Recognizing Physical Activities for Spinal Cord Injury Rehabilitation Using Wearable Sensors.

Area of Science:

• Biomedical engineering
• Rehabilitation science
• Machine learning for healthcare

Background:

• Activity recognition is crucial for rehabilitation but underutilized for spinal cord injury (SCI) patients.
• Patient self-reporting for adherence is unreliable and can impact rehabilitation outcomes.
• Objective monitoring of physical activities is needed to enhance SCI rehabilitation efficacy.

Purpose of the Study:

• To develop and implement a systematic activity recognition method for monitoring physical activities during SCI rehabilitation.
• To improve the accuracy and reliability of rehabilitation adherence assessment.
• To leverage machine learning for analyzing sensor data in SCI recovery.

Main Methods:

• A novel dynamic segmentation technique was employed to divide raw sensor data into fragments, outperforming traditional sliding window methods.
• A machine learning approach was adopted to build a predictive model for activity recognition.
• The method was validated using data from a single wrist-worn accelerometer.

Main Results:

• The proposed activity recognition method demonstrated high effectiveness in identifying all examined physical activities.
• The system achieved an excellent overall accuracy of 96.86% in activity recognition.
• Dynamic segmentation proved superior to sliding window techniques for this application.

Conclusions:

• The developed activity recognition method is effective for monitoring SCI rehabilitation.
• This approach offers a more objective and accurate way to assess patient adherence and progress.
• The findings support the integration of advanced activity recognition technologies in SCI recovery programs.