Brain metabolism model predicts Alzheimer tau in humans
Trained on 1,088 participants, the machine learning framework predicted flortaucipir-PET with R² values of 0.34 to 0.71 and classified tau stages with 79% to 95% accuracy.
In a study of 1,088 human participants spanning normal aging and Alzheimer’s disease and related disorders, researchers developed a machine learning framework to predict tau pathology. The framework integrates a computational model of brain function with FDG-PET imaging to estimate Alzheimer-type tau accumulation measured by flortaucipir-PET. The authors validated the model across three groups: a longitudinal subset of 215 participants, 295 individuals from the Alzheimer’s Disease Neuroimaging Initiative, and 264 individuals from a heterogeneous clinical cohort. Across participant-level tests, the model predicted flortaucipir-PET retention with R² values ranging from 0.34 to 0.71. It also separated low from high tau stages with 79% to 95% accuracy and estimated tau burden in clinical patients lacking tau PET scans.
Why it matters
Tau accumulation is a major driver of neurodegeneration in Alzheimer’s disease and cognitive aging. Linking metabolic PET imaging to tau distribution helps illuminate how functional brain changes correspond to pathological protein spread.
Caveats
Prediction accuracy varied across test settings, with R² values as low as 0.34. In addition, the heterogeneous clinical validation cohort included only 28 individuals with direct tau PET imaging.
The paper
A model of brain function to predict tau across Alzheimer’s disease and related disorders
Corriveau-Lecavalier N, Zhang G, Botha H et al.
Nature Communications · 2 Oct 2026
