HAI Book 2025 - Flipbook - Page 397
Lutz, Olivia
97
Evaluation of the Uni-tau and CenTauR frameworks in new datasets
for tau PET harmonization in Alzheimer9s disease
Olivia Lutz1, Emily Olafson1, Matteo Tonietto2, Gregory Klein2, Tharick Pascoal3, Guilherme
Negrini3, Guilherme Povala3, Antoine Leuzy4, Lars Lau Raket5, Nick Doren2, Sandra Sanabria
Bohorquez1, Edmond Teng1
1
Research and Early Development (gRED), Genentech, Inc, South San Francisco, CA, US
Research and Early Development (pRED), Hoffmann-La Roche, Basel, CH
3
Department of Psychiatry, University of Pittsburgh, Pittsburgh, PA, US
4
Critical Path for Alzheimer’s Disease (CPAD) Consortium, Critical Path Institute, Tucson, AZ, US
5
Eli Lilly and Company, Indianapolis, IN, US
2
Introduction: Several tau-PET tracers are used in Alzheimer9s disease (AD) research and therapy, but SUVR
metrics are not directly comparable across tracers. To date, two methods, CenTauR and Uni-tau, have been
developed to transform SUVR values into a harmonized tau-PET scale. Neither has been evaluated in independent
data. Our objective was to evaluate the CenTauR and Uni-tau scale by evaluating whether tracer bias was
detectable in harmonized data.
Methods: [18F]FTP tau-PET images from the Alzheimer's Disease Neuroimaging Initiative (ADNI) and [18F]MK-6240
tau-PET images from the Lantheus database were used to create matched samples of AD/A´+ and cognitively
unimpaired (CU)/A´- individuals (Table 1). Samples were matched on age, sex, and MMSE. SUVRs calculated using
the inferior cerebellum as a reference were converted to CenTauR/Uni-tau using equations for the meta temporal
region. Tracer effect was estimated with the F-statistic, comparing linear models with SUVR/CenTauR/Uni-tau as
the dependent variable with and without 8tracer9 as an independent variable, controlling for age/sex/MMSE. Lowhigh tau-PET cut-offs were determined by fitting Gaussian mixture models on the combined AD/A´+ and CU/A´samples.
Results: In SUVR, a strong tracer effect was detectable in the AD/A´+ and CU/A´- samples (Figure 1). In CenTauR
and Uni-tau, no tracer effect was detected in the AD/A´+ samples. In CU/A´- samples, there was a large tracer
effect for CenTauR (F = 56.5, p < 0.001) and a small tracer effect for Uni-tau (F = 4.46, p = 0.04) (Figure 1). Cut-offs
were more similar in Uni-tau than CenTauR (Table 2).
Conclusion: Our study found that while both harmonization methods remove tracer bias in AD/A´+ patients, Unitau appears to more effectively harmonize CU/A´- individuals than CenTauR. This was also indicated by more
consistent low-high tau-PET cut-offs using Uni-tau.
HAI2025 - 397
