Nuances in [18F] flutemetamol reader methodology: High inter- and intra reader agreement rates

Abstract

Confidence in the ability of the end user to designate a PET amyloid scan either positive or negative by simple visual inspection is paramount to the approvability by global regulators. GEHC has developed a simple system of identifying the presence or absence of cortical binding based on the reference region binding level and the contrast between white matter and cortical grey matter. This provides a means to a binary yes or no decision for amyloid. These methods have been applied throughout the Flutemetamol development program and have demonstrated that readers blinded to clinical information have a high inter and intra reader agreement rate. [18F] Flutemetamol shows high uptake into global cortical regions in the majority of Alzheimer's subjects. In healthy subjects non-specific white matter uptake is apparent and the reader is trained to recognise structures showing uptake such as pons and thalamus that are not related to amyloid uptake. In cases showing amyloid uptake the reader is referred to regions such as the frontal, cingulate, precuneus, parietal and striatum.Phase I images were used for training readers and methods described above were used to read a wider selection images from Alzheimers’ (n = 27), Mild Cognitive Impairment (n = 20) and Healthy Volunteers (n = 25) scanned with [18F] Flutemetamol. Five independent readers were trained to classify scans as abnormal (high cortical uptake) and normal (low cortical uptake with only white matter non-specific identified). Inter-reader agreement between the 5 blinded readers was 98.7% (in 288 images examined there was complete agreement in 284). The Fleiss Kappa score, a measure of agreement between the readers was 0.96 indicating excellent reader agreement. 10% of the images read were subject to a randomised re-read for each reader. Of the total of 40 re-reads only one image was read with a different result by a single reader. Brain biodistribution of [18F]flutemetamol in abnormal (significant fibrillar amyloid) and normal (none/negligible fibrillar amyloid) images is easily assessed by the described read methodology. These methods were devised in phase 1 and validated in phase II resulting in a robust and highly consistent method for classifying PET ß-amyloid images.