CLASSIFYING AUTISM SPECTRUM DISORDER BASED ON SCANPATHS AND SALIENCY

Individuals suffering from autism spectrum disorder (ASD) demonstrate viewing patterns that are often different from those exhibited by control subjects, especially in the context of emotional or social stimuli. Previous studies with such content typically relied on precise hand-labelling of the regions of interest in displayed images, thus limiting the transferability of performed analyses onto new, unlabelled images. In contrast to this, we propose an approach that classifies the viewing behaviour of individuals as likely associated with either ASD or typical development in a fully automatic fashion, relying on scanpath features and analytically predicted saliency. Our analysis further demonstrates that gaze data for images with multiple faces have a substantially higher discriminative power than other image groups.