Fig. 5.
Cases in which there was disparity between the pathology dichotomy as normal or abnormal and PET negative or positive were associated with low reader confidence (Panel a) and inter-reader agreement (Panel b). False-negative cases (FN) where the majority interpreted the PET image as negative in the presence of an abnormal neuritic plaque burden were associated with low inter-reader agreement (P < 0.0001, Kruskal Wallis test). False-positive (FP) cases were associated with low reader confidence (P < 0.0001, Kruskal Wallis test). Inter-reader agreement was determined as Fleiss’ kappa coefficient P i (see text for details) which for 5 readers is 1 when all 5 readers are in agreement, 0.6 when the agreement is 4:1 and 0.4 when the agreement is split 3:2. Reader confidence was recorded as a 5-point scale (1-5) with 5 being most certain and 1 being the least certain. Panel c In the 3 false-positive cases, although neuritic plaque frequency was below threshold, β-amyloid in the form of diffuse plaques was comparable to mCERADSOT cases (true positives; TP). One case also had high % area stained for β-amyloid by IHC but was low intensity and the readers called this case true negative (TN) (reader ratio 5:0, Case 13) suggesting that in this case cortical β-amyloid levels in the absence of any neuritic plaques were insufficient to produce a positive image by PET. Another case had sufficient neuritic plaques (mCERADSOT 2.7) even though the % area stained was relatively low. For all panels, boxes represent mean +/- 1 standard error and whiskers represent 95% confidence interval. Open circles represent outlier values and asterisks represent extreme values (see Materials and methods for details)