A Stroke Specific Neural Network to Estimate Continuous Gait Speed Using a Single Inertial Measurement Unit

Gait speed is a key metric of gait function and recovery in stroke survivors. This study presents a novel approach for estimating gait speed using a single inertial measurement unit (IMU) placed on the pelvis and a Nonlinear Autoregressive Network with Exogenous Inputs (NARX). To validate the network, datasets from two distinct cohorts of stroke survivors were collected: a chronic group of 40 community-dwelling individuals, and an inpatient group of 42 participants recruited from an inpatient rehabilitation facility (IRF). The NARX model was initially developed using data from the chronic group and subsequently refined using data from the inpatient group. The model's performance was evaluated using data from the 10-Meter Walk Test (10MWT) and the 6-Minute Walk Test (6MWT). Results demonstrated high temporal alignment between the NARX-estimated gait speeds and those derived from a reference system (i.e., Xsens MVN) during the 10MWT, with a median cross-correlation coefficient of 0.98 and a median mean squared error of 0.05 (m/S)<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>. When applied to the 6MWT, the NARX model accurately estimated the total distance walked with a median absolute error of 41.66 m. The proposed method enables continuous gait speed and distance monitoring in stroke survivors with minimal setup using a single IMU. This approach would facilitate in-person and remote patient monitoring, offering valuable insights for personalized rehabilitation strategies for stroke survivors in IRFs.