Prolonged Monitoring Of Speed And Balance Control During Gait Using An IMU With Deep-Learning Techniques

PURPOSE: This study aims to develop a new approach based on deep learning techniques for prolonged monitoring of speed and balance control during gait using a waist-worn inertial sensor (IMU), and to use the proposed device to quantify the effects of speed on dynamic balance control variability. METHODS: Twenty-four healthy young adults wore an IMU on their waist and walked on a walkway at three different speeds: preferred walking speed (PWS), slower walking speed, and faster walking speed. The motions of the center of mass (COM) and center of pressure (COP) were recorded to quantify dynamic gait balance by calculating the COM-COP inclination angles (IA) relative to the vertical and their rates of change (RCIA). These data, along with the IMU data, were used to train and validate a seq2seq model. The accuracy of the model's predictions on speeds and IA-related variables was described by relative root-mean-square errors (rRMSE). For each subject, phase angles (PA) at each speed were obtained from the phase plots of the model-predicted IA. The deviation phase (DP) of the PA was then calculated to quantify gait balance variability, with a lower DP indicating smaller variability. The effects of speed on DP values were analyzed for both double-limb support and single-limb support phases. RESULTS: The proposed model showed high accuracy for speed and IA-related variables, with mean rRMSEs for the model-predicted speed, IA, and RCIA being 6.3%, 3.5%, and 4.1%, respectively (Fig. 1). Balance control variability was significantly affected by gait speed, with lowest DP values observed during double-limb support at PWS (Fig. 1). CONCLUSION: This study is the first to show the feasibility of simultaneously monitoring speed and balance control during gait using an IMU. The proposed approach showed high accuracy and holds significant potential for prolonged, real-life monitoring of gait speed and balance control variability in natural environments for both research and clinical applications. Supported by: This study was funded by the National Science and Technology Council, Taiwan (NSTC 110-2221-E-002-027-MY3 and 113-2221-E-002 -043 -MY3)