Early detection of autism spectrum disorder: gait deviations and machine learning

Area of Science:

• Neurodevelopmental Disorders
• Biomarkers for Early Detection
• Computational Biomechanics

Background:

• Autism Spectrum Disorder (ASD) diagnosis relies on social and communication impairments, often diagnosed late.
• Delayed diagnosis of ASD misses crucial early intervention opportunities.
• Gait abnormalities in children with ASD present a potential early biomarker.

Purpose of the Study:

• To assess gait patterns in children with ASD using RGB camera-based pose estimation.
• To identify statistically significant gait parameters differentiating ASD from typically developing (TD) children.
• To develop and evaluate machine learning models for ASD classification based on gait data.

Main Methods:

• Utilized MediaPipe (MP) for single RGB camera-based pose estimation to analyze gait.
• Collected gait data from 32 children with ASD and 29 TD children.
• Employed four machine learning algorithms, including binomial logistic regression, for classification.

Main Results:

• Children with ASD showed significantly reduced step length and right elbow angle, and increased right shoulder angle compared to TD children.
• Machine learning models were trained using statistically significant gait parameters.
• Binomial logistic regression achieved the highest classification accuracy of 0.82.

Conclusions:

• Gait analysis via RGB camera pose estimation is a viable method for assessing children with ASD.
• Specific gait parameters can differentiate ASD from TD children.
• Machine learning models, particularly logistic regression, show promise for early ASD detection using gait biomarkers.