A New Approach For Monitoring Gait Balance Control Stability Using An IMU With Deep-Learning Techniques

PURPOSE: This study proposed a new approach based on a waist-worn inertial measurement unit (IMU) and deep learning techniques for monitoring and quantifying gait balance control stability in healthy young and older adults. METHODS: Thirteen healthy young and thirteen older adults wore an IMU on their waist and walked on a walkway while the motions of the centre of mass (COM) and centre of pressure (COP) were measured by a motion analysis system. The COM-COP coordination describing the balance control was quantified by the angular motions of the COM-COP vector about the vertical at the COP in terms of the phase angles (PA) of the inclination angles (IA) from the vertical and their rates of change (RCIA). The deviation phase (DP) of the PA curves described the coordination variability. A novel Physics-Informed Gated Recurrent Unit (PIGRU) model, a type of artificial neural network, was developed, trained and validated by the measured IAs and IMU data. With the IMU data as input, the PIGRU predicted IA and RCIA for calculating PA and DP. The relative root-mean-squared errors (rRMSEs) of model-predicted IA, RCIA and DP were calculated to indicate the model’s accuracy. Independent t-tests compared the DP between groups (α = 0.05). RESULTS: The means (SD) of the rRMSEs of the predicted IA, RCIA and DP were 3.1% (0.6%), 3.9% (1.0%), and 4.2% (1.2%), respectively. With the IMU data, the PIGRU model showed different COM-COP coordination patterns between groups (Fig. 1A), with the older group showing significantly greater DP during single limb support (Fig. 1B). CONCLUSION: This study proposed a novel approach based on a PIGRU model and a single waist-worn IMU for quantifying whole-body balance control during walking. The new approach demonstrated high prediction accuracy with an ability to distinguish between young and older adults, indicating a great potential for real-life monitoring of dynamic balance control stability during level walking.This research was funded by the Ministry of Science and Technology, Taiwan (MOST 110-2221-E-002-027-MY3).