A comparison of two novel visual read‐based methods for high tau identification from flortaucipir‐PET

Abstract

AbstractBackgroundTo compare the performances of two novel visual read‐based methods for flortaucipir‐PET in patients with Alzheimer’s Disease (AD) to identify high tau accumulation, defined as AD‐signature weighted neocortical Standardized Uptake Value ratio (SUVr)>1.46.MethodBaseline flortaucipir‐PET scans displaying advanced‐AD tau‐patterns (increased neocortical activity beyond posterior lateral temporal and occipital regions with bindings 1.65‐times higher [1.65x‐visual threshold] than cerebellar average) were utilized from the confirmatory phase of A05 study (N = 78, NCT02016560). Four expert human‐readers conducted reads using two novel visual read‐based methods while being blinded to each other and all quantitative results. The first approach (SUVrmax‐method) requires a reader to visually identify and hand‐draw focal regions‐of‐interest on frontal areas with the highest bindings and calculate an SUVr between the identified frontal maximum and average cerebellum binding. The second approach (Enhanced Read‐method) requires the reader to evaluate binding elevation compared to cerebellum with an enhanced visual threshold, either across gray matter (global, 3.795x‐visual threshold) or only in the frontal lobe (2.805x‐visual threshold), compared to the standard 1.65x‐visual threshold specified in the FDA‐approved Tauvid™ label. Methods were compared to the standard quantitative high tau determination as standard‐of‐truth.ResultBased on a previously found ROC‐determined cut point using computer‐generated SUVrmax values, the SUVrmax‐method showed median 88% Positive Percent Agreement (PPA), 87% Negative Percent Agreement (NPA) and 87% overall accuracy across raters. The Enhanced Read‐method showed median 75% PPA, 88% NPA and 85% overall accuracy and median 81% PPA, 91% NPA and 88% overall accuracy for global and frontal approaches, respectively.ConclusionBoth novel flortaucipir‐PET visual read‐based methods showed comparable and promising performances in identifying AD with high tau, do not require computationally intensive approaches and can be implemented with different imaging software.