Machine Learning-Based Early Prediction Model for Autism Spectrum Disorder in Infants Using Acoustic Feature

Area of Science:

• Biomedical Engineering
• Developmental Psychology
• Machine Learning

Background:

• Early detection of autism spectrum disorder (ASD) is crucial for timely intervention.
• Infant vocalizations offer a potential non-invasive avenue for early ASD screening.

Purpose of the Study:

• To develop and validate a machine learning model for early ASD detection in infants using acoustic features.
• To identify key acoustic biomarkers indicative of ASD in early infancy.

Main Methods:

• Prospective cohort study involving infants aged 9-18 months from an ASD sibling cohort.
• Acoustic feature extraction (4368 features) followed by LASSO regression for dimensionality reduction (39 features).
• Support vector machine (SVM) classifier development with tenfold cross-validation and evaluation of four kernel functions.

Main Results:

• The SVM model with a sigmoid kernel achieved high performance: 92.86% sensitivity, 93.33% specificity, and 93.18% accuracy.
• Significant differences in spectral and energy-related acoustic features were observed in infants diagnosed with ASD (p < 0.01).
• The study identified 39 key acoustic features for ASD prediction.

Conclusions:

• Acoustic features derived from infant vocalizations can serve as reliable, non-invasive biomarkers for early ASD detection.
• The developed SVM model shows significant potential for early ASD screening and facilitating timely intervention strategies.