Analysis of Machine Learning-Based Assessment for Elbow Spasticity Using Inertial Sensors

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Machine Learning in Healthcare

Background:

• Spasticity is a common neurological impairment affecting limb movement and physical rehabilitation outcomes.
• Clinical assessment of spasticity often relies on subjective measures like the modified Ashworth scale (MAS).
• Objective, quantitative methods are needed for precise spasticity evaluation.

Purpose of the Study:

• To develop and validate a machine learning-based method for quantifying elbow spasticity severity.
• To compare the performance of various machine learning algorithms using wearable sensor data.
• To provide a clinically comparable alternative to the MAS for spasticity assessment.

Main Methods:

• Collected inertial data (acceleration, rotation) from patients' elbows using wearable inertial measurement units during passive stretch tests.
• Employed machine learning algorithms including decision trees, random forests (RFs), support vector machines, linear discriminant analysis, and multilayer perceptrons.
• Evaluated algorithms using different segmentation techniques and feature sets to classify spastic movement degree.

Main Results:

• A Random Forest (RF) algorithm achieved high accuracy, reaching up to 95.4% in classifying spastic movement severity.
• The proposed method demonstrated the potential to generate a clinically meaningful index of spasticity.
• Wearable technology combined with machine learning proved effective for objective spasticity measurement.

Conclusions:

• Wearable sensors and machine learning offer a viable approach for objective and accurate spasticity assessment.
• This technology can empower patients to monitor their spasticity levels outside clinical settings.
• The developed method provides a promising tool for enhancing physical rehabilitation and patient self-management.