Machine learning approaches in automated infant General Movements Assessment: A scoping review

Area of Science:

• Medical Imaging and Machine Learning
• Developmental Pediatrics
• Neurology

Background:

• Automated General Movements Assessment (GMA) using machine learning (ML) and deep learning (DL) is emerging for early cerebral palsy (CP) detection.
• Existing literature requires synthesis to understand factors influencing automated GMA performance.

Purpose of the Study:

• To evaluate and synthesize ML and DL methodologies in automated infant GMA for early CP prediction.
• To link methodological choices to performance variations in automated GMA.

Main Methods:

• A scoping review was conducted following PRISMA/PRISMA-ScR guidelines.
• Literature from 2015-2025 was searched across major databases.
• Included studies were analyzed for factors influencing GMA classification performance.

Main Results:

• Forty-three studies met inclusion criteria, predominantly using DL for pose reconstruction and ML/DL for classification.
• Smaller datasets achieved higher accuracy (up to 100%) in general movement classification than larger ones.
• All studies reported >90% accuracy for CP classification.

Conclusions:

• DL methods generally outperform ML, but performance varies due to data quality, recording conditions, and methodological choices.
• This review provides a comprehensive synthesis linking methodology to performance, generalizability, and interpretability.
• Offers a framework and guidance for developing robust automated GMA systems for early CP detection.