The Computational Modeling of Perceptual Biases of Children with ASD in Naturalistic Settings

Background: Several studies have shown that individuals with ASD exhibit abnormalities in processing visual information. However, because these studies typically employ elementary perceptual stimuli, such as contrast gratings, it is not clear how applicable they are to natural settings.

Objectives: In this study we use computational techniques developed for modeling visual attention in order to compare the utilization of basic perceptual features by children with ASD versus typically-developing controls as these children view naturalistic scenes. We also access higher level contextual influences by manipulating the presented scenes and gauging the impact of the manipulation on feature usage.

Methods: 40 children diagnosed with ASD (age(months): M=45.0, SD=7.6) and 22 typically-developing controls (age(months): M=44.6, SD=5.8) were presented scenes depicting play interactions between a child and an adult. A computational model of visual attention was used to decompose scenes into elementary perceptual features of intensity, orientation, motion, and color. These features were matched to the gaze positions of the children viewing the scenes. Scenes were presented either upright or inverted and with-sound or mute.

Results: Consistent with laboratory results, children with ASD used more intensity information and less motion information than typical controls. In addition, in ASD, the use of motion was positively correlated with VMA and NVMA. Finally, both groups increased usage of intensity information when the scene was inverted and focused more on motion when sound was added.

Conclusions: The results suggest that abnormalities detected in ASD using basic perceptual stimuli in laboratory settings extend to how children with ASD respond to visual stimuli in natural settings. Furthermore, the relationship between motion usage and behavioral indices of functioning offers the possibility that early abnormalities in perception may have developmental or cognitive associations. Finally, the response of both groups to context modulation suggests common mechanisms for reshaping and redirecting visual attention.