Inverted Faces: An Indicator to Identify ASD and TD

Background: Eye tracking technique is widely used to assist in identifying ASD due to its non-invasive and comfortable characteristic. When seeing a normal human face, ASD and TD children show big different interests, especially in eyes and mouths. Inverted human faces are interesting stimulus that cause human brain having different response than normal faces. But there’s no experiment to show whether responses to inverted faces can be an indicator to identify ASD and TD.

Objectives: Our goal is to investigate whether there are significant differences between ASD and TD children on their response to inverted faces, and to set up a model identifying ASD and TD.

Methods: The experiment used SMI desktop eye tracking system with frequency 250Hz. It was designed to have 12 pictures of human faces in same size and luminance with different gender and ages. Starting with a 10 second black picture, then 9 pictures were showed continuously with 2 seconds on each, and last 3 pictures with 4 seconds on each. In the series, the 3^rd ,7^th, 10^th, 11^th ,12^th pictures were inverted faces. Fixation time on Area of Interest (AOI) of eyes, mouths, and faces were collected, as well as pupillary data. Data were analyzed later by SPSS and MATLAB.

Total 25 subjects of 3-6 years old children with 11 ASD and 14 TD provided valid data on this experiment. ASD children had diagnosis from hospital and double checked by CARS (Child Autism Rate Scale).

Results: Based on the experimental data, 13 factors were extracted, and 5 of them showed significant difference for ASD and TD group. Principal Component Analysis showed that 3 of these factors were critical. Naive Bayes Classifier was constructed with these 3 critical factors, and its performance was evaluated by leave-one-out method. The model gave a result of an accuracy of 88.0% and an AUC of 92.2% on its prediction.

Conclusions: Eye responses on inverted faces is a sensitive indicator to identify ASD and TD. It can be used to build up a model to assist in ASD screening on 3-6 years old children.