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

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder diagnosed by clinicians and experts through questionnaires, observations, and interviews. Current diagnostic practices focus on social and communication impairments, which often emerge later in life. This delay in detection results in missed opportunities for early intervention. Gait, a motor behavior, has been previously shown to be aberrant in children with ASD and may be a biomarker for early detection and diagnosis of ASD. The current study assessed gait in children with ASD using a single RGB camera-based pose estimation method by MediaPipe (MP). Data from 32 children with ASD and 29 typically developing (TD) children were collected. The ASD group exhibited significantly reduced step length and right elbow° and increased right shoulder° relative to TD children. Four machine learning (ML) algorithms were employed to classify the ASD and TD children based on the statistically significant gait parameters. The binomial logistic regression (Logit) performed the best, with an accuracy of 0.82, in classifying the ASD and TD children. The present study demonstrates the use of gait analysis and ML techniques for the early detection of ASD.