Balance Assessment Using a Handheld Smartphone with Principal Component Analysis for Anatomical Calibration

Abstract

Most balance assessment studies using inertial measurement units (IMUs) in smartphones use a body strap and assume the alignment of the smartphone with the anatomical axes. To replace the need for a body strap, we have used an anatomical alignment method that employs a calibration maneuver and Principal Component Analysis (PCA) so that the smartphone can be held by the user in a comfortable position. The objectives of this study were to determine if correlations existed between angular velocity scores derived from a handheld smartphone with PCA functional alignment vs. a smartphone placed in a strap with assumed alignment, and to analyze acceleration score differences across balance poses of increasing difficulty. The handheld and body strap smartphones exhibited moderately to strongly correlated angular velocity scores in the calibration maneuver (r = 0.487-0.983, p < 0.001). Additionally, the handheld smartphone with PCA functional calibration successfully detected significant variance between pose type scores for anteroposterior, mediolateral, and superoinferior acceleration data (p < 0.001).

Keywords: Principal Component Analysis; accessibility; anatomical calibration; balance assessment; inertial measurement unit (IMU); smartphone; wearables.

Figure 1

Participant performing balance pose during experiment while wearing a smartphone in a body strap with assumed alignment with the participant’s anatomical axes. Smartphone was aligned on the participant’s back with the phone’s Z-axis facing in the participant’s anterior direction and Y-axis in the participant’s superior direction.

Figure 2

(a) Handheld smartphone held against chest, with participant’s anatomical axes displayed. (b) Overlay of handheld smartphone with calculated anteroposterior (AP, shown in blue), mediolateral (ML, shown in red), and superoinferior (SI, shown in yellow) axes after calibration maneuver PCA analysis.

Figure 3

Participants performed the forward flexion maneuver before every balance pose with handheld and body strap smartphones (not pictured) recording inertial data. (a) Participant standing in upright position. (b) Participant bent about their ML axis in forward-flexion. The forward-flexion calibration maneuver consists of moving from position (a) to (b) and back to (a).

Figure 4

Images of three balance poses of increasing difficulty. (a) Double-leg stance (DL); expected to be least difficult. (b) Tandem stance (T); expected to be of moderate difficulty. (c) Single-leg stance (SL); expected to be most difficult.

Figure 5

Smartphone manual data trimming. (a) Beginning of calibration maneuver. (b) End of calibration maneuver. (c) Beginning of jump, end of balance pose. Also shown, 30 s balance pose between points (b) and (c), and trimming to 24 s period starting 3 s after (b) and before (c). Figure key shows color coding for X, Y, and Z axis inertial data.

Figure 6

22 AP-Z RMS angular velocity scores between handheld and body strap smartphone. A positive trendline was produced (r = 0.491, p < 0.001).

Figure 7

22 ML-X RMS angular velocity scores between handheld and body strap smartphone. A positive trendline was produced with the calibration method (r = 0.983, p < 0.001).

Figure 8

22 SI-Y RMS angular velocity scores between handheld and body strap smartphone. A positive trendline was produced with the calibration method (r = 0.487, p < 0.001).

Figure 9

RMS acceleration scores in each direction and for each balance pose for the handheld smartphone; mean and 1 standard deviation shown. ** = significant differences (p < 0.0167) between two or more balance pose pairs in post-hoc Dunn’s pairwise comparisons.

Figure 10

RMS acceleration scores in each direction and for each balance pose for the body strap smartphone; mean and 1 standard deviation shown. * = significant differences (p < 0.001) between all 3 balance poses in Kruskal–Wallis test. ** = significant differences (p < 0.0167) between two or more balance pose pairs in post-hoc Dunn’s pairwise comparisons.

Figure 11

Sample of angular velocity data during the forward-flexion maneuver before (left) and after (right) PCA functional alignment of the handheld smartphone. (Left) Filtered angular velocity data from the handheld smartphone in the inertial X, Y, and Z axes prior to PCA calibration. (Right) Data in the AP, ML, and SI axes after PCA calibration.