Abstract:Objective: Autism spectrum disorder (ASD) is a heterogeneous neurodevelopmental disorder which affects the developmental trajectory in several behavioral domains, including impairments of social communication and stereotyped behavior. Unlike typically developing children who can successfully obtain the detailed facial information to decode the mental status with ease, autistic children cannot infer instant feelings and thoughts of other people due to their abnormal face processing. In the present study, we tested the other-race face, the own-race strange face and the own-race familiar face as stimuli material to explore whether ASD children would display different face fixation patterns for the different types of face compared to TD children. We used a machine learning approach based on eye tracking data to classify autistic children and TD children. Methods: Seventy-seven low-functioning autistic children and eighty typically developing children were recruited. They were required to watch a series face photos in a random order. According to the coordinate frequency distribution, the K-means clustering algorithm divided the image into 64 Area Of Interest (AOI) and selected the features using the minimal redundancy and maximal relevance (mRMR) algorithm. The Support Vector Machine (SVM) was used to classify to determine whether the scan patterns of different faces can be used to identify ASD, and to evaluate classification models from both accuracy and reliability. Results: The results showed that the maximum classification accuracy was 72.50% (AUC = 0.77) when 32 of the 64 features of unfamiliar other-race faces were selected; the maximum classification accuracy was 70.63% (AUC = 0.76) when 18 features of own-race strange faces were selected; the maximum classification accuracy was 78.33% (AUC = 0.84) when 48 features of own-race familiar faces were selected; The classification accuracy of combining three types of faces reached a maximum of 84.17% and AUC = 0.89 when 120 features were selected. Conclusions: There are some differences between low-functioning autistic children and typically developing children in the processing of the own-race and other-race faces by the machine learning approach, which might be a useful tool for classifying low-functioning autistic children and TD children.

One size does not fit all: detecting attention in children with autism using machine learning

Attention-Focused Eye Gaze Analysis to Predict Autistic Traits Using Transfer Learning

Early identification of autism spectrum disorder based on machine learning with eye-tracking data

Investigation of Eye-Tracking Scan Path as a Biomarker for Autism Screening Using Machine Learning Algorithms

Automatic Classification Of Children With Autism Spectrum Disorder By Using A Computerized Visual-Orienting Task

Visual Attention Analysis and Prediction on Human Faces for Children with Autism Spectrum Disorder

Classification of Children With Autism and Typical Development Using Eye-Tracking Data From Face-to-Face Conversations: Machine Learning Model Development and Performance Evaluation

Learning Visual Attention to Identify People with Autism Spectrum Disorder.

Autistic Spectrum Traits Detection and Early Screening: A Machine Learning Based Eye Movement Study.

Eye Tracking Biomarkers for Autism Spectrum Disorder Detection using Machine Learning and Deep Learning Techniques: Review

A novel multi-modal model to assist the diagnosis of autism spectrum disorder using eye-tracking data

Using visual attention estimation on videos for automated prediction of autism spectrum disorder and symptom severity in preschool children

Enhanced multilevel autism classification for children using eye-tracking and hybrid CNN-RNN deep learning models

Applying Eye Tracking with Deep Learning Techniques for Early-Stage Detection of Autism Spectrum Disorders

Classifying Individuals With Asd Through Facial Emotion Recognition And Eye-Tracking

Efficient Machine Learning Models for Early Stage Detection of Autism Spectrum Disorder

Attention Analysis in Robotic-Assistive Therapy for Children With Autism

Identifying Children with Autism Spectrum Disorder Based on Gaze-Following

Ensemble Modeling of Multiple Physical Indicators to Dynamically Phenotype Autism Spectrum Disorder

Developing an Artificial Intelligence Based Model for Autism Spectrum Disorder Detection in Children

The study of the differences between low-functioning autistic children and typically developing children in the processing of the own-race and other-race faces by the machine learning approach