Classiﬁcation of Wheelchair Related Shoulder Loading Activities from Wearable Sensor Data: A Machine Learning Approach

Sensors, 2022 · DOI: 10.3390/s22197404 · Published: September 29, 2022

Assistive Technology Bioinformatics Rehabilitation

Simple Explanation

Shoulder problems are common in manual wheelchair users, leading to limitations in daily activities and increased healthcare costs. These issues are linked to long-term upper limb reliance and "shoulder load." Estimating daily shoulder load requires knowing which activities are performed and how they are executed. This study aims to develop a method for classifying wheelchair-related shoulder-loading activities using wearable sensor data. The researchers trained deep learning networks on sensor data from participants performing relevant activities. The algorithm showed high accuracy in classifying these activities, suggesting it can be used to estimate daily shoulder load.

Study Duration

Not specified

Participants

10 able bodied participants

Evidence Level

Not specified

Key Findings

1
The trained deep learning algorithm achieved an overall accuracy of 98% in classifying wheelchair-related shoulder-loading activities (SL-ADL).
2
Omitting EMG data did not negatively impact the performance of the trained algorithms, suggesting it is not required for classification.
3
Using data from only one sensor (forearm) led to slightly better performance than using data from the upper arm sensor, although both were worse than using all sensors.

Research Summary

This study developed and validated a methodology for classifying wheelchair-related shoulder-loading activities (SL-ADL) using wearable sensor data and deep learning. A deep learning model, incorporating GRU and biLSTM layers, achieved high accuracy (over 98%) in classifying SL-ADL based on data from multiple sensors. The study found that EMG data was not essential for classification, and while using data from all five sensors yielded the best results, a single forearm sensor still provided reasonable performance.

Practical Implications

Shoulder Load Monitoring

Enables real-life monitoring of shoulder load in manual wheelchair users, which is crucial for preventing shoulder problems.

Activity-Specific Interventions

Facilitates the identification of specific activities contributing to high shoulder load, allowing for targeted interventions.

Sensor Selection Guidance

Provides guidance on the optimal sensor setup for classifying wheelchair-related activities, balancing accuracy and practicality.

Study Limitations

1
The study used able-bodied participants trained in wheelchair activities, which may not fully represent experienced wheelchair users.
2
The data was collected under controlled laboratory conditions, which may not reflect the variability of real-life settings.
3
The algorithm classifies data into eight predefined activities, not accounting for other activities performed in real-life settings.

Your Feedback

Was this summary helpful?

Back to Assistive Technology