S1 Fig -

(a) Eye gaze heat map of the participants overlayed on a segmented version of the original 22q11DS image (see S1 Text for original image sources). The leftmost image is the average heat map of the successful clinicians (top) and non-clinicians (bottom). We apply two different noise-thresholds: a low threshold to remove possibly spurious visual interests (middle), and then a high threshold that more clearly indicates facial regions with high visual attentions (right). (b) Saliency maps of image with key regions that affect the classifier accuracy. We apply a low (left) and high coverage-threshold (right) on the saliency maps so that the area covered is approximately the same as the low and high noise-threshold, respectively.

(TIF)