AbstractBackgroundCognitive decline in Alzheimer Disease (AD) is strongly correlated to the spatiotemporal pattern of tau‐protein neurofibrillary tangles (NFTs). Evidence suggests tau seeding spreads transneuronally in a prion‐like manner, a biological process distinct from NFT aggregation in previously‐affected regions. We previously proposed a method for quantifying global tau spatial spread (TSS) using tau positron emission tomography (PET) which found TSS is particularly informative in preclinical AD. The spread of tau pathology into specific regions may therefore be key to the onset of cognitive decline.Methods445 older adult and 21 younger control participants were recruited. Cognition was assessed with the Knight ADRC‐PACC and Tau‐PET images were parcellated using the Schaeffer400 atlas. For each region, TSS was calculated as the proportion of voxels with significant levels of tau. Abnormal voxels were identified by z‐scoring tau‐PET voxel‐wise relative to corresponding voxels in younger controls and then thresholded at z>1.96 for significance. This study predicted participant PACC from regional TSS using random forest regression (trained with 100 repeated 10‐folds). Feature importance (FI) was calculated with Mean Decrease in Impurity to identify regions in which tau spread is most significant to cognitive impairment.ResultsRegional TSS demonstrated highly saturated regions in the limbic and temporal lobes with progressively lower levels in peripheral regions, largely sparing the frontal lobe (Figure 1). However, the importance of regional spread in predicting cognition did not reflect the same spatial pattern (Figure 2). Four regions emerged with the highest FI in the limbic and default networks, indicating the spread of tau within these brain regions is key to predicting cognitive impairment (Table 1).ConclusionOur results suggest that tau pathology has spread extensively within medial and temporal regions early in AD. Regional TSS can be used to identify peripheral regions in which tau pathology is currently spreading into. As expected, limbic regions in the temporal lobe were critical to predicting participant cognitive impairment. However, regions in the temporal and parietal lobes belonging to the default network were additionally informative. Spread of tau pathology into new peripheral brain regions should therefore be evaluated when predicting future AD‐related cognitive decline.