Alzheimer's Imaging Consortium.

Area of Science:

• Neuroimaging
• Biomarkers
• Alzheimer's Disease Research

Background:

• The Centiloid framework is widely adopted for harmonizing amyloid-PET quantification and facilitating multi-cohort data comparison.
• It aids in standardizing amyloid imaging interpretation across research and clinical settings.

Purpose of the Study:

• To evaluate the global implementation of the Centiloid framework by comparing Centiloid distributions and positivity thresholds across diverse cohorts.
• To establish standardized cutoffs for amyloid-PET positivity using meta-analysis.

Main Methods:

• Data from 37 publicly available cohorts (n=41,678) were collected, including Centiloid values, demographics, and visual reads.
• Gaussian mixture models (GMM) were used to determine Centiloid peaks and cutoffs (mean + 2SD of the lower Gaussian) for each cohort.
• Random effects meta-analyses combined data across cohorts, with subgroup analyses examining effects of processing pipelines, radiotracers, and study settings. Correspondence with visual reads was assessed using Cohen's kappa.

Main Results:

• A common Centiloid peak was identified at 1 CL, with significant heterogeneity in the second peak (meta-analysis outcome: 64 CL).
• The meta-analysis yielded a GMM-based cutoff of 19 CL (95% CI: 16-21 CL), with no significant differences across radiotracers or study settings.
• A visual read-based cutoff of 24 CL (95% CI: 21-27 CL) showed strong correspondence with Centiloids (kappa=0.86), but high inter-study heterogeneity (I²>80%) was observed.

Conclusions:

• Meta-analysis-derived Centiloid cutoffs are consistent with existing literature.
• Significant heterogeneity across studies highlights the need to investigate contributing factors and raises concerns about the universal application of common cutoffs.