Evaluation of a visual read method for flortaucipir PET Scans

Objectives: To evaluate a clinically applicable visual read method for flortaucipir PET scans.

Methods: Five imaging physicians were trained in-person using a predefined read methodology to visually interpret flortaucipir PET scans. After scaling images to cerebellum, they were instructed to evaluate regions of the neocortex (the posterolateral temporal (PLT), occipital, parietal and frontal regions) for increased tracer uptake, and interpret scans as either not consistent (tAD-) or consistent with an AD pattern (tAD+ or tAD++) using the following criteria: tAD-: no increased activity or increased activity isolated to the mesial temporal, anterior lateral temporal, and/or frontal region(s) tAD+: increased activity in the PLT or occipital region(s) tAD++: increased activity in the parietal/precuneus region(s) or frontal region with PLT, occipital, or parietal region(s) Read categories were derived from longitudinal studies, where increased neocortical uptake in the PLT was associated with amyloid positivity, and activity beyond the PLT/occipital regions was associated with greater cognitive decline. After training on demonstration and practice cases, all readers independently read 105 scans from end-of-life subjects enrolled in a Phase 3 clinico-pathological trial (study A16) plus 17 scans from a supplemental academic cohort (total n=122). The primary analyses tested sensitivity and specificity for a flortaucipir PET scan visual interpretation consistent with AD (tAD+/tAD++) to detect AD tau pathology (Braak V/VI) and high level of AD neuropathologic change in patients who came to autopsy (n=82).

Results: 82 cases came to autopsy and had evaluable scans. Median sensitivity and specificity for a flortaucipir PET AD pattern to identify Braak V/VI pathology were 89.1% and 83.3%, respectively. The median sensitivity and specificity for identifying high AD neuropathic change (NIA-AA path criteria) were 95.1% and 80.5%, respectively. Interrater reliability as measured by Fleiss’ kappa for all interpreted scans (n=122) was 0.80 and overall agreement was 90.3%. Results from additional studies will also be presented.

Conclusions: There was substantial agreement among readers. Using the read method, readers were able to identify scans associated with AD pathology at autopsy.