Caregivers experience lower back pain due to their awkward postures while handling patients. Therefore, a monitoring system to supervise caregivers' postures using wearable sensors is being developed. This study proposed a postural recognition method for caregivers during postural change while handling a patient on a bed. The proposed method recognizes foot positions and arm movements by a machine learning algorithm using inertial data on the trunk and foot pressure data obtained from wearable sensors. An experiment was conducted to evaluate whether the proposed method could recognize three foot positions and three arm movements. Participants provided postural change for a simulated patient on a bed (patient: supine to lateral recumbent) under nine conditions, including different combinations of the three foot positions and three arm movements; the experiment was repeated ten times for each condition. Experimental results showed that the proposed method using an artificial neural network with all features obtained from an inertial measurement unit and insole pressure sensors could recognize arm movements and foot positions with an accuracy of approximately 0.75 and 0.97, respectively. These results suggest that the proposed method can be used in a monitoring system tracking a caregiver's posture.

Because caregivers often experience lower back pain caused by lumbar load from patient handling, monitoring this load can help prevent pain. Erector spinae muscle activity, which is measured and monitored as lumbar load, is commonly measured by electromyography (EMG). However, EMG’s electrodes can cause skin irritation and be uncomfortable. Therefore, measuring muscle activity without electrodes is necessary. In this study, we propose a method for estimating erector spinae muscle activity using wearable sensors, specifically inertial and shoe-type force sensors. Inertial sensors measure acceleration and angular velocity of the trunk. Shoe-type force sensors measure vertical force of the feet. A regression model obtained from a machine learning algorithm can predict erector spinae muscle activity using inertial and force data. In our experiment, we evaluated the accuracy of our method by comparing sensor data with surface EMG data in patient handling. Results show that this method can measure erector spinae muscle activity with a small error (less than 5% maximal voluntary contractions) and a significantly high correlation with actual value (r = 0.891, p <0.05). In addition, a Bland-Altman plot showed no fixed and proportional errors. These findings indicate that our proposed method can accurately monitor the lumbar loads of caregivers.