Crunching the numbers. Responsibly.
PURPOSE: Positron emission tomography combined with magnetic resonance imaging (PET/MR) has not yet achieved the level of adoption of PET/CT. This study aimed to harmonise PET imaging protocols across a national PET/MR network and to quantitatively assess whether PET/MR can achieve reliability comparable to PET/CT. While previous PET test-retest studies have demonstrated good repeatability, they have typically been limited to small cohorts or restricted site configurations. METHODS: We conducted a multi-site harmonisation and rigorous test-retest study across the network of eight PET/MR scanners. Thirty-seven healthy older participants (65-90 years) underwent harmonised one-hour amyloid PET/MR scans using either [[Formula: see text]F]flutemetamol or [[Formula: see text]F]florbetaben on two occasions. Retest scans were performed under conditions of same-site repeatability or multi-site reproducibility. Harmonised acquisition and reconstruction protocols were applied, and amyloid burden was quantified on the Centiloid (CL) scale. RESULTS: CL values across 74 scans showed excellent test-retest agreement (ICC = 0.968), improving to 0.987 after exclusion of one attenuation correction related outlier. Mean test-retest variability was 2.58%. No statistically significant differences were observed across repeatability versus reproducibility conditions, scanner types, or tracers. CL measurements were highly consistent with three independent blinded visual reads. CONCLUSION: This study demonstrates that harmonised PET/MR achieves high reliability comparable to PET/CT. Although the accuracy of attenuation maps requires checks, this study supports the use of PET/MR for quantitative amyloid imaging in research and therapeutic trials, and provides a valuable open resource of image and raw PET/MR data for further methodological development.
CL values across 74 scans showed excellent test-retest agreement (ICC = 0.968), improving to 0.987 after exclusion of one attenuation correction related outlier.
direct repeatability estimates support the direction, though precision intervals and prospective power planning are absent
No statistically significant differences were observed across repeatability versus reproducibility conditions, scanner types, or tracers.
small uneven subgroups and non-significance cannot establish equivalence across conditions, vendors, or tracers
This study demonstrates that harmonised PET/MR achieves high reliability comparable to PET/CT.
the comparison is indirect because the study includes no head-to-head PET/CT arm
CL measurements were highly consistent with three independent blinded visual reads.
high agreement and kappa support the claim, with unexplained restriction from 74 to 72 scans
Derived from the full evaluation — not a separate score.
Strengths
The study implements common PET/MR acquisition and reconstruction procedures across eight sites, two vendors, and two amyloid tracers, including same-site and cross-site retest conditions.
↳ Methods, opening study-design paragraphs; Results, Table 2
The full-cohort ICC of 0.968 is reported alongside the post-exclusion ICC of 0.987, and the attenuation-map cause of the outlier is investigated rather than silently removed.
↳ Results, Test-retest Reproducibility; Figure 1; Discussion, attenuation-correction paragraph
Three trained readers were blinded to participant identity and test–retest status, with complete reader agreement and visual–quantitative agreement statistics reported.
↳ Methods, Visual Assessment; Results, Figures 1 and 4
Limitations
The Abstract and Conclusion claim reliability comparable to PET/CT, but the study contains no direct PET/CT arm and the Discussion explicitly acknowledges that absence.
↳ Abstract, Conclusion; Discussion, limitations paragraph
Non-significant tests in small, uneven groups are used to describe consistency across conditions, scanners, sites, and tracers, without a prospective equivalence margin or precision target.
↳ Results, Table 2 and Figures 2–3; Discussion, paragraph 3
The paper reports 74 scans overall but 72 in Figure 4, while the mixed model uses 36 participants rather than the 37 completed pairs. These differences are not explicitly reconciled.
↳ Abstract; Methods, Statistical Analysis; Results, Figure 4
The eight-site, two-vendor design and transparent reporting of the attenuation-correction outlier support a useful network-level repeatability contribution. The primary ICC and TRV findings are credible for healthy older volunteers, but small subgroup cells, long and variable retest intervals, and absent precision planning limit claims of consistency across scanners, sites, and tracers. The main downward adjustment arises from the unqualified PET/CT-comparability claim, which is not identified by a direct comparator and conflicts with the limitation acknowledged in the Discussion. The study therefore supports conditional research use more securely than broad therapeutic-trial readiness.
Nabu’s assessment, alongside the field’s view.
Are you an author of this paper?
Sound3.3
Confidence mediumThe eight-site, two-vendor, two-tracer design adds network-level PET/MR repeatability and reproducibility evidence beyond smaller or more restricted prior studies. The advance remains bounded by 37 healthy participants and small condition-specific groups.
“across all eight DPUK sites equipped with hybrid PET/MR scanners”
Randomised retest allocation, harmonised acquisition, Centiloid processing, Bland–Altman analysis, and blinded multi-reader assessment support the primary repeatability estimate. No prospective precision target or ICC confidence interval is reported, retest intervals extend to 482 days, and small subgroup tests are interpreted too strongly.
“63.19 (112.59) [2–482]”
The study design and principal ICC and TRV results are readily traceable, but the headline PET/CT comparison is stronger than the indirect evidence permits. Unexplained 74-versus-72 scan accounting and 37-versus-36 participant accounting further reduce reporting precision.
“harmonised PET/MR achieves high reliability comparable to PET/CT”
The Discussion substantively compares prior test–retest studies and acknowledges attenuation correction, population, tracer, and reconstruction limitations. Its admission that no direct PET/CT comparison was performed conflicts with the unqualified PET/CT-comparability conclusion, binding the Positioning score at 2.8.
“the absence of a direct head-to-head comparison with PET/CT”
Caveats4 of 4 checks
The main repeatability results remain interpretable, but scan and participant counts are not consistently reconciled across analyses. The mixed-model reporting also contains an unexplained degrees-of-freedom issue for a two-level visual-read factor.
Ethics approval, ARSAC authorisation, and written consent are declared. Data-access wording is inconsistent between controlled-access procedures and claims that the resource is openly available.
Flags: 3 declared / 5 total
48 references in manuscript 48 of 48 checkable references found in an index the reference list may be incompletely parsed; counts are indicative
No retraction notice found in Retraction Watch.
Sources: Retraction Watch ✓
Where this paper’s evidence sits on the path from initial observation to real-world use.
The protocols were implemented across operating sites and vendors, placing the work beyond a single-centre feasibility study. Healthy volunteers, omitted TOF reconstruction, and no direct PET/CT comparison prevent deployment-level validation for therapeutic trials.
“absence of a direct head-to-head comparison with PET/CT”
AI-generated, human-governed. Something look off? Contact us to request a review.