Application of Inertial Measurement Units and Machine Learning Classification in Cerebral Palsy: Randomized...

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Machine Learning in Healthcare

Background:

• Cerebral palsy (CP) affects movement and posture, impacting millions globally and in Australia.
• Traditional tools like goniometers and inclinometers are used for joint angle measurement in CP research.
• Current methods can be time-consuming and challenging, particularly for pediatric populations.

Purpose of the Study:

• To develop a digital solution for mass data collection in children with CP using inertial measurement units (IMUs).
• To apply machine learning (ML) algorithms to classify CP movement features and assess therapy effectiveness.
• To reduce the time required for data classification by eliminating the need for Euler, quaternion, and joint measurements.

Main Methods:

• Custom IMUs were developed to record wrist movements in two age groups (approaching 3 and 15 years) of participants with and without CP.
• IMU data were utilized to calculate wrist joint angles and range of motion.
• Nine ML algorithms were employed to classify CP-associated movement features and evaluate treatment efficacy (e.g., wrist extension).

Main Results:

• Wrist joint angle calculations were successfully performed and validated against Vicon motion capture.
• ML algorithms classified CP movement features using raw IMU data.
• The Random Forest algorithm achieved 87.75% accuracy for the older age group, while C4.5 decision tree achieved 89.39% for the younger group.

Conclusions:

• IMUs show potential for accurate active range of motion data collection in children with CP, overcoming challenges of goniometric methods.
• Positive anecdotal feedback suggests IMUs could be valuable for ongoing hand movement monitoring in children.
• The developed digital solution offers a promising avenue for efficient and accurate CP assessment and therapy evaluation.