Recognizing Physical Activities for Spinal Cord Injury Rehabilitation Using Wearable Sensors

Sensors, 2021 · DOI: https://doi.org/10.3390/s21165479 · Published: August 14, 2021

Assistive Technology Neurorehabilitation Bioinformatics

Simple Explanation

This paper introduces an activity recognition method for monitoring rehabilitation exercises of individuals with spinal cord injury using wearable sensors. The method uses raw sensor data divided into fragments using a dynamic segmentation technique, offering improved recognition performance. A machine learning approach was used to develop the method and build a predictive model.

Study Duration

Not specified

Participants

10 healthy individuals (3 male, 7 female; aged 29 ± 5.5 years)

Evidence Level

Not specified

Key Findings

1
The proposed method achieved an overall accuracy of 96.86% in recognizing physical activities.
2
Random Forest (RF) outperformed other classifiers, demonstrating high effectiveness in recognizing activities.
3
The dynamic segmentation technique enabled higher classification performance compared to the sliding window approach.

Research Summary

This study introduces a novel application of activity recognition to assist in the rehabilitation of SCI patients. The empirical results indicate the effectiveness of the method in recognizing all the activities considered in the study. Compared to the sliding window, our approach achieved higher performance in recognizing all of the physical activities under study.

Practical Implications

Telerehabilitation Programs

The method's effectiveness and applicability in healthcare self-management makes it suitable for telerehabilitation programs.

Automated Rehabilitation Assessments

Data from wearable sensors, used with classification algorithms, contributes to a valuable technology for performing automated rehabilitation assessments.

Overcoming Self-Reported Measures Limitations

The method overcomes the limitation of patients’ self-reported measures and surveys to verify the completeness of rehabilitation activities.

Study Limitations

1
Only time-domain features were used, limiting potential improvements from frequency-domain analysis.
2
The absence of a feature selection method could lead to less efficient implementation of the predictive model.
3
Additional signal processing techniques are needed for attenuating artifacts from body movements.

Your Feedback

Was this summary helpful?

Back to Assistive Technology