Measuring functional hand use in children with unilateral cerebral palsy using accelerometry and machine learning

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Pediatric Neurology

Background:

• Unilateral cerebral palsy (CP) significantly impacts functional hand use in children.
• Objective and quantitative methods are needed to assess hand function in daily activities.
• Current assessment methods may not fully capture the nuances of functional hand use during naturalistic tasks.

Purpose of the Study:

• To investigate the efficacy of wearable sensors for measuring functional hand use in children with unilateral CP.
• To compare machine learning algorithms with traditional activity count methods for analyzing hand function data.
• To determine the feasibility of using accelerometry for quantitative monitoring of hand use.

Main Methods:

• Collected dual wrist-worn accelerometry data from 10 children with unilateral CP during play sessions.
• Utilized video observers to label functional and non-functional hand use instances.
• Compared machine learning models against activity count approaches for classifying hand movements.

Main Results:

• The best-performing machine learning model achieved high precision and recall (F1=0.896) when trained individually, correlating strongly (r=0.990) with video observations.
• Individualized machine learning models demonstrated significant correlation and strong agreement with expert video analysis.
• Performance decreased significantly (F1=0.584) in a leave-one-subject-out validation, indicating a need for robust individualization.
• The activity count method failed to differentiate functional from non-functional hand use and showed no significant correlation with video observations.

Conclusions:

• Wearable accelerometry combined with machine learning holds potential for quantitative monitoring of functional hand use in children with unilateral CP.
• Further development is needed to enhance the generalizability of machine learning models.
• Traditional activity count methods are insufficient for distinguishing functional from non-functional hand use in this population.