Dimensionality of Gaze Patterns Towards Faces and Objects in Toddlers with ASD

Background: Eye-tracking is a prominent tool in psychological and cognitive research. However, the analysis of eye-tracking data is confounded by choices in how fixations should be defined, i.e. how densely packed points must be in order to be considered a fixation. Our previous work (Shic, Scassellati, & Chawarska, 2008), has shown that by manipulating fixation parameters, the fixation statistics between ASD and TD toddlers can reversed. There exists evidence that scale-free fractal structure may derive from gaze patterns of viewing naturalistic stimuli.

Objectives: (1) To explore the nature of toddlers’ scanning patterns during complex scene viewing, independent of fixation parameters. (2) To apply this technique to examine scale-free gaze behaviors in toddlers with Autism Spectrum Disorders (ASD) and typical development (TD) when looking at faces and blocks.

Methods:  We presented 6 trials (10s/trial) with pictures of faces and 6 trials with block designs to TD toddlers (n = 12, Mean age=23.8(±2.8) months) and toddlers with ASD (n=10, Mean age=22.7 (±3.7) months). We examined the log of the number of fixations (n) as a function of the log of box size (s) ranging from .1 deg. to 10 deg. This is a greedy box-counting algorithm for fractal dimensionality, equivalent to an adaptation of fixation identification algorithms at multiple distance scales.

Results: The smallest R-squared in our linear regression is R²=.944, with an average of R²=.992, suggesting that the scanning patterns were strongly self-similar and that measures of gaze pattern fractal dimensionality through box counting is robust. A mixed linear model on the box count dimensionality of our eye tracking data revealed a main effect of stimulus class (p<.01) with a higher dimensionality for blocks than faces. An interaction was found for group and stimulus class (p<.05) driven by a lower dimensionality for faces than blocks in TD toddlers, but no such effect for toddlers with ASD. There was a trend for toddlers with ASD to exhibit higher dimensionalities for scanning faces (p=.08). Our results are consistent with lower dimensionalities being associated with more directed patterns of scanning.

Conclusions:  Our use of fractal dimensionality may offer a route to more robust, more informative, and less biased approaches towards eye-tracking analysis. Consistent with previous experiments, our study, which reports on box counting fractal dimensionality, suggests that TD toddlers use different basic attentional strategies to process faces and blocks, whereas toddlers with ASD may be employing similar distributional strategies for both stimuli classes.