A new wearable stretch sensor array for 3D spine model visualization during therapeutic exercise

Mobile visualization of the spine during therapeutic exercise may unlock the potential benefits of biofeedback for home-based therapy and fitness programs. In this paper we present the design and validation of a new approach to wearable sensors to close the digital-physical gap between sensor data and 3D spine posture by supporting animation of a customizable 3D thoracolumbar spine model that may eventually enable a mobile, virtual reality (VR) visual biofeedback. First, we propose a linear model of the dependency between four stretch signals from the dorsal surface and the angular positions of the thoracolumbar spine on the sagittal, coronal, and transverse planes. We then describe validation experiments and demonstrate the concept by animating the spine model with monoaxial, biaxial and tri-axial motions. The linear model was validated using a 3-way comparison of exercise video, sensor signals, and spine model animation. The computed angular positions were consistent with the video recording, and the animation of the model was visually accurate with a mean absolute error of 3.62 ° for single axis motions and 8.74 ° for dual axis motions. At the end, we provide a discussion on improvements to the linear model performance for cross-axial interactions, plus an outlook of future work.