Comparison of [18F]-florbetapir PET visual interpretation and quantitation to detect amyloid pathology at the community level

Introduction In research and clinical trial applications, [ 18 F]-florbetapir positron emission tomography (PET) imaging data are commonly evaluated with quantitation-based approaches. In clinical practice, scans are primarily interpreted visually by nuclear medicine physicians or radiologists (community readers), often with varying amyloid PET expertise. This study investigated the concordance between visual reads and quantitation of florbetapir PET scans at the community level. Methods The analysis included 337 florbetapir PET scans from 2 studies: AV45-QP (96 scans each interpreted by 80 community readers, generating 7680 binary reads) and AV45-RTP01 (241 scans and interpretations submitted by 29 community readers, generating 237 binary reads after removing indeterminate or uninterpretable scans). Inter-reader reliability (Fleiss' κ) was assessed in AV45-QP. Area under the receiver operating characteristic curve (AUC) analysis evaluated the association between quantitation and visual reads, with optimal thresholds determined by Youden's index and concordance evaluated with Cohen's κ. Results There was substantial inter-reader agreement in AV45-QP (Fleiss' κ = 0.71). The median of individual readers' optimal Centiloid (CL) thresholds was 24.7 (range: 12.84–59.44 CL); 91.3% of readers had optimal thresholds between 20 and 50 CL. In AV45-RTP01, the optimal threshold was 22.2 CL. Visual reads, compared with quantitation, showed a median AUC of 0.90 and median Cohen's κ of 0.72 for AV45-QP, and AUC and Cohen's κ of 0.91 and 0.76, respectively, for AV45-RTP01. Conclusions Community readers showed substantial but imperfect inter-reader agreement and concordance with quantitation of florbetapir PET scans. Inter-reader disagreement and visual–quantitative discordance were non-negligible, supporting complementary roles for visual interpretation and quantitation.