Augmenting amyloid PET interpretations with quantitative information improves consistency of cerebral amyloid detection


      Establishing reliable methods for interpreting the presence or absence of elevated cerebral amyloid plaque on PET scans is increasingly important for radiologists with the greater availability of molecular PET imaging in clinical practice. We examined the value of adding quantitative amyloid PET information to the interpretation of amyloid-PET scans.


      A total of 60 nondemented (CDR 0) adults over age 65 (mean age 73.6 years) were evaluated with amyloid-PET imaging using florbetapir as part of screening for a Alzheimer’s disease prevention study at the University of Kansas Alzheimer’s Disease Center. Images were first interpreted visually (Visual Read) as either ‘elevated’ or ‘not elevated’ using FDA-approved methods by two different raters. Images were then re-evaluated after the reader considered quantitative analyses (VisualQuant Read) using MIMneuro software to compare the image to a standard atlas. We examined the frequency of interpretation changes after quantitative information was considered and whether this information improved the inter-rater agreement of interpretations. We also examined how the Visual Reads and VisualQuant Reads compare to a purely quantitative read where a scan was determined to be elevated based on an SUVR>1.1 in six regions of interest (anterior cingulate, inferior medial frontal, lateral temporal, posterior cingulate, precuneus, and superior parietal). Simple percent agreement and Cohen’s kappa were used to measure agreement.


      60 scans were interpreted by two blinded raters. The initial Visual Read was changed after the introduction of quantitative information in 6.7% (n=4) of scans for Rater 1 and 11.7% (n=7) for Rater 2. Initial Visual Reads were changed from non-elevated to elevated in 9 of the 11 (81.8%) scans that were changed. Agreement of reads across raters was 90% (54 of 60; Kappa=0.75) for Visual Reads and increased to 95% (n=57, Kappa=0.89) for the VisualQuant Reads. Quantitative only assessments (SUVR >1.1) were concordant with VisualReads (86.7% for Rater 2 and 90.0% for Rater 1) but concordance was higher with the VisualQuant Reads (95.0% - 96.7%).


      Augmenting the radiological interpretation of amyloid PET scans with quantitative information appears to improve the consistency in interpretations for the early detection of the presence of cerebral amyloid accumulation.