Predicting losses of balance during upright stance: evaluation of a novel approach based on wearable accelerometers

The study of postural sway during quiet stance has proved to be a useful approach to investigate the function of the balance system. Recent studies have suggested that providing information on postural sway to vestibular patients through biofeedback may improve their balance awareness and therefore reduce their risk of falling. One drawback common to these approaches is related to timing: informing a patient about current balance conditions may not allow enough time to react and avoid a fall. Here we propose a new technique for predicting relevant balance related events based on the recording of inertial information on trunk and thigh movement using wearable devices. We have developed a regressive model for the prediction of quiet stance dynamics of the center of body mass (CM), based on these sensory data. Our preliminary results show that, with careful signal processing, such approach may allow to learn quiet stance dynamics based on the inverted pendulum model and use it in predicting critical balance conditions with a few hundreds of milliseconds advance. When these predictions are then used for event-detection the system provides accurate results and is thus promising for the development of a fall prevention device.