Toward Predicting Infant Developmental Outcomes From Day-Long Inertial Motion Recordings

Area of Science:

• Developmental Pediatrics
• Machine Learning in Healthcare
• Infant Motor Development

Background:

• Infant mortality rates are decreasing, leading to more infants with neuromotor challenges surviving.
• Motor, social, and cognitive development are interconnected, with challenges in one area impacting others.
• Early identification of developmental differences is crucial for timely intervention.

Purpose of the Study:

• To classify infants as typically developing (TD) or at-risk (AR) using movement data.
• To predict future developmental outcomes for at-risk infants.
• To assess the contribution of inertial motion data to classification and prediction accuracy.

Main Methods:

• Collected day-long inertial motion recordings from 12 TD infants and 24 AR infants.
• Employed machine learning classifiers (decision trees, k-nearest neighbors, support vector machines).
• Used movement recordings and demographic data for classification and outcome prediction.

Main Results:

• Achieved 94.4% accuracy in classifying infants as TD or AR.
• Attained 89.5% accuracy in predicting future outcomes for AR infants.
• Inertial motion data significantly improved prediction accuracy compared to demographic data alone.

Conclusions:

• Machine learning analysis of infant movement is a promising tool for monitoring developmental progress.
• Early identification of infants at risk for developmental delays can be enhanced through motion analysis.
• This approach supports stakeholders in identifying infants who may benefit from early intervention.