Comparison of two preprocessing methods for ¹⁸F-Flortaucipir PET quantification in Alzheimer's disease

Eur J Nucl Med Mol Imaging. 2025 Jul 19. doi: 10.1007/s00259-025-07452-3. Online ahead of print.

ABSTRACT

BACKGROUND: Tau-Positron Emission Tomography (PET) has become central in Alzheimer's disease (AD) research and clinical settings. Multiple preprocessing pipelines for tau-PET quantification have been described, with satisfactory performance but direct comparisons remain scarse. Our study evaluates the comparability of two commonly used PET preprocessing methods, respectively in native and standard spaces, in quantifying tau deposition and in their ability to discriminate AD patients.

METHODS: 209 subjects were included from the Geneva memory clinic including cognitively unimpaired (CU) individuals, mild cognitive impairment (MCI) and dementia patients. Images were processed in native and standard space using inferior cerebellar grey matter as reference region. Standardized uptake value ratios (SUVR) were extracted from AD-specific regions. Correlations between SUVR obtained by different methods and plasma biomarkers were assessed. ROC analyses compared the ability of the two methods to discriminate visually assessed tau status, amyloid-positive cognitively impaired from amyloid-negative CU, and subjects with declining cognition over time.

RESULTS: SUVR from the two methods were strongly correlated across all regions. However, SUVR values obtained with standard space method showed higher values. SUVR in the medial temporal lobe from native space processing provided a greater accuracy in discriminating positive scans and identifying subjects with cognitive decline. For all other analyses methods performed equally well. The correlation with plasma biomarkers was comparably high with both methods.

CONCLUSION: While preprocessing in native and standard space is adequate for quantifying ¹⁸F-Flortaucipir PET and for discriminating AD patients, higher accuracy can be obtained in the mesial temporal regions and to predict cognitive decline using processing in native space.

PMID:40681899 | DOI:10.1007/s00259-025-07452-3