Deep learning model to recognize the different progression condition patterns of manual wheelchair users for prevention of shoulder pain

Wheelchair user will use different propulsion strategies to control in a variety of progression conditions that may induce the shoulder pain. The hypothesis of this study is that wheelchair user in different progression conditions has different ways to control the wheelchair. The purpose of this study is to use the accelerometer to recognize the movement of the wheelchair. It can be easily used to define the different progression condition in order to know the cause of the inducement of shoulder pain. The researchers collected acceleration data during the wheelchair progression in rough and smooth distinguishing surfaces: (1) outdoor grassland and; (2) indoor flatland. Researchers transformed the acceleration data into spectrogram files and training convolutional neural network (CNN) deep learning model to accurately recognize and predict wheelchair user’s wheelchair location. As the results, the wheelchair user’s medial-lateral direction of acceleration is expected to present more significant features than the front-back motion when being related to progression condition. At the same time, the vertical direction of acceleration also reflected the wheelchair vibration during different surface of progression condition.