Validation of Body-Worn Sensors for Gait Analysis During a 2-min Walk Test in Children

Abstract

Introduction: Instrumented gait mat systems have been regarded as one of the gold standard methods for measuring spatiotemporal gait parameters. However, their portable walkways confine walking to a restricted area and limit the number of gait cycles collected. Wearable inertial sensors are a potential alternative that allow more natural walking behavior and have fewer space restrictions. The objective of this pilot study was to establish the concurrent validity of body-worn sensors against the portable walkway system in older children.

Methods: Twenty-one participants (10 males) 7-17 years old performed 2-min walk tests at a self-selected and fast pace in a 25-m-long hallway, while wearing three inertial sensors. Data collection were synchronized between devices and the portions of the walk when subjects passed on the walkway were used to compare gait speed, stride length, gait cycle duration, cadence, and double support time. Regression models and Bland-Altman analysis were completed to determine agreement between systems for the selected gait parameters.

Results: Gait speed, cadence, gait cycle duration, and stride length as measured by inertial sensors demonstrated strong agreement overall. Double support time was found to have lower validity due to a combined bias of age, height, weight, and walking pace.

Conclusion: These results support the validity of wearable inertial sensors in measuring gait speed, cadence, gait cycle duration, and stride length in children 7 years old and above during a 2-min walking test. Future studies are warranted with a broader age range to thoroughly represent the pediatric population.

Keywords: accelerometer; pediatrics; wearable technology.

Figure 1 —

This diagram is a representation (not to scale) of the experimental setup. The GAITRite mat was placed in the middle of the 25-m hallway and connected to a computer with the GAITRite software via the signal box. APDM Mobility Lab released an output trigger from the Opal access point to the GAITRite software via the signal box.

Figure 2 —

Bland–Altman plots (a) compare GAITRite data with APDM Opal inertial sensor measurements and Forest plots (b) present regression coefficients of gait speed (in meters per second). In the Bland–Altman plots, dashed lines mark the two SD boundary of the GAITRite—Opal bias. Square and triangle scatter points represent FP and self-selected RP, respectively. In the forest plots of the regression coefficients, standardized effect sizes are presented where the means are denoted by open circles. The thick and thin lines represent 50% and 95% posterior credible intervals, respectively. FP = fast pace; RP = regular pace.

Figure 3 —

Bland–Altman plots (a) compare GAITRite data with APDM Opal inertial sensor measurements and Forest plots (b) present regression coefficients of cadence (in steps per minute). Scatter points decorated by pace. In the Bland–Altman plots, dashed lines mark the two SD boundary of the GAITRite—Opal bias. Square and triangle scatter points represent FP and self-selected RP, respectively. In the forest plots of the regression coefficients, standardized effect sizes are presented where the means are denoted by open circles. The thick and thin lines represent 50% and 95% posterior credible intervals, respectively. FP = fast pace; RP = regular pace.

Figure 4 —

Bland–Altman plots (a) compare GAITRite data with APDM Opal inertial sensor measurements and Forest plots (b) present regression coefficients of gait cycle duration (in seconds). In the Bland–Altman plots, dashed lines mark the two SD boundary of the GAITRite—Opal bias. Square and triangle scatter points represent FP and self-selected RP, respectively. In the forest plots of the regression coefficients, standardized effect sizes are presented where the means are denoted by open circles. The thick and thin lines represent 50% and 95% posterior credible intervals, respectively. FP = fast pace; RP = regular pace.

Figure 5 —

Bland–Altman plots (a) compare GAITRite data with APDM Opal inertial sensor measurements and Forest plots (b) present regression coefficients of stride length (in meters). In the Bland–Altman plots, dashed lines mark the two SD boundary of the GAITRite—Opal bias. Square and triangle scatter points represent FP and self-selected RP, respectively. In the forest plots of the regression coefficients, standardized effect sizes are presented where the means are denoted by open circles. The thick and thin lines represent 50% and 95% posterior credible intervals, respectively. FP = fast pace; RP = regular pace.

Figure 6 —

Bland–Altman plots (a) compare GAITRite data with APDM Opal inertial sensor measurements and Forest plots (b) present regression coefficients of double support time (%GCT). In the Bland–Altman plots, dashed lines mark the two SD boundary of the GAITRite—Opal bias. Square and triangle scatter points represent FP and self-selected RP, respectively. In the forest plots of the regression coefficients, standardized effect sizes are presented where the means are denoted by open circles. The thick and thin lines represent 50% and 95% posterior credible intervals, respectively. FP = fast pace; RP = regular pace; %GCT = percentage of gait cycle time.