Quantitation of PET signal as an adjunct to visual interpretation of florbetapir imaging

PURPOSE: This study examined the feasibility of using quantitation to augment interpretation of florbetapir PET amyloid imaging. METHODS: A total of 80 physician readers were trained on quantitation of florbetapir PET images and the principles for using quantitation to augment a visual read. On day 1, the readers completed a visual read of 96 scans (46 autopsy-verified and 50 from patients seeking a diagnosis). On day 2, 69 of the readers reinterpreted the 96 scans augmenting their interpretation with quantitation (VisQ method) using one of three commercial software packages. A subset of 11 readers reinterpreted all scans on day 2 based on a visual read only (VisVis control). For the autopsy-verified scans, the neuropathologist's modified CERAD plaque score was used as the truth standard for interpretation accuracy. Because an autopsy truth standard was not available for scans from patients seeking a diagnosis, the majority VisQ interpretation of the three readers with the best accuracy in interpreting autopsy-verified scans was used as the reference standard. RESULTS: Day 1 visual read accuracy was high for both the autopsy-verified scans (90%) and the scans from patients seeking a diagnosis (87.3%). Accuracy improved from the visual read to the VisQ read (from 90.1% to 93.1%, p 90% on day 1). Accuracy in interpreting the autopsy-verified scans also increased from the first to the second visual read (VisVis group). However, agreement with the reference standard (best readers) for scans from patients seeking a diagnosis did not improve with a second visual read, and in this cohort the VisQ group was significantly improved relative to the VisVis group (change 5.4% vs. -1.1%, p < 0.0001). CONCLUSION: These results indicate that augmentation of visual interpretation of florbetapir PET amyloid images with quantitative information obtained using commercially available software packages did not reduce the accuracy of readers who were already performing with above average accuracy on the visual read and may improve the accuracy and confidence of some readers in clinically relevant cases.