AbstractBackgroundIdentifying age‐inappropriate cortical neurodegeneration on structural MRI is a key component of Alzheimer’s disease (AD) assessment. However, widely available reference standards that enable objective interpretation are lacking. BrainChart (www.BrainChart.io; https://github.com/BrainChart/Lifespan) calculates age and sex adjusted centile scores for structural MRI metrics using normative data from 101,457 participants across the lifecourse and can be applied across different datasets. We applied this approach to regional cortical thickness estimates and explored its utility in identifying patients with AD associated neurodegeneration.MethodCross‐sectional BrainChart cortical thickness centile scores were calculated for 241 individuals with pathologically confirmed AD (median scan time prior to death 5.8 years), as well as 1432 cognitively normal participants (CN) from the NACC dataset. Area under the receiver operating characteristic curve (AUC) analyses were performed to identify the cortical regions that best identified patients with pathologically confirmed AD. This approach was then applied to 305 participants from the ADNI‐3 cohort, all of whom underwent clinical amyloid (florbetapir) PET and tau (flortaucipir) PET within 6 months of structural MRI (CN = 185; Mild Cognitive Impairment (MCI) = 95); AD = 26).ResultThe top ten performing regions for differentiating pathologically confirmed AD from CN in the NACC dataset were entorhinal, middle temporal, inferior parietal, isthmus cingulate, fusiform, supramarginal, inferior temporal, superior temporal, precuneus and temporal pole. AUCs ranged from 0.72 – 0.77 (figure 1 and table 1). Combining these ten regions into a single model significantly improved diagnostic separability (AUC = 0.81, figure 1). Optimal cut‐points for individual regions ranged from the 25th centile to the 35th centile, with sensitivities ranging from 57.7% to 68.9%, and specificities ranging from 72.6% to 80.9%. Using progressively lower centile scores increased specificity but decreased sensitivity (table 1). In ADNI‐3 participants, a combined model of the same ten regions derived from the NACC dataset was able to discriminate amyloid and tau pet positive AD patients from amyloid and tau negative CN (AUC = 0.98), as well as amyloid and tau positive MCI patients from amyloid and tau negative CN (AUC = 0.85) (figure 2).ConclusionBrainChart age and sex adjusted cortical thickness centile scores, derived from an extensive normative dataset, represent a generalizable method for objectively identifying cortical neurodegeneration in AD.