Machine-Learning-Based Methodology for Estimation of Shoulder Load in Wheelchair-Related Activities Using Wearables

Sensors, 2023 · DOI: 10.3390/s23031577 · Published: February 1, 2023

Simple Explanation

This study focuses on estimating shoulder load in manual wheelchair users (MWUs) during daily activities using wearable sensors and machine learning. The goal is to create a non-laboratory-based method for continuous shoulder load monitoring. Researchers collected data from able-bodied participants performing typical MWU activities while wearing inertial measurement units (IMUs) and electromyography (EMG) sensors. These sensors recorded movement and muscle activity. A neural network was trained to predict shoulder load based on sensor data, and the predictions were compared against shoulder load calculated using musculoskeletal modeling. The study explored different sensor setups and machine learning models to find the most accurate approach.

Study Duration

Not specified

Participants

10 able-bodied participants

Evidence Level

Not specified

Key Findings

1
A subject-specific biLSTM model trained on a sparse sensor setup (upper arm IMU, WC IMUs, and EMG) yielded the most promising results, with a mean correlation coefficient of 0.74 ± 0.14 and a relative root-mean-squared error of 8.93% ± 2.49%.
2
Shoulder-load profiles predicted by the model had a high mean similarity of 0.84 ± 0.10 with those determined by musculoskeletal modeling, indicating good agreement in the overall shoulder loading patterns.
3
The study demonstrates the feasibility of using wearable sensors and neural networks to estimate shoulder load during wheelchair-related activities, offering a potential tool for monitoring and intervention.

Research Summary

This study developed a machine-learning-based methodology to estimate shoulder load in wheelchair-related activities using wearable sensors. The approach involved collecting data from participants performing daily activities while equipped with IMUs and EMG sensors. A neural network was trained to predict shoulder joint reaction forces (SJRF) based on sensor data, and the model's performance was evaluated using various cross-validation strategies, sensor setups, and model architectures. The final model, a subject-specific biLSTM trained on a sparse sensor setup, achieved promising results in estimating shoulder load, demonstrating the potential of this methodology for real-world monitoring and intervention.

Practical Implications

Personalized Shoulder Load Monitoring

The developed methodology could enable continuous monitoring of shoulder load in MWUs during daily life, allowing for personalized interventions to prevent shoulder pain and injuries.

Activity-Specific Load Analysis

By identifying activities associated with high shoulder load, targeted strategies can be developed to reduce the risk of joint overload during specific tasks.

Remote Rehabilitation and Prevention

The wearable sensor-based approach allows for remote monitoring and intervention, improving access to rehabilitation services and promoting proactive shoulder health management.

Study Limitations

1
The methodology was developed using data from able-bodied participants, requiring validation with spinal-cord-injured individuals.
2
The ground-truth SJRF was based on musculoskeletal modeling, which relies on assumptions and may have limitations in accuracy.
3
A generalizable algorithm would be preferable to a subject-speciﬁc algorithm as it is less resource-intensive.

Your Feedback

Was this summary helpful?

Back to Bioinformatics