Correlation Of Amyloid Pet In Centiloid Units With Neuropathological Findings In Alzheimer'S Disease

482 Introduction: The Centiloid scale has been derived to standardise amyloid tracer positron emission tomography (PET) reporting, and has not yet been validated against neuropathological data. We aimed to determine the accuracy of Centiloid scores compared with post-mortem neuropathological evaluation, and thus determine the optimal Centiloid units (CL) threshold ranges in the assessment of Alzheimer9s disease (AD) in relation to neuropathological data, expert visual reads, and clinicopathological diagnosis. Methods: Forty-nine deceased subjects with various dementia diagnoses from previous dementia imaging studies were identified from the databases of two research units. Subjects had both an antemortem PET scan with either 11CPiB or 18F-florbetaben; and post-mortem brain neuropathological evaluation. CL results were derived. The author CR, blinded to CL values and neuropathological data, visually read all scans. Neuropathological evaluation yielded a C score based on densities of none, sparse, moderate or frequent neuritic plaques in the inferior temporal lobe as per Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) guidelines. Final clinicopathological diagnoses were also recorded. CL values were firstly correlated with binary C score categories (high vs low; and none vs any; where “high” = moderate and frequent, “low” = none and sparse, “none” = none, and “any” = sparse, moderate and frequent). Secondly, CL values were correlated with binary visual read (negative or positive). Thresholds of 15, 20, 25, 30 and 35 CL were tested to determine optimal Centiloid unit value performance. Finally, cases of AD as determined by clinicopathological diagnosis were correlated with CL values, excluding cases of mixed dementia. Results: Correlation of Centiloid values with binary C score categories (high vs low) yielded an optimal threshold of between 15 and 20 CL (sensitivity: 1.00; specificity: 0.90; PPV 0.86; NPV 1.00; accuracy 0.94) for ruling out the presence of amyloid plaques, and a threshold of 35 CL (sensitivity: 0.95; specificity: 0.93; PPV 0.90; NPV 0.97; accuracy 0.94) as having the highest specificity for supporting a high C score result (moderate or frequent plaques). Against binary C score categories (none vs any), the optimal threshold found for ruling out the presence of amyloid plaques was between 15 and 20 CL (sensitivity: 0.88; specificity: 0.96; PPV 0.95; NPV 0.89; accuracy 0.92). Correlation of expert visual read yielded optimal threshold values of between 25 and 30 CL (accuracy 1.00). Correlation of CL values with final clinicopathological diagnosis in seventeen AD cases yielded a median CL result of 87.7 (IQR ±42.2), with \u003e90% of cases scoring \u003e45 CL. Conclusion In this cohort, a CL threshold between 15 and 20 accurately reflected the absence of amyloid plaques on neuropathological examination. Visual reads were uniformly positive when CL values were greater than 25 CL, and uniformly negative when lower. For C score correlation, a threshold of 35 CL provided optimal specificity, a threshold of 20 CL yielded optimal sensitivity, and optimal accuracy was equivalent at thresholds of 20 and 35 CL. Greater than 90% of AD cases diagnosed on final clinicopathological criteria had a score of \u003e45 CL. A limitation of the study is that only two of forty-nine subjects had results between 15 and 35 CL, restricting the ability to more tightly define thresholds. Further research is needed to refine and validate these thresholds.