PET Randoms, Dead Time, and NECR

PET count-rate performance is a tug-of-war between physics that help and physics that hurt. True coincidences build the image and grow in proportion to activity; random coincidences carry no useful position information and grow with the square of activity; and dead time throws away counts exactly when the detectors are busiest. The Noise-Equivalent Count Rate (NECR) folds all three into one figure of merit that predicts image signal-to-noise — and it explains why injecting more activity is not always injecting more signal.¹²³

Introduction

Every PET image is built from coincidence events: two 511 keV annihilation photons detected close enough in time to be assigned to a line of response between the two detectors. In an ideal world, every recorded coincidence would be a true — both photons from the same annihilation, neither scattered — and more injected activity would simply mean more counts and a cleaner image.³

The real world is messier. Some coincidences are scattered, so the line of response is mispositioned. Some are random, two photons from two unrelated annihilations that happen to land in the same timing window, carrying no real spatial information at all. And at high count rates the detectors and electronics cannot keep up: dead time causes counts to be lost or piled up. Scatter and randoms add background variance; dead time removes signal. Push the activity higher and these effects grow faster than the useful signal does.³⁷

The medical physics community needed one number that captures the net effect of all of this on image quality, and that number is the NECR. This article explains the three competing processes — trues versus scatter and randoms, and dead time — defines the NECR and shows how it is computed, works through the count-rate math, and connects it to the practical questions physicists and technologists actually face: how much activity to inject, why big patients are hard, and what the NEMA NU 2 count-rate test is really telling you. DRPS supports this work through PET/CT and nuclear medicine physics and accreditation support across Florida, Maryland, Virginia, Washington DC, California, and Nevada.

Topic Explanation

The three kinds of coincidence

A PET system records a coincidence whenever two events are detected within a short coincidence timing window of each other. Those coincidences fall into three physically distinct categories:³

Trues are what you want. Scatter and randoms are background: they add events that do not correspond to a real annihilation along that line of response, and even after correction they leave behind added statistical noise. The whole game of count-rate performance is maximizing trues relative to the variance contributed by scatter and randoms — while dead time quietly erodes the trues you actually record.³

For the correction that handles scattered events specifically, see our companion discussion in PET/CT attenuation correction; this article focuses on randoms and dead time, the two effects that dominate at high count rate.

Why randoms are the villain at high activity

The defining feature of random coincidences is how fast they multiply. Because a random requires two independent single events to fall in the same window, the random rate depends on the product of two singles rates — and singles scale with activity, so randoms scale with the square of activity. Trues, coming from single annihilations, scale only linearly. The consequence is stark: doubling the activity roughly doubles the trues but roughly quadruples the randoms.³⁷

That is why the randoms-to-trues ratio climbs relentlessly with activity, and why simply injecting more radiopharmaceutical eventually makes the data worse, not better.

Key Technical Principles

The randoms rate equation

For two detectors with singles rates $S_1$ and $S_2$ and a coincidence timing window (resolving time) $\tau$, the rate of random coincidences between them is:

$$R=2\tau S_1{S}_2$$

The factor of $2$ arises because either photon can arrive first within the window of width $\tau$ on each side. Since $S_1$ and $S_2$ each scale with activity $A$, the random rate scales as $A^2$, while the true rate scales as $A$.³⁷

A worked illustration makes the scaling concrete. Take a representative narrow coincidence window for a modern LSO/LYSO system, $\tau =4\text{ns}$, and two detector elements each seeing a singles rate of $1.0\times 10^6\text{cps}$:

$$R=2\times (4\times 10^{-9}\text{s})\times (1.0\times 10^6\text{cps})\times (1.0\times 10^6\text{cps})=8.0\times 10^3\text{cps}$$

Now double the activity, so each singles rate doubles to $2.0\times 10^6\text{cps}$:

$$R=2\times (4\times 10^{-9})\times (2.0\times 10^6)\times (2.0\times 10^6)=3.2\times 10^4\text{cps}$$

The activity doubled, but the randoms rate quadrupled. This is the quadratic penalty in action, and it is why the timing window matters so much: halving $\tau$ halves the randoms rate directly. Narrow coincidence windows and good time-of-flight timing resolution — for example, a measured TOF timing resolution of about 527 ps FWHM on a modern LSO time-of-flight system — are among the most powerful tools for suppressing randoms.⁶

Correcting randoms: delayed window versus singles

Two methods dominate randoms estimation, and the choice affects the NECR:³

• Delayed coincidence window. A second timing window is opened, shifted in time by an interval much greater than $\tau$, so it can only record coincidences between unrelated events — i.e., randoms. Subtracting this delayed estimate from the prompt data removes the randoms bias accurately, but because both the prompt and delayed measurements carry Poisson noise, the subtraction adds statistical variance.
• Singles-based estimation. The randoms rate is computed from the measured singles rates and $\tau$ using $R=2\tau S_1{S}_2$. Because singles are counted at very high rates, this estimate is statistically smoother (lower variance) than the delayed-window method, at the cost of relying on an accurate model of the timing window.

Dead time: paralyzable and non-paralyzable

After a detector registers an event, it needs time to process it. During that interval it may miss or distort a second event. This is dead time, and it is modeled two ways. For a true event rate $n$, a measured rate $m$, and a characteristic dead time ${\tau }_d$:³

$$m=n{e}^{-n{\tau }_d}\text{(paralyzable)}$$

$$m=\frac{n}{1+n{\tau }_d}\text{(non-paralyzable)}$$

In the non-paralyzable model, each event is followed by a fixed dead interval and the measured rate saturates toward $1/{\tau }_d$. In the paralyzable model, an event arriving during the dead interval extends it, so at very high rates the measured rate actually peaks and then falls — the detector "paralyzes." Real PET front ends behave somewhere between these idealizations, but both capture the essential point: at high activity, the recorded true rate no longer tracks the true rate, and count losses set in.³⁷

Putting it together: the Noise-Equivalent Count Rate

The NECR was introduced to relate count rate to image signal-to-noise in a single number. It is the count rate that a hypothetical scanner with no scatter and no randoms would need in order to produce the same signal-to-noise ratio as the actual system. For true, scatter, and random rates $T$, $S$, and $R$, the NECR is:¹

$$\text{NECR}=\frac{T^2}{T+S+R}$$

This is the form used when the randoms estimate is essentially noiseless (for example, a low-noise singles-based estimate). When randoms are subtracted using the online delayed-coincidence window, the noise of that subtraction adds a second $R$ term:

$$\text{NECR}=\frac{T^2}{T+S+2R}$$

The $2R$ in the denominator is the statistical penalty for real-time delayed-window randoms subtraction, and it is why the correction method must be stated when NECR values are compared.¹³ Image variance scales inversely with NECR, so a system or protocol with a higher NECR yields a lower-noise reconstruction for the same acquisition.

A quick numerical example ties the pieces together. Suppose at a given activity a scanner records $T=300\text{kcps}$, $S=120\text{kcps}$, and $R=180\text{kcps}$. With a noiseless randoms estimate:

$$\text{NECR}=\frac{300^2}{300+120+180}=\frac{90,000}{600}=150\text{kcps}$$

With online delayed-window subtraction:

$$\text{NECR}=\frac{300^2}{300+120+2(180)}=\frac{90,000}{780}\approx 115\text{kcps}$$

The same raw data yields a lower effective count rate once the randoms-subtraction noise is included — a roughly 23% penalty in this example.

Why NECR peaks

Now combine the scaling laws. As activity rises, $T$ grows linearly (until dead time bends it over), while $R$ grows quadratically and $S$ grows roughly with $T$. The numerator $T^2$ climbs, but the denominator eventually climbs faster because of the $R$ (and dead-time losses in $T$). The result is a curve that rises, reaches a peak NECR, and then declines. The peak value and the activity concentration at which it occurs are the headline metrics of count-rate performance: they tell you the best signal-to-noise the scanner can deliver and the activity beyond which more is counterproductive.¹⁵⁷ On a modern total-body PET system, for example, peak NECR has been reported on the order of 2 Mcps under total-body imaging conditions, alongside a very high sensitivity of about 174 kcps/MBq.⁵

Clinical Impact

The NECR curve is not an academic abstraction — it sets the optimal injected activity and explains why some scans are inherently harder. Because NECR peaks and then falls, there is a most-favorable activity for a given scanner and patient; injecting beyond it raises patient and staff dose without improving, and possibly degrading, the effective signal-to-noise ratio.⁷

Patient size is the clearest clinical consequence. A larger patient attenuates more of the trues while still generating singles that drive randoms and dead time, so the NECR achievable in a heavy patient is substantially lower than in a small one. Analyses of clinical F-18 FDG data show NECR falling on the order of about 1.4% per kilogram of body weight, with the ratio of randoms to trues rising markedly across the adult weight range. The practical implication is that big patients are better served by longer acquisition times than by simply more activity — more activity mostly feeds the quadratic randoms term.⁷

This is also why count-rate performance connects to quantitation. If randoms and dead time are not accurately corrected, the standardized uptake values (SUVs) a clinic reports can drift, especially at high count rates or in bulky patients. Harmonization and accreditation programs such as EANM/EARL exist precisely to keep SUV recovery consistent across systems, and stable count-rate corrections are part of what makes that possible.⁸ For the downstream metric itself, see PET SUV quantification.

Practical Optimization Tips

Optimize activity to the scanner, not to habit

Use the scanner's count-rate response to choose activity, rather than a fixed historical value. The goal is to operate near — not far beyond — the NECR peak for your patient population, balancing image quality against dose and throughput.⁷

Scale scan time with patient size

For heavier patients, add acquisition time (or bed-position time) rather than piling on activity. Because NECR falls with weight, matching data quality across body sizes generally means longer scans for large patients.⁷

Exploit time-of-flight and a tight timing window

Good TOF timing resolution and a narrow coincidence window suppress the randoms rate and improve effective NECR. When evaluating or upgrading a system, timing resolution is a first-order determinant of high-count-rate performance.⁶

Know your randoms-correction method

Whether your system uses delayed-window or singles-based randoms estimation changes the noise it injects and the NECR expression that applies. Interpret vendor NECR specifications with the correction method in mind, and be consistent when comparing systems.¹³

Test count-rate performance at acceptance and track it

Measure count-rate performance at acceptance using NEMA NU 2 methodology and re-verify it periodically. AAPM Task Group 126 provides clinically practical acceptance and quality-assurance procedures for PET/CT built in the spirit of NEMA NU 2, using common phantoms and freely available software.²⁴

Common pitfalls to avoid

• Assuming more activity is always better. Past the NECR peak, extra activity adds dose and randoms, not signal.⁷
• Comparing NECR values without the correction method. The $2R$ penalty makes online-subtracted NECR lower than a noiseless estimate for the same data.¹
• Ignoring patient weight in protocol design. NECR falls with body mass; a one-size activity does not fit all.⁷
• Neglecting timing-window and TOF effects. A wider window or poorer timing resolution multiplies randoms.⁶
• Skipping count-rate testing at acceptance. Without a baseline, later count-rate drift is hard to detect.²

Regulatory Considerations

PET count-rate performance sits inside both a radioactive-materials framework and an imaging-accreditation framework. The positron-emitting radiopharmaceuticals used in PET are byproduct material, so their possession and medical use fall under 10 CFR Part 35 (or the equivalent Agreement State program), with dose limits under 10 CFR Part 20 — the same regulatory backbone that governs any nuclear medicine service.³ Of the states DRPS serves, Florida, Maryland, Virginia, California, Nevada, Pennsylvania, New York, and New Jersey are NRC Agreement States, while Washington DC and Delaware are regulated directly by the NRC for byproduct material.

The performance side is governed by measurement standards and accreditation. NEMA NU 2-2024, Performance Measurements of Positron Emission Tomographs, is the current standard that defines how spatial resolution, sensitivity, scatter fraction, count losses (dead time), randoms, and NECR are measured, using the 70-centimeter line-source count-rate phantom.² AAPM Task Group 126 translates those measurements into acceptance-testing and quality-assurance procedures for clinical PET/CT.⁴ Accreditation programs — the ACR PET/CT accreditation program in the United States, and EANM/EARL harmonization in Europe — expect documented performance and stable quantitation as a condition of accredited status.⁸ A qualified or board-certified medical physicist typically performs the acceptance testing and periodic surveys that verify count-rate performance and the associated corrections; for the broader survey, see our overview of PET/CT NEMA NU 2 performance testing.

Frequently Asked Questions (FAQs)

What is NECR in PET?

The Noise-Equivalent Count Rate (NECR) is a single figure of merit for PET count-rate performance. It is the count rate an ideal, background-free system would need to achieve the same signal-to-noise ratio as the real system once scatter and random coincidences are accounted for. NECR combines the true, scatter, and random coincidence rates into one number, and because image variance scales inversely with NECR, a higher NECR generally means a better-quality data set for a given acquisition.

Why do random coincidences increase faster than true coincidences?

A true coincidence comes from a single annihilation, so its rate scales linearly with activity. A random coincidence comes from two unrelated annihilations whose photons happen to arrive within the coincidence timing window, so its rate depends on the product of two independent singles rates and scales with the square of activity. As you inject more activity, randoms grow quadratically while trues grow only linearly, so the randoms fraction rises and eventually dominates.

What is dead time in a PET scanner?

Dead time is the interval after a detected event during which the detector and its electronics cannot fully process a second event. When the count rate is high, events overlap and some are lost or mispositioned, so the measured count rate falls below the true rate. Dead time is modeled as paralyzable or non-paralyzable behavior, and it is the reason the measured true-coincidence rate saturates and then declines at very high activity.

Why does NECR peak and then fall as activity increases?

At low activity, trues dominate and NECR rises with activity. As activity climbs, random coincidences grow with the square of activity and dead time begins discarding true counts, so the numerator of NECR grows more slowly while the denominator grows faster. The two effects overtake the linear gain in trues, so NECR reaches a peak and then declines. The peak NECR and the activity at which it occurs characterize the scanner's count-rate performance.

How are random coincidences corrected?

The two common methods are the delayed coincidence window and singles-based estimation. The delayed-window method opens a second timing window shifted far enough in time that it can only capture random events, then subtracts that estimate from the prompt data; it is accurate but adds statistical noise. Singles-based estimation computes the randoms rate from the measured singles rates and the timing window, which is less noisy. The correction method affects the NECR expression used.

How is count-rate performance measured?

Count-rate performance, including scatter fraction, count losses from dead time, randoms, and NECR, is measured using the NEMA NU 2 standard with a 70-centimeter line-source phantom filled with a decaying activity. Data are acquired repeatedly as the activity decays, and the true, scatter, random, and NEC rates are plotted against activity concentration. The AAPM Task Group 126 report provides acceptance-testing and quality-assurance procedures built in the spirit of NEMA NU 2 for clinical PET/CT systems.

Does a higher injected dose always improve PET image quality?

No. Because NECR peaks and then falls, injecting more activity past the peak does not improve, and can degrade, the effective signal-to-noise ratio, while increasing patient and staff dose. The optimal activity depends on the scanner's count-rate response and the patient, and heavier patients reach a lower NECR, which is why larger patients may need longer scan times rather than simply more activity.

Key Takeaways

• Three processes compete. Trues build the image and scale with activity; scatter and randoms add background; dead time removes counts at high rates.³
• Randoms scale quadratically. With $R=2\tau S_1{S}_2$, doubling activity roughly quadruples randoms while only doubling trues.³⁷
• Dead time bends the true rate over. Paralyzable and non-paralyzable models both predict that recorded trues saturate or fall at high activity.³
• NECR is the figure of merit. $\text{NECR}=T^2/(T+S+R)$, or $T^2/(T+S+2R)$ with online delayed-window randoms subtraction; image variance scales inversely with NECR.¹
• NECR peaks. More activity past the peak adds dose and randoms, not signal — a direct argument for count-rate-aware protocols.⁷
• Measure and accredit. NEMA NU 2-2024 defines the count-rate test; AAPM TG-126 and accreditation programs turn it into practice.²⁴⁸

Conclusion

The NECR is one of the most useful numbers in PET physics because it compresses a genuinely complex interplay — linear trues, quadratic randoms, and rate-dependent dead time — into a single quantity that tracks image signal-to-noise. Understanding it changes how a program behaves: it reframes "how much activity should we inject?" from a habit into a count-rate-aware decision, it explains why large patients need longer scans rather than bigger doses, and it gives acceptance testing a defensible metric to measure and trend.

A strong PET program treats count-rate performance as something to be measured at acceptance, verified over time, and used in protocol design — not as a specification that lives only in the vendor's brochure. When randoms, dead time, and NECR are understood and monitored, injected activity, scan time, and image quality can be balanced deliberately, to the benefit of both image quality and radiation safety.

How DRPS Can Help

Diagnostic Radiation Physics Services supports PET/CT and nuclear medicine programs with acceptance testing and annual performance surveys, NEMA NU 2 count-rate and NECR evaluation, scatter and randoms correction verification, SUV calibration and harmonization checks, protocol and injected-activity optimization, and the documentation that accreditation requires — delivered by board-certified medical physicists through our PET/CT and nuclear medicine physics and accreditation support services.

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

A strong PET program is not just about passing a count-rate test. It is about using the count-rate physics to inject the right activity, scan for the right time, and defend the image quality behind every SUV.

Related Resources

• PET/CT NEMA NU 2 performance testing
• PET/CT attenuation correction
• PET/CT daily QC and calibration
• Time-of-flight PET imaging
• PET spatial resolution and positron range
• PET/CT and nuclear medicine physics
• Accreditation support

References

1. Strother SC, Casey ME, Hoffman EJ. Measuring PET scanner sensitivity: relating countrates to image signal-to-noise ratios using noise equivalent counts. IEEE Trans Nucl Sci. 1990;37(2):783-788. doi:10.1109/23.106715. doi.org
2. National Electrical Manufacturers Association. NEMA NU 2-2024: Performance Measurements of Positron Emission Tomographs (PETs). Rosslyn, VA: NEMA; 2024. nema.org
3. Tarantola G, Zito F, Gerundini P. PET instrumentation and reconstruction algorithms in whole-body applications. J Nucl Med. 2003;44(5):756-769. PubMed
4. Lopez BP, Jordan DW, Kemp BJ, Kinahan PE, Schmidtlein CR, Mawlawi OR. PET/CT acceptance testing and quality assurance: Executive summary of AAPM Task Group 126 Report. Med Phys. 2021;48(2):e31-e35. doi:10.1002/mp.14656. doi.org
5. Spencer BA, Berg E, Schmall JP, et al. Performance evaluation of the uEXPLORER total-body PET/CT scanner based on NEMA NU 2-2018 with additional tests to characterize PET scanners with a long axial field of view. J Nucl Med. 2021;62(6):861-870. doi:10.2967/jnumed.120.250597. doi.org
6. Jakoby BW, Bercier Y, Conti M, Casey ME, Bendriem B, Townsend DW. Physical and clinical performance of the mCT time-of-flight PET/CT scanner. Phys Med Biol. 2011;56(8):2375-2389. doi:10.1088/0031-9155/56/8/004. doi.org
7. Watson CC, Casey ME, Bendriem B, et al. Optimizing injected dose in clinical PET by accurately modeling the counting-rate response functions specific to individual patient scans. J Nucl Med. 2005;46(11):1825-1834. PubMed
8. Kaalep A, Sera T, Oyen W, et al. EANM/EARL FDG-PET/CT accreditation — summary results from the first 200 accredited imaging systems. Eur J Nucl Med Mol Imaging. 2018;45(3):412-422. doi:10.1007/s00259-017-3853-7. doi.org
9. American Association of Physicists in Medicine. PET/CT Acceptance Testing and Quality Assurance. AAPM Report No. 126 (Task Group 126). College Park, MD: AAPM; 2019. aapm.org
10. U.S. Nuclear Regulatory Commission. 10 CFR Part 35: Medical Use of Byproduct Material. ecfr.gov