The Effect of Sensor Location on Estimation of Spatiotemporal Parameters of Gait

Sensor placement of IMUs and Apple devices at the trunk, wrist, thigh, shank, and foot locations.

Objective

An estimated 11% of U.S. adults have a mobility disability with that number expected to grow. A person's mobility has a large effect on their quality of life, thus evaluating abnormalities in movement is critical for rehabilitation and clinical decision-making. Current gait analysis methods, which include marker-based motion capture systems and force platforms, are typically expensive and limited to a laboratory setting.

New wearable sensing technologies, such as inertial measurement units (IMU), have emerged as a promising yet cost-effective alternative for laboratory gait assessments. However, there remains a need for more comprehensive studies to characterize IMU sensor performance across various IMU locations and their unique algorithms for spatiotemporal estimations. Accordingly, this study aims to evaluate the performance of IMUs at typical locations (trunk, thigh, shank, and foot) and various walking conditions (overground, treadmill, braced), and validate them against the gold-standard motion capture as well as two commercial wearable devices (Apple Watch and iPhone).

Design

Five healthy subjects were fitted with seven IMU sensors, an Apple Watch and an iPhone. The IMU sensors (XSENS, 100 Hz sampling frequency) were securely fixed to the subjects’ trunk (1), lateral thighs (2), lateral shanks (2), and feet (2) (over shoes) while the Apple products were fixated on the subject's right side waist (iPhone) and left wrist (Apple Watch). Before any tests were conducted, subjects were asked to walk a distance of 20 feet three times at a self-selected “comfortable” speed, which was recorded as baseline. Walking trials were recorded at various conditions as follows: three treadmill trials (baseline speed, 10% above baseline speed, 10% below baseline speed), one overground walking trial (baseline speed), and two walking trials with a knee brace at baseline (treadmill, overground). Each IMU location was processed separately to extract spatiotemporal parameters.

Spatiotemporal parameters were holistically selected based on clinical relevance and availability from the Apple Health App.

Application

We found that walking speed had an effect on spatial parameter accuracy. We expected accuracy to increase with speed for all parameters, but trends were only found for spatial parameters on two sensors. The Apple Watch was more accurate at slow speeds than fast speeds (RMSE 0.028 vs. 0.197 m/s, p = 0.02), while the Shank IMU was more accurate at fast speeds than slow speeds (RMSE 0.543 vs. 0.478 m/s, p = 0.04). Additionally, there was no significant difference between other walking condition combinations and almost all of the accuracy difference was due to IMU location. Across all walking conditions, sensors positioned further from the subjects’ COM were less accurate in estimating spatial parameters. In addition to spatial parameters, DST was also sensitive to sensor type but not walking conditions.

(A) The iPhone was the least accurate and the shank IMU was the most accurate for estimating temporal parameters relying on ground-contact events. (B) Sensors closer to the body’s center of mass were more accurate than those placed more distally for estimation of gait velocity.

Overall, walking conditions had a minimal effect on accuracy with the exception of walking speed on only the Watch and Shank IMU. Sensor location was the major determining factor of accuracy for both spatial and temporal estimations: sensors closest to the COM had the highest accuracy for spatial, while DST, the least accurate temporal parameter, was most accurate at the Shank IMU. Although previously unevidenced in biomechanics research, the decrease of spatial parameter accuracy for IMUs further from the body COM aligns with biomechanical and clinical understanding of gait. The Apple Watch and iPhone, representing the commercially-accessible market, provide acceptable results for regular daily activities, while an IMU pairing at the Shank and Trunk locations can provide substantial data for estimating the full range of spatial and temporal parameters in clinical and research use cases. With these novel findings, we hope to raise awareness of the emerging role of wearable technologies in health monitoring and enable the development of new patient monitoring systems.

Overall, consumer devices estimated gait velocity well, but sensors located at the shank (i.e., research-grade) are better for temporal parameters.

*This work was presented at the IEEE International Conference on Biomedical and Health Informatics (BHI ‘23); Poster presentation titled “The Effect of Sensor Location on Spatiotemporal Parameters of Gait.”

Page updated

Google Sites

Report abuse