A new bio-inspired metric based on eye movement data for classifying ASD and typically developing children

In this paper, we propose a new bio-inspired metric for classifying autism spectrum disorder (ASD) children and typically developed (TD) children. The model used in the Saliency4ASD Grand Challenge at ICME 2019 uses linear regression and prior probability to process distance and time data respectively. Unfortunately, this model performs unsatisfactorily because the visual attention characteristics of ASD and TD children are similar under certain circumstances. Therefore, we screen stimulus materials to select these with significant differences between eye movement distribution of ASD and TD children. We calculate the SSIM value of the ASD and TD data of each picture and conduct the subjective experiments to classify the stimulus materials into two categories: the images with the similar attention map for ASD and TD children; and the images with the dissimilar attention map for ASD and TD children. Owing to the biological property of eye, a viewing angle will be formed when people are observing a picture. Meanwhile, gazing at one point of longer time means more attention. Thus, we pick the point of the longest fixation time for each data group and extract the patch centered on this point. Three point-add strategies are afterward utilized to add points on this patch. Subsequently, a new bio-inspired metric based on graph theory is developed. Experimental results show that the new model outperforms our previous model with a classification accuracy of 72.3%.