Background: Variable quantitative thresholds have been proposed to define amyloid positivity on PET, but few studies have compared these thresholds to post-mortem amyloid burden as the “standard of truth.”
      Methods: We included 43 individuals who underwent PIB-PET during life and brain autopsy (mean age at PET=66.8±8.0, T autopsy -T PET =3.0±1.8 years). Patients were evaluated clinically at UCSF (N=36) and UC Davis (N=7), and all underwent PET at Lawrence Berkeley Laboratory. Clinical diagnoses included Alzheimer's disease (AD, N=10), frontotemporal dementia (N=24), other dementia (N=2), mild cognitive impairment (N=5) and cognitively normal (N=2). Global amyloid was measured with a PIB Index (gray cerebellum reference) via 0-90 minute Distribution Volume Ratios (DVR, N=39) and/or 50-70 min Standardized Uptake Value Ratios (SUVR, N=41). Amyloid at autopsy was classified using CERAD (N=43) and NIA-Reagan criteria (N=33, 10 cases unclassifiable). We compared conservative PIB thresholds in the literature (SUVR high =1.40, DVR high =1.20) and more liberal proposed thresholds (SUVR low =1.21, DVR low =1.08, see Villeneuve et al., submitted) to CERAD frequent neuritic plaques (NP) and NIA-Reagan high-likelihood AD. Receiver operator characteristic (ROC) analyses were performed to empirically derive optimal thresholds.
      Results: The liberal DVR low and SUVR low thresholds captured all but one patient with CERAD frequent NP, and each threshold misclassified one patient with lower amyloid burden (sensitivity 93-94%, specificity 96%, see Figure - blue line). Conservative thresholds were considerably less sensitive (DVR high 71%, SUVR high 77%) but 100% specific (Figure - red line). Liberal thresholds performed similarly well in capturing NIA-Reagan high-likelihood AD (sensitivity 92%, specificity 95% for both DVR low and SUVR low), whereas conservative thresholds were again less sensitive (DVR high 75%, SUVR high 85%) but 100% specific. ROC analyses yielded areas under the curve ranging from 0.929-0.971 (all p<0.001). The optimal thresholds derived from ROC were nearly identical to our a priori liberal thresholds (DVR=1.09 and SUVR=1.20-1.21).
      Conclusions: Liberal thresholds for PIB-positivity were highly sensitive and specific for high-burden amyloid pathology. Conversely, more widely used conservative thresholds appeared to be overly stringent in defining amyloid positivity. More work is needed to verify these findings in subjects with intermediate pathology and independent cohorts.