Upper extremity post-stroke motion quality estimation with decision trees and bagging forests.

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Data Science

Background:

• Stroke is a primary cause of long-term disability, necessitating effective rehabilitation strategies.
• Understanding movement quality in real-world settings is crucial for optimizing recovery after stroke.
• Current methods may not fully capture the nuances of motor performance in post-stroke individuals.

Purpose of the Study:

• To develop and compare two computational approaches for quantifying upper extremity movement quality in stroke survivors.
• To establish a mathematical framework linking objective sensor data to clinical performance metrics.
• To evaluate the predictive accuracy of different machine learning models for analyzing rehabilitation data.

Main Methods:

• Utilized accelerometer and gyroscope sensor data to capture upper extremity movements.
• Developed a mathematical framework to translate sensor data into clinically relevant quality measures.
• Employed two distinct machine learning approaches: decision trees and Bootstrap Aggregating (Bagging) forests.
• Compared the performance and predictive ability of the two analytical methods.

Main Results:

• Both decision trees and Bagging forests could extract clinically meaningful quality measures from sensor data.
• Bootstrap Aggregating forest approaches demonstrated superior predictive ability compared to decision trees.
• This superiority was particularly evident with unstable datasets, characteristic of post-stroke data.

Conclusions:

• Objective quantification of movement quality using wearable sensors is feasible and valuable for stroke rehabilitation.
• Machine learning, specifically Bagging forests, offers a robust method for analyzing complex sensor data in this population.
• The findings suggest that advanced ensemble methods may be more effective for evaluating motor performance in individuals recovering from stroke.