HAI Book 2025 - Flipbook - Page 608
Tissot, Cécile
142
Best practices for tau-PET tracer harmonization in Alzheimer's
research
Cécile Tissot1, Hsin-Yeh Tsai1, Joseph Giorgio2, Ganna Blazhenets3, Theresa Harrison2, Emily
Olafson4, Sandra Sanabria4, Antoine Leuzy5, Vincent Doré6, William Jagust1,2, Pedro RosaNeto7, Tharick Pascoal8, Suzanne Baker1
1
Lawrence Berkeley National Laboratory, Berkeley, CA, US
University of California, Berkeley, Berkeley, CA, US
3
University of California, San Francisco, San Francisco, CA, US
4
Genentech, South San Francisco, CA, US
5
Lund University, Lund, SE
6
Australian e-Health Research Centre, Melbourne, AU
7
McGill University, Montreal, QC, CA
8
University of Pittsburgh, Pittsburgh, PA, US
2
Background: Harmonization of tau-PET tracers is critical for consistency in Alzheimer9s disease (AD) research.
With variability in tracers and their tau pathology affinities, standardized practices are needed to ensure
comparability across studies. This consensus paper, based on discussions with PET imaging experts, aims to
establish best practices for tracer harmonization, focusing on key performance metrics to be reported.
Methods: Discussions emphasized selecting appropriate metrics to assess harmonization quality. Key metrics
include R² values, effect sizes between groups (e.g., A´+ vs. A´-), and test-retest reliability. The voxel-wise versus
region-of-interest (ROI) approach was debated, especially for addressing regional asymmetries in tau deposition.
The importance of non-linear models to capture early-stage tau accumulation and reduce bias in intermediate tau
groups was discussed, along with the relevance of cognitive outcomes for clinical trials.
Results: It was agreed that analyses should include a training dataset to develop harmonization methods, then
apply these to a test dataset. In addition to R², experts recommend reporting effect sizes and harmonized data
slopes as standard metrics to compare tracers and populations and assess harmonization effectiveness. Nonlinear models were favored for capturing early tau accumulation and preserving dynamic ranges, particularly in
intermediate-to-high tau individuals. Effect sizes and correlations with cognition should be reported before and
after transformation. Voxel-wise harmonization was preferred for maintaining regional specificity, though ROIbased methods might suffice for advanced tau stages. Selected ROIs include Braak stages, meta-temporal and
whole cortex. Longitudinal stability was emphasized, considering the need for simple yet effective harmonization
methods that preserve disease-relevant signals.
Discussion: Experts stressed the importance of reporting key harmonization metrics, including R² values and
effect sizes, to ensure robust cross-tracer comparisons. Voxel-wise approaches offer precision but must remain
flexible, depending on the research question. This supports both academic research and industry applications,
promoting effective data integration in AD studies.
Keywords: harmonization, consensus, tau-PET
HAI2025 - 608