Machine learning's effectiveness in evaluating movement in one-legged standing test for predicting high autistic...

Area of Science:

• Neuroscience
• Developmental Psychology
• Biomedical Engineering

Background:

• Growing evidence links autism spectrum disorder (ASD) to motor impairments, particularly balance deficits.
• The one-legged standing test (OLST) is a common assessment for balance.
• Machine learning (ML) offers objective analysis of movement data for clinical assessments.

Purpose of the Study:

• To evaluate the efficacy of ML in analyzing OLST movement data for predicting high autistic traits.
• To identify key movement variables from OLST that are predictive of autistic traits.

Main Methods:

• 126 children (64 boys, 62 girls) performed the OLST on a pressure sensor.
• Full-body images and foot pressure data were collected.
• A Support Vector Machine (SVM) algorithm was employed to classify participants into high and low autistic trait groups.

Main Results:

• ML models achieved perfect accuracy (1.000) for sensitivity and specificity using proposed and combined variables.
• Shoulder, hip, and trunk movements were identified as significant predictors of balance status in children with high autistic traits.
• Higher Social Responsiveness Scale scores correlated with increased probability of belonging to the high autistic trait group.

Conclusions:

• ML analysis of OLST movement data, particularly shoulder and waist movements, is effective for predicting high autistic traits.
• Further research incorporating diverse balance metrics is recommended to enhance predictive capabilities.