter trials to be valid.

Does the radiotracer affect the partial volume effect?

Yes. Positron range and prompt-gamma emissions differ by radionuclide, which changes effective resolution and quantification. F-18 has a short positron range and typically gives the highest recovery coefficients (smallest PVE), while higher-energy positron emitters such as Ga-68 have longer range and larger blurring. I-124 adds complications from prompt-gamma coincidences and a low positron branching ratio.

Key Takeaways

• PVE is a systematic, size-dependent bias, not noise. It arises from finite spatial resolution (spill-out/spill-in) and voxel sampling.¹
• Recovery coefficients quantify it. RC is the ratio of measured to true concentration; measured RCs for small spheres can fall well below 0.5.³
• The 2–3 × FWHM rule predicts where it bites. Objects smaller than two to three times the system FWHM are progressively underestimated.¹
• Direction depends on contrast. Hot-in-cold underestimates; cold-in-hot can overestimate.
• Correction is possible but conditional. RC-based correction, PSF reconstruction, deconvolution, and anatomy-based methods all help, but each has accuracy and noise trade-offs.⁶⁷
• Standardization is the safeguard. NEMA NU-2 phantom characterization and EARL harmonization keep SUVs comparable across scanners, reconstructions, and sites.⁵¹⁰¹¹

Conclusion

The partial volume effect is the reason a PET scanner's quantitative promise has fine print: SUV is accurate only for objects large compared with the system resolution, and it is biased — usually downward — for everything smaller. Because the bias is systematic and predictable, it is also measurable and, to a degree, correctable. Recovery coefficients turn an abstract blurring phenomenon into a number a physics program can characterize, monitor, and apply.

A defensible quantitative PET program measures its own recovery curves on the NEMA phantom, ties clinical SUVs to a documented and harmonized reconstruction, reports whether values are partial-volume corrected, and educates readers about the size dependence of SUV. Done consistently, this turns PET quantification from a source of subtle error into a reliable biomarker for staging, response, and dosimetry.

How DRPS Can Help

Diagnostic Radiation Physics Services helps PET/CT programs make their quantitation trustworthy. Our board-certified medical physicists perform NEMA NU-2 image-quality and recovery-coefficient testing, support EARL and ACR accreditation, evaluate reconstruction settings for partial volume and Gibbs-artifact behavior, and advise on partial volume correction for response assessment and dosimetry. These services are delivered through PET/CT and nuclear medicine physics, accreditation support, and medical physics consulting.

DRPS supports facilities across Florida, Maryland, Virginia, Washington DC, California, Nevada, Pennsylvania, New York, New Jersey, and Delaware — see our service locations or contact us.

Related Resources

• PET SUV quantification
• EARL SUV harmonization
• PET/CT NEMA NU-2 performance testing
• PET/CT attenuation correction
• Time-of-flight PET imaging
• PET/CT and nuclear medicine physics
• Medical physicist consulting

References

1. Soret M, Bacharach SL, Buvat I. Partial-volume effect in PET tumor imaging. J Nucl Med. 2007;48(6):932-945. doi:10.2967/jnumed.106.035774. doi.org
2. Boellaard R, Delgado-Bolton R, Oyen WJG, et al. FDG PET/CT: EANM procedure guidelines for tumour imaging: version 2.0. Eur J Nucl Med Mol Imaging. 2015;42(2):328-354. doi:10.1007/s00259-014-2961-x. doi.org
3. Krempser AR, Ichinose RM, Miranda de Sá AMFL, Velasques de Oliveira SM, Carneiro MP. Recovery coefficients determination for partial volume effect correction in oncological PET/CT images considering the effect of activity outside the field of view. Ann Nucl Med. 2013;27(10):924-930. doi:10.1007/s12149-013-0773-x. doi.org
4. Anouan KJ, Lelandais B, Edet-Sanson A, Ruan S, Vera P, Gardin I, Hapdey S. 18F-FDG-PET partial volume effect correction using a modified recovery coefficient approach based on functional volume and local contrast. Q J Nucl Med Mol Imaging. 2017;61(3):301-313. doi:10.23736/S1824-4785.17.02756-X. doi.org
5. Kafkaletos A, Mix M, Sachpazidis I, et al. The significance of partial volume effect on the estimation of hypoxic tumour volume with [18F]FMISO PET/CT. EJNMMI Phys. 2024;11(1):43. doi:10.1186/s40658-024-00643-1. doi.org
6. Grecchi E, O'Doherty J, Veronese M, Tsoumpas C, Cook GJ, Turkheimer FE. Multimodal partial-volume correction: application to 18F-fluoride PET/CT bone metastases studies. J Nucl Med. 2015;56(9):1408-1414. doi:10.2967/jnumed.115.160598. doi.org
7. Jentzen W, Freudenberg L, Bockisch A. Quantitative imaging of 124I with PET/CT in pretherapy lesion dosimetry: effects impairing image quantification and their corrections. Q J Nucl Med Mol Imaging. 2011;55(1):21-43. PubMed
8. Chomet M, Schreurs M, Vos R, et al. Performance of nanoScan PET/CT and PET/MR for quantitative imaging of 18F and 89Zr as compared with ex vivo biodistribution in tumor-bearing mice. EJNMMI Res. 2021;11(1):57. doi:10.1186/s13550-021-00799-2. doi.org
9. Rezaei S, Ghafarian P, Jha AK, Rahmim A, Sarkar S, Ay MR. Joint compensation of motion and partial volume effects by iterative deconvolution incorporating wavelet-based denoising in oncologic PET/CT imaging. Phys Med. 2019;68:52-60. doi:10.1016/j.ejmp.2019.10.031. doi.org
10. National Electrical Manufacturers Association. NEMA NU 2: Performance Measurements of Positron Emission Tomographs (PET). Rosslyn, VA: NEMA. nema.org
11. EANM Research Ltd (EARL). EARL FDG-PET/CT accreditation program and recovery-coefficient specifications. earl.eanm.org