Monitoring wheelchair propulsion patterns: feasibility and validity of using wearable sensors.

Currently, there is a need to understand the characteristics of manual wheelchair propulsion patterns in the daily life of users and the impact of these patterns on repetitive strain injury of the shoulders. This study aimed to develop a method for remote and long-term monitoring of wheelchair propulsion techniques. We used a hand-mounted inertial measurement unit (IMU) to identify propulsion patterns in manual wheelchair users. IMU data was collected from 12 participants (7 males and 5 females), including 8 experienced and 4 inexperienced manual wheelchair users. We applied continuous wavelet transform (CWT) for feature extraction and used Support Vector Machine (SVM) and Multilayer Perceptron (MLP) Neural Network for pattern classification. SVM with a linear kernel achieved 89% accuracy, 78% F1-score, 78% precision, and 78% recall. SVM with a polynomial kernel achieved 94% accuracy, 88% F1-score, 88% precision, and 89% recall, while the MLP reached 95% accuracy, 89% F1-score, 89% precision, and 89% recall. Neither the participants’ wheelchair experience nor their gender significantly affected the performance of the classifiers. These findings suggest that the proposed IMU and propulsion patterns classification method can be used across different user profiles for remote and long-term monitoring of wheelchair propulsion patterns to better understand shoulder overuse risk in daily life.