An Algorithmic Approach for Detecting Neuromotor Developmental Disabilities in Infants from Wearable Sensor Data

Area of Science:

• Developmental Pediatrics
• Machine Learning in Healthcare
• Wearable Sensor Technology

Background:

• Recruiting infants for health research is challenging, limiting dataset size for conventional machine learning (ML).
• Previous studies on limited infant motion data raise reliability concerns for ML model performance.
• Infant developmental trajectories require reliable methods for early identification and prediction.

Purpose of the Study:

• To differentiate infants at-risk (AR) for developmental disabilities from typically developing (TD) infants using motion data.
• To predict developmental outcomes in at-risk infants at 24 months using wearable sensor data.
• To develop a robust analytical approach for small, sensitive pediatric datasets.

Main Methods:

• Analysis of full-day motion recordings from TD infants (N=12) and AR infants (N=24).
• Introduction of a novel algorithmic approach to extract 'Motifs' from raw motion signals.
• Utilizing Motif abundance as indicators for group differentiation and outcome prediction.

Main Results:

• Motif analysis effectively distinguished AR infants from TD infants.
• The approach demonstrated capability in predicting developmental outcomes for AR infants.
• Novel Motif extraction proved reliable even with a limited dataset size.

Conclusions:

• Motif-based analysis offers a promising, reliable method for infant developmental research with limited data.
• This approach enhances the potential of wearable technology for early identification of developmental risks.
• Future research can leverage Motif analysis for larger-scale pediatric health studies.