Clinical evaluation of SPARE‐TauAD, an AV‐1451‐derived tau index

Background The deposition of abnormal proteins defines neurodegenerative diseases. Among these proteins, tau deposits define frontotemporal lobar degeneration (FTLD) tau, and together with Aβ, it defines the more prevalent Alzheimer’s disease (AD). PET scans with ligands binding AD tau neurofibrillary tangles (NFT) topographically quantify the amount of pathology. However, for clinical purposes, this information needs to be summarized into easily applicable and meaningful indices. Here we develop a machine learning‐derived tau PET index, the Spatial Pattern of Abnormality for Recognition of Tau in AD (SPARE‐TauAD). Method 603 subjects were included in the study. Their clinical diagnoses were: 366 cognitively normal (CN), 177 mild cognitive impairment (MCI), and 60 dementia subjects. These subjects were classified as Aβ positive and negative based on CSF Aβ1‐42 levels or florbetapir PET summary composite. SPARE‐TauAD was derived using a support vector machine classifier comparing AV‐1451 PET scans in 141 Aβ negative CN subjects against 108 Aβ positive MCI and dementia subjects. We then evaluated its association with CSF t‐tau, global cognition (measured using ADAS‐Cog 13), clinical diagnosis, and clinical progression. Result SPARE‐TauAD and CSF t‐tau showed a moderate correlation (rho=0.51, p