Abstract
PURPOSE: Among visually Tau-PET-positive scans, large variation exists in the size of the visually tau-positive area. Here, we propose a metric quantifying the spatial extent of visual Tau-PET-positivity, termed "TAU-SPEX", and evaluate associations with visual read status, neurofibrillary tangle (NFT) pathology at autopsy, and cognition. METHODS: [(18)F]flortaucipir data from 1,645 participants (aged 71.9 ± 8.2 years, 50.3% females) from four cohorts were visually read as positive or negative. TAU-SPEX was calculated as the percentage of gray matter voxels with suprathreshold Tau-PET uptake (using a threshold identical to that used for visual reading) in a spatially unconstrained whole-brain mask. We additionally computed Tau-PET SUVr in a whole-brain and temporal meta-region. We tested the performance of TAU-SPEX for distinguishing visually tau-negative from tau-positive participants and for distinguishing participants with and without NFT Braak-V/VI pathology at autopsy (n = 18), and tested associations of TAU-SPEX with concurrent and longitudinal cognition. The performance of TAU-SPEX was compared to SUVr. RESULTS: TAU-SPEX demonstrated strong performance in distinguishing tau-negative from tau-positive participants (AUC: 0.97). Moreover, TAU-SPEX showed high accuracy, sensitivity, specificity, positive predictive value and negative predictive value (all > 0.90) for identifying tau-positive participants, and showed high sensitivity (87.5%) and specificity (100.0%) for identifying participants with NFT Braak-V/VI pathology. TAU-SPEX was moderately associated with concurrent (β=-0.36 [-0.29, -0.43], p < 0.001) and longitudinal (β=-0.19 [-0.15, -0.22], p < 0.001) cognition. Across analyses, TAU-SPEX generally outperformed SUVr. CONCLUSION: TAU-SPEX was strongly associated with visual read, NFT Braak-V/VI pathology and cognition, and might be useful in clinical settings as a potential adjunct to [(18)F]flortaucipir visual interpretation.