RT Journal Article
SR Electronic
T1 Can incorporation of a quantitative estimate of cortical to cerebellar SUVr as an adjunct to visual interpretation improve the accuracy and reliability of florbetapir PET scan interpretation?
JF Journal of Nuclear Medicine
JO J Nucl Med
FD Society of Nuclear Medicine
SP 245
OP 245
VO 55
IS supplement 1
A1 Pontecorvo, Michael
A1 Devous, Michael
A1 Arora, Anupa
A1 Devine, Marybeth
A1 Lu, Ming
A1 Joshi, Abhinay
A1 Breault, Christopher
A1 Skovronsky, Daniel
A1 Mintun, Mark
A1 Heun, Stephen
YR 2014
UL http://jnm.snmjournals.org/content/55/supplement_1/245.abstract
AB 245 Objectives We evaluated whether quantitation by FDA-cleared software as an adjunct to visual interpretation (VisQ method) improves accuracy of florbetapir scan interpretation in lower accuracy readers (<90% accuracy using visual techniques alone). Methods 22 physicians were trained to quantitateflorbetapir images as an adjunct to visual interpretation. Each reader then visually interpreted, without access to quantitation information, 46 florbetapir PET images from subjects autopsied within one year of PET scan and 50 images from patients seeking diagnosis for cognitive decline. Next, the readers reviewed all 96 images again, using a software package to calculate a quantitative estimate (SUVr) of florbetapir activity and applied the VisQ method. For each case, the reader compared the SUVr to normal ranges. For scans in which the initial visual read and quantitative result were discordant, readers reassessed their original interpretation in light of the quantitation, reinspected images and verified ROI placement (a potential source of error in SUVr calculation) to resolve any discordance. They then provided a final VisQ interpretation. Results With visual interpretation, the mean accuracy (compared to pathology truth standard) on the 46 autopsy verified cases was 89.5%. 7 of the 22 readers were identified as lower accuracy (< 90%) from the visual reads. In these 7 readers, accuracy improved significantly with VisQ from a mean of 81.7% to 88.8% (p=0.0029). Across all 22 readers, accuracy improved from 89.5% to 93.8% (p<0.0001). Interreader agreement across all 96 cases increased (p < 0.05) from a Fleiss kappa of 0.72 to 0.79. Conclusions Accuracy and interreader reliability significantly increased after applying the VisQ method that incorporates quantitative estimates of florbetapir PET amyloid binding (SUVr). To test the generality of these findings studies are ongoing with additional software packages.