The Centiloid Scale for Amyloid PET

The Centiloid scale converts amyloid PET uptake from any tracer, scanner, or analysis pipeline onto a single 0-to-100 unit, anchored so that young amyloid-negative brains read near 0 and typical Alzheimer's disease reads near 100. It is the practical answer to a hard quantitation problem: four amyloid radiopharmaceuticals in common use, dozens of processing pipelines, and no way to compare the raw numbers until they are mapped onto a common axis.

Introduction

Amyloid PET has been remarkably consistent in its qualitative, visual findings across centers, but the quantitative numbers reported as outcome measures have varied widely. Some of that variation is trivial, such as the choice of units; some is scanner dependent; and of course different tracers yield different numbers for the same amyloid burden ¹. For a research program, a memory-clinic registry, or a multicenter trial, that variation is a real obstacle: a standardized uptake value ratio (SUVR) of 1.5 means one thing for Pittsburgh compound B and something quite different for florbetapir.

The Centiloid scale, introduced by the Centiloid Project working group in 2015, solves this by scaling the outcome of any particular analysis method or tracer onto a 0-to-100 range anchored by young controls and typical Alzheimer's disease (AD) patients ¹. This guide explains the physics and the arithmetic behind the scale, how tracer-specific conversion equations are built, what Centiloid thresholds mean against neuropathology, and what quality control a qualified medical physicist should expect before a site reports Centiloids clinically or in a trial.

DRPS supports PET/CT and nuclear medicine programs across Florida, Maryland, Virginia, Washington DC, California, Nevada, Pennsylvania, New York, New Jersey, and Delaware through its PET/CT and nuclear medicine physics and medical physics consulting services.

Topic Explanation

What problem does the Centiloid scale solve?

The Centiloid scale is a unit of standardized amyloid burden, not a new measurement. It does not change how a scan is acquired. It changes how the final number is expressed, so that quantitative amyloid burden can be compared across tracers, scanners, reconstruction methods, and analysis pipelines.

Four amyloid PET radiopharmaceuticals are widely used. The carbon-11 tracer Pittsburgh compound B (PiB) is the historical research standard but is limited by the 20-minute half-life of C-11 to sites with an on-site cyclotron. Three fluorine-18 tracers — florbetapir, florbetaben, and flutemetamol — can be produced centrally and shipped, which is why they dominate clinical practice ^{1, 6}. Each was approved by the FDA for visual estimation of amyloid plaque density in patients being evaluated for cognitive decline ^{7, 8, 9}.

The difficulty is that these tracers do not produce interchangeable numbers. They differ in nonspecific white-matter binding, in the optimal target and reference regions, and in dynamic range. A site that acquires florbetapir today and switches to flutemetamol next year cannot simply compare the old and new SUVR values. The Centiloid scale gives them a common currency.

For readers new to amyloid imaging in the clinic, our companion post on amyloid and tau brain PET covers the clinical interpretation, while this article focuses on the quantitative framework. The same harmonization philosophy underlies our discussion of EARL SUV harmonization for FDG PET.

From counts to SUVR

Every Centiloid value begins as an SUVR. The SUVR is the ratio of mean radiotracer concentration in a target region to the mean concentration in a reference region:

The standard Centiloid target region is a large, predominantly neocortical volume of interest, and the standard reference region is the whole cerebellum ^{1, 2}. The reference region matters enormously: it should contain little or no specific amyloid binding, so that its signal reflects nonspecific delivery and clearance. Dividing by it cancels much of the between-subject and between-scanner variation in injected activity, body size, and uptake time. The arithmetic is the same idea behind the standardized uptake value ratios used elsewhere in PET; see our primer on PET SUV quantification.

Because SUVR is a ratio of regional means from a single static scan, it is sensitive to how those regions are drawn, how the brain is spatially normalized to the template, and how the images were reconstructed. Those are exactly the steps a Centiloid pipeline must standardize.

Key Technical Principles

The anchoring equation

The core of the Centiloid method is a two-point linear calibration. The working group defined the scale so that the mean SUVR of a cohort of young, amyloid-negative controls (aged 45 years or younger, chosen so they are almost certainly amyloid negative) maps to 0 Centiloids, and the mean SUVR of a cohort of typical AD patients maps to 100 Centiloids ¹. For any individual scan analyzed with the standard PiB method, the Centiloid value is:

where SUVR(ind) is the individual's SUVR, SUVR(YC-0) is the mean of the young-control anchor group, and SUVR(AD-100) is the mean of the typical-AD anchor group. This is nothing more than a linear rescaling: subtract the zero point and divide by the span between the two anchors.

Substituting the published PiB anchor values gives the standard PiB equation ²:

A worked example

Suppose a standard PiB analysis returns a cortical SUVR of 1.50 against the whole-cerebellum reference. Then:

A value of about 46 Centiloids sits well above the "no neuritic plaque" range and within the range that neuropathology studies associate with at least moderate plaque burden, which we quantify below. The same SUVR of 1.50 measured with a different tracer would convert to a different Centiloid value, because each tracer has its own equation.

Building a tracer-specific equation

A fluorine-18 tracer is placed on the scale by scanning subjects with both that tracer and PiB, then regressing the two SUVR measurements and propagating the result through the PiB calibration ¹. This is why the published conversion equations differ by tracer and by the exact volumes of interest used.

The crucial physics point is that the conversion equation is tied to a specific pipeline: a specific target VOI, a specific reference region, a specific spatial-normalization template, and a specific reconstruction. Applying florbetapir's equation to a flutemetamol SUVR, or applying a whole-cerebellum equation to a pons-referenced SUVR, produces a wrong Centiloid value even though the arithmetic "runs." A Centiloid value is only meaningful if the entire chain that produced the SUVR matches the chain used to derive the equation.

Why the reference region and partial-volume effects matter

Because SUVR is a ratio, any error in the reference-region signal scales the whole result. Reference regions that are small, poorly placed, or contaminated by spill-in from adjacent high-uptake tissue (for example, cerebellar white matter or the occipital cortex) bias every Centiloid value computed from them. Partial-volume effects — the blurring of activity across the finite spatial resolution of the scanner — further distort both target and reference signals, and they differ between scanners with different resolution and reconstruction. For a deeper treatment, see our post on the PET partial-volume effect.

This is also why a Centiloid pipeline cannot be validated on image appearance alone. The only way to know a pipeline is correct is to show that its young-control cohort reads near 0 CL with low variance and that its typical-AD cohort reads near 100 CL, reproducing the anchors of the method ¹.

Clinical Impact

Thresholds against neuropathology

The value of the Centiloid scale is that, because it is anchored and tracer-independent, thresholds derived in one study can be interpreted across tracers. A widely cited autopsy comparison examined Centiloid values against post-mortem neuritic-plaque density and clinicopathological diagnosis ³. Its findings give a practical interpretive framework:

These numbers come from a specific tracer set and pipeline, so a site should not treat them as a universal switch. They do, however, illustrate why the scale is useful: a "gray zone" roughly between 10 and 30 CL is a recognized region of uncertainty, and a quantitative value helps a reader understand where a borderline visual read sits.

Separately, a florbetapir standardization study reported that a threshold of about 24 Centiloids discriminated none-or-sparse from moderate-to-frequent plaques in autopsy-confirmed data, consistent with the general picture above ⁴.

Longitudinal and multicenter use

The strongest use case for Centiloids is change over time and comparison across sites. Because the scale is anchored, a 10-Centiloid change means the same thing whether it was measured with florbetapir in one trial and flutemetamol in another, provided both pipelines are properly calibrated. This is why disease-modifying-therapy trials and large observational cohorts increasingly report amyloid burden and amyloid change in Centiloids rather than tracer-specific SUVR. Amyloid PET is also used clinically to confirm the presence of amyloid pathology in patients being considered for amyloid-targeting therapies, and serial quantitative measurement is a natural fit for monitoring; the physics of making those serial numbers comparable is exactly what the Centiloid scale provides.

The scale supports, but does not replace, the visual read

For routine clinical interpretation, each F-18 amyloid tracer is approved for a visual read performed according to its prescribing information, and that visual read remains the basis of the clinical result ^{7, 8, 9}. The appropriate use criteria from the Society of Nuclear Medicine and Molecular Imaging and the Alzheimer's Association define the clinical scenarios in which amyloid PET is and is not appropriate, independent of quantitation ⁶. A Centiloid value adds reproducible, comparable context — especially for borderline scans, follow-up, and research — but it is not a substitute for the validated visual interpretation.

Practical Optimization Tips

A medical physicist supporting a Centiloid program should work through the pipeline in the same order the data flow.

1. Lock down the standard regions

Use the standard Centiloid target and reference volumes distributed by the Centiloid Project, not locally drawn regions, unless a local pipeline has been formally cross-calibrated and its own equation derived ^{1, 2}. Document the template, the target VOI, and the reference region.

2. Use the correct tracer-specific equation

Match the conversion equation to the exact tracer, VOI set, and reference region in use. If the site changes any element of the pipeline, the equation must be re-derived or re-validated; it is not transferable.

3. Validate against the anchors

Demonstrate that a young-control group reads near 0 CL with low variance and that typical-AD cases read near 100 CL. A young-control mean that drifts away from 0, or a large young-control variance, is the clearest early sign that spatial normalization or reference placement is wrong ¹.

4. Harmonize across scanners and reconstructions

Reconstruction algorithm, iterations, post-filter, and resolution modeling all change quantitative uptake. A site that reports Centiloids from more than one scanner, or that upgrades reconstruction software, should verify quantitative agreement before and after, using a common phantom and a fixed processing standard. This is the amyloid analog of the harmonization work routine for FDG SUV; the principles mirror those in EARL SUV harmonization.

5. Keep acquisition inside the validated window

Centiloid equations assume the scan was acquired in the tracer's standard post-injection window (for example, the 50-to-70-minute window used in several PiB and NAV4694 calibrations) ². Scanning outside that window changes the SUVR and invalidates the conversion.

Common pitfalls to avoid

• Mixing pipelines. Reporting a Centiloid value from one tracer's equation applied to another tracer's SUVR.
• Changing the reference region silently. Whole cerebellum, cerebellar gray, and pons give different SUVR; each needs its own equation.
• Treating thresholds as universal. Neuropathology-derived cutoffs depend on the study's tracer and pipeline.
• Ignoring reconstruction drift. A software upgrade can shift Centiloids without any change in the patient.
• Using Centiloids as a standalone diagnosis. The scale quantifies amyloid burden; it does not diagnose dementia, and it does not replace the approved visual read.

Regulatory Considerations

Amyloid PET radiopharmaceuticals are FDA-approved diagnostic drugs, and the Centiloid scale is a quantitative method layered on top of an approved visual-read indication, not a separate regulatory product. The three F-18 tracers — florbetapir (Amyvid), flutemetamol (Vizamyl), and florbetaben (Neuraceq) — are each approved for estimating amyloid neuritic-plaque density by PET in adults being evaluated for Alzheimer's disease and other causes of cognitive decline, and each prescribing information specifies a visual interpretation method ^{7, 8, 9}.

Quantitative reporting intended for clinical use should therefore be positioned as decision support alongside the approved read, with the pipeline, equation, and thresholds documented. For research and clinical-trial use, the SNMMI/Alzheimer's Association appropriate use criteria define when amyloid PET is indicated at all, which is logically prior to any quantitation ⁶.

From a radiation safety standpoint, amyloid PET uses the same byproduct-material framework as any other PET radiopharmaceutical. C-11 PiB requires an on-site cyclotron and PET radiopharmacy, while the F-18 agents are received as unit doses. Medical use of these byproduct materials is governed by 10 CFR Part 35 under the NRC or the equivalent Agreement State program: of the states DRPS serves, Florida, Maryland, Virginia, California, Nevada, Pennsylvania, New York, and New Jersey are NRC Agreement States that license medical use under their own radiation-control rules, while Washington DC and Delaware are regulated directly by the NRC. A site adding amyloid imaging should confirm its authorized users, dose calibrator QC, and PET imaging physics are in place as part of PET/CT and nuclear medicine physics support. For the underlying scanner performance that any quantitative program depends on, see PET/CT NEMA NU-2 performance testing and PET/CT ACR phantom QC.

Frequently Asked Questions (FAQs)

What is the Centiloid scale in amyloid PET?

The Centiloid scale is a standardized 0-to-100 unit for amyloid PET. It is anchored so that the average of young, amyloid-negative controls reads about 0 Centiloids and the average of typical Alzheimer's disease patients reads about 100 Centiloids. Any tracer or analysis method can be converted onto the scale, so quantitative amyloid burden can be compared across sites, scanners, and radiopharmaceuticals.

Why can't I just compare SUVR values between amyloid tracers?

Standardized uptake value ratio (SUVR) depends on the tracer, the reference region, the target region, the scanner, and the processing pipeline. The same brain can yield very different SUVR numbers with florbetapir versus flutemetamol or Pittsburgh compound B. The Centiloid scale removes that tracer and method dependence by mapping each pipeline's SUVR onto the same anchored 0-to-100 axis.

What Centiloid value is considered amyloid positive?

There is no single universal cutoff, but published neuropathology comparisons suggest that below about 10 Centiloids reflects essentially no neuritic plaque, above about 20 Centiloids indicates at least moderate plaque density, and roughly 50 Centiloids or more best confirms a neuropathological diagnosis of Alzheimer's disease. The exact threshold a site uses should match its tracer, pipeline, and clinical question.

How is a Centiloid value actually calculated?

First the pipeline computes an SUVR using the standard cortical target region and a standard reference region such as the whole cerebellum. Then that SUVR is inserted into a tracer-specific linear equation derived against the Pittsburgh compound B standard. The result is a Centiloid value. The equation differs by tracer and reference region, so it must be the validated equation for that exact pipeline.

Does the Centiloid scale replace the visual read?

No. For clinical interpretation, amyloid PET is read visually according to the tracer's approved prescribing information, and the quantitative value supports but does not replace that read. The Centiloid scale is most valuable for research, longitudinal follow-up, multicenter trials, and harmonizing quantitative reporting, and it should be used alongside, not instead of, the validated visual interpretation.

What quality control does a Centiloid pipeline need?

The pipeline must reproduce the standard Centiloid regions and reference region, use the correct tracer-specific conversion equation, and demonstrate that its young-control values fall near 0 Centiloids with low variance. Spatial normalization, reference-region placement, scanner calibration, and reconstruction settings all affect the result, so each should be documented and periodically verified by a qualified medical physicist.

Can Centiloid values be compared across scanners and reconstructions?

Only if each scanner and reconstruction is harmonized. Different reconstruction algorithms, filters, and resolution models change quantitative uptake, so a site that reports Centiloids across multiple scanners should verify agreement using a common phantom and processing standard, in the same spirit as EARL-style harmonization for FDG PET SUV.

Key Takeaways

• The Centiloid scale is an anchored unit, not a new measurement. It rescales a tracer's SUVR so that young controls read ~0 and typical AD reads ~100 CL.
• SUVR is tracer- and pipeline-dependent; Centiloids are not — if done correctly. The conversion equation is tied to a specific target region, reference region, template, and reconstruction.
• Each tracer needs its own validated equation. PiB, florbetapir, florbetaben, flutemetamol, and NAV4694 have distinct conversions, and the reference region must match the equation.
• Neuropathology anchors interpretation. Below ~10 CL means no neuritic plaque; above ~20 CL means at least moderate plaque; ~50 CL or more best confirms AD.
• Validation means hitting the anchors. A correct pipeline puts young controls near 0 CL with low variance.
• Centiloids support the visual read; they do not replace it, and they are only comparable across scanners after harmonization.

Conclusion

The Centiloid scale is a small piece of arithmetic resting on a careful piece of physics. The arithmetic — a two-point linear rescaling anchored at 0 and 100 — is easy. The physics is where programs succeed or fail: the reference region, the spatial normalization, partial-volume behavior, the reconstruction, and the acquisition window all feed the SUVR that the arithmetic then rescales. Get the pipeline right, and a site can report amyloid burden that is comparable across tracers, scanners, and years, which is exactly what longitudinal follow-up and multicenter work demand. Get it wrong, and the Centiloid value is a precise-looking number built on an unvalidated foundation.

For nuclear medicine programs standing up quantitative amyloid imaging, the practical message is to treat the Centiloid pipeline as an instrument that must be commissioned and QC'd, not a spreadsheet formula. The anchors are the acceptance test: young controls near 0, typical AD near 100.

How DRPS Can Help

Diagnostic Radiation Physics Services helps PET/CT and nuclear medicine programs implement defensible quantitative imaging. For amyloid PET, this can include reviewing the Centiloid pipeline and conversion equations, verifying scanner calibration and reconstruction harmonization, supporting EARL-style quantitative QC, confirming dose calibrator and PET performance testing, and documenting the program for research or clinical use through our PET/CT and nuclear medicine physics and medical physics consulting services.

DRPS supports facilities across our service locations, including Florida, Maryland, Virginia, Washington DC, California, Nevada, New York, Pennsylvania, New Jersey, and Delaware. If your site is adding amyloid imaging or standardizing quantitative reporting, contact us to review the pipeline before the first clinical or trial read.

Related Resources

• Amyloid and tau brain PET
• PET SUV quantification
• EARL PET SUV harmonization
• PET partial-volume effect
• PET/CT NEMA NU-2 performance testing
• PET/CT ACR phantom QC
• PET/CT and nuclear medicine physics services
• Medical physicist consulting

References

1. Klunk WE, Koeppe RA, Price JC, et al. The Centiloid Project: standardizing quantitative amyloid plaque estimation by PET. Alzheimers Dement. 2015;11(1):1-15.e1-4. doi:10.1016/j.jalz.2014.07.003. PubMed
2. Rowe CC, Jones G, Doré V, et al. Standardized Expression of 18F-NAV4694 and 11C-PiB β-Amyloid PET Results with the Centiloid Scale. J Nucl Med. 2016;57(8):1233-1237. doi:10.2967/jnumed.115.171595. PubMed
3. Amadoru S, Doré V, McLean CA, et al. Comparison of amyloid PET measured in Centiloid units with neuropathological findings in Alzheimer's disease. Alzheimers Res Ther. 2020;12(1):22. doi:10.1186/s13195-020-00587-5. PubMed
4. Navitsky M, Joshi AD, Kennedy I, et al. Standardization of amyloid quantitation with florbetapir standardized uptake value ratios to the Centiloid scale. Alzheimers Dement. 2018;14(12):1565-1571. doi:10.1016/j.jalz.2018.06.1353. PubMed
5. Bourgeat P, Doré V, Rowe CC, et al. A universal neocortical mask for Centiloid quantification. Alzheimers Dement (Amst). 2023;15(3):e12457. doi:10.1002/dad2.12457. PubMed
6. Johnson KA, Minoshima S, Bohnen NI, et al. Appropriate use criteria for amyloid PET: a report of the Amyloid Imaging Task Force, the Society of Nuclear Medicine and Molecular Imaging, and the Alzheimer's Association. J Nucl Med. 2013;54(3):476-490. doi:10.2967/jnumed.113.120618. PubMed
7. U.S. Food and Drug Administration. Amyvid (florbetapir F 18 injection) — Drug Approval Package (NDA 202008). 2012. accessdata.fda.gov
8. U.S. Food and Drug Administration. Vizamyl (flutemetamol F 18 injection) prescribing information (NDA 203137). 2013. accessdata.fda.gov
9. U.S. Food and Drug Administration. Neuraceq (florbetaben F 18 injection) — Approval Package (NDA 204677). 2014. accessdata.fda.gov
10. Global Alzheimer's Association Interactive Network (GAAIN). The Centiloid Project standard analysis data and regions. gaain.org