Optimizing early detection of beta‐amyloid accumulation with PET using spatial extent

Abstract

AbstractBackgroundAs clinical trials shift towards prevention, there has been increasing interest in improving early detection of beta‐amyloid (Aβ), below standard global PET positivity thresholds. While many studies have demonstrated that some regions appear more vulnerable to the earliest deposits, those regions have varied widely across studies. We sought to develop more flexible metrics for identifying subthreshold individuals at risk for progression to global Aβ positivity by evaluating a wide range of early regional targets and expanding our measurement beyond tracer retention (DVR/SUVR) to include spatial extent.Method269 clinically‐normal older adults were included from the Harvard Aging Brain Study, focusing on 150 initially globally PIB‐ adults with longitudinal PET. Region‐specific positivity thresholds were computed using iterative longitudinal two‐level linear mixed effect models that identified the cutoff beyond which local accumulation significantly increases over time. Next, we identified all regions significantly predictive of future global accumulation in initially PIB‐ adults and generated 3 aggregate masks at liberal (p<.05), moderate (p<.001) and conservative a‐levels (Bonferroni p<.0007). DVR and spatial extent (number of elevated ROIs) were computed in each mask. Receiver operator characteristic (ROC) curve analysis determined sensitivity/specificity of each metric to predict future global positivity. External validation was performed using globally florbetapir‐negative (FBP‐) healthy controls from the Alzheimer’s Disease Neuroimaging Initiative (ADNI, n=211).ResultRegional variations in measurement reliability were detected (Fig1) and accounted for by employing accumulation‐based regional positivity thresholds. A broad set of ROIs were identified as potential early targets based on prediction of global accumulation, encompassing bilateral cingulate, medial frontal/parietal cortices and left‐lateralized lateral frontal/parietal/temporal cortices (Fig2). ROC analyses demonstrated that spatial extent outperformed DVR in predicting progression to global PIB+ within 3 years (Fig3), with 82% sensitivity and 97% and 98.5% specificity, respectively, for the two larger aggregate masks. External validation in ADNI (Fig4) again demonstrated the superior performance of spatial extent (SE=100%, SP=91.6%) in predicting progression to global positivity within 3 years.ConclusionOur findings demonstrate that due to a high level of heterogeneity in early Aβ deposition, measures of spatial extent across a broad set of neocortical regions are more sensitive to detect early Aβ than traditional DVR/SUVR.