Fig. 3.

Accuracy of atrophy detections in relation to the gold standard reading. Blue indicates the accuracy using a strict definition of agreement while red displays accuracy levels with a more liberal criterion (see main text for detail). The reported regions are indicated above each panel. Numbers in brackets indicate the percentage of cases remaining after excluding those with conflicting ratings of the two gold standard readers. The readers (R1-R7) are ordered by increasing experience from left to right. For each reader, the left bar indicated performance without and the right bar with z-scores. The trendlines indicate qualitatively the relation between experience and accuracy without (dashed trendlines) and with (solid trendline) atrophy scores. The transparent horizontal blue and red line denote the chance level 0.2 for full and 0.46 partial accuracy, respectively.