Characterization of Infants' General Movements Using a Commercial RGB-Depth Sensor and a Deep Neural Network Tracking...

Area of Science:

• Neurology
• Biomedical Engineering
• Developmental Pediatrics

Background:

• Cerebral palsy is the most common childhood neuromotor disorder.
• Diagnosis often relies on visual assessment of general movements (GM) in infancy, requiring specialized training.
• Large-scale implementation of GM assessment is challenging due to training requirements.

Purpose of the Study:

• To explore the feasibility of evaluating infant general movements (GM) in a home environment using automated analysis.
• To assess if quantitative metrics derived from 3D movement trajectories can predict movement disorders.
• To investigate the use of low-cost RGB-D sensors for in-home infant movement analysis.

Main Methods:

• Utilized a commercial RGB-D sensor to record infant general movements (GM) in a home setting.
• Applied open-source markerless motion tracking to estimate 2D trajectories of points of interest (PoI).
• Developed a novel method to reconstruct 3D PoI trajectories using depth sensor data.
• Calculated nine GM metrics (eight established, one novel) from PoI trajectories at 3, 4, and 5 months of age.
• Compared automated metrics against clinical evaluations by pediatric specialists.

Main Results:

• Successfully estimated meaningful GM metrics from 3D infant movement trajectories recorded at home.
• Demonstrated the potential of these metrics for early identification of movement disorders.
• The automated analysis showed promise in evaluating infant GM in a familiar environment.

Conclusions:

• Automated analysis of infant general movements (GM) using low-cost sensors and 3D trajectory reconstruction is feasible in a home setting.
• Quantitative GM metrics derived from this method show potential for early detection of movement disorders, including cerebral palsy.
• This approach may facilitate large-scale, accessible screening for infant neuromotor conditions.