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

Abstract

Shoulder problems (pain and pathology) are highly prevalent in manual wheelchair users with spinal cord injury. These problems lead to limitations in activities of daily life (ADL), labor- and leisure participation, and increase the health care costs. Shoulder problems are often associated with the long-term reliance on the upper limbs, and the accompanying “shoulder load”. To make an estimation of daily shoulder load, it is crucial to know which ADL are performed and how these are executed in the free-living environment (in terms of magnitude, frequency, and duration). The aim of this study was to develop and validate methodology for the classification of wheelchair related shoulder loading ADL (SL-ADL) from wearable sensor data. Ten able bodied participants equipped with five Shimmer sensors on a wheelchair and upper extremity performed eight relevant SL-ADL. Deep learning networks using bidirectional long short-term memory networks were trained on sensor data (acceleration, gyroscope signals and EMG), using video annotated activities as the target. Overall, the trained algorithm performed well, with an accuracy of 98% and specificity of 99%. When reducing the input for training the network to data from only one sensor, the overall performance decreased to around 80% for all performance measures. The use of only forearm sensor data led to a better performance than the use of the upper arm sensor data. It can be concluded that a generalizable algorithm could be trained by a deep learning network to classify wheelchair related SL-ADL from the wearable sensor data.

Learning Activities of Daily Living from Unobtrusive Multimodal Wearables: Towards Monitoring Outpatient Rehabilitation

“…Device D, capturing acceleration, heart rate (HR), and respiratory rate (RR) data on the chest, was deemed irrelevant for ADL classification in manual wheelchair users using the proposed methodology and features. This aligns with prior studies that solely utilized kinematic data for activity classification in manual wheelchair users, thereby excluding non-kinematic modalities like HR and RR [10], [11]. In [10] it was similarly demonstrated that non-kinematic data in the form of muscle activity do not aid in ADLs classification.…”

“…This aligns with prior studies that solely utilized kinematic data for activity classification in manual wheelchair users, thereby excluding non-kinematic modalities like HR and RR [10], [11]. In [10] it was similarly demonstrated that non-kinematic data in the form of muscle activity do not aid in ADLs classification. Additionally, device A1 (pressure mat -backrest), did not enhance the performance.…”

“…Real-life monitoring of ADLs in manual wheelchair users with an SCI has recently been explored [10], [11]. Both groups utilized data from inertial measurement units (IMUs) and a neural network to classify ADLs.…”

“…Moreover, their feature selection capability during training prevents overfitting in high-dimensional feature spaces, enhancing overall performance and interpretability. In [10] three body-worn and two wheelchair-attached IMU sensors were used for classifying eight shoulder-related ADLs. Such IMU-based sensory systems are frequently utilized for the classification of ADLs, nonetheless, these systems are limited in their ability to track vital signs and thus, do not furnish insight into the user's overall health and well-being.…”

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

“…To evaluate sensor data, especially in a sparse measurement setup, machine learning techniques are necessary [ 20 , 21 ]. In recent years, IMU and machine learning techniques, including artificial neural networks (ANN), have been successfully applied in both classification and prediction tasks in biomechanical settings [ 21 , 22 , 23 , 24 , 25 , 26 , 27 ]. Traditional machine learning algorithms such as Random Forest, Support Vector Machine, etc., require time-intensive feature engineering and manual feature extraction.…”