Kinect-based assessment of lower limb kinematics and dynamic postural control during the star excursion balance test

Gait & Posture 58 (2017) 421–427

Contents lists available at ScienceDirect

Gait & Posture journal homepage: www.elsevier.com/locate/gaitpost

Full length article

Kinect-based assessment of lower limb kinematics and dynamic postural control during the star excursion balance test

MARK

Moataz Eltoukhya, Christopher Kuenzeb, Jeonghoon Oha, Savannah Wootena, ⁎ Joseph Signorilea,c, a b c

Department of Kinesiology and Sport Sciences, School of Education & Human Development, University of Miami, Coral Gables, FL 33143, USA Department of Kinesiology, School of Education, Michigan State University, East Lansing, MI 48824, USA Center on Aging, Miller School of Medicine,1695 N.W. 9th Avenue, Suite 3204, Miami, FL, 33136, USA

A R T I C L E I N F O

A B S T R A C T

Keywords: Postural control SEBT Microsoft kinect

Assessments using dynamic postural control tests, like the Star Excursion Balance Test (SEBT), in combination with three-dimensional (3D) motion analysis can yield critical information regarding a subject’s lower limb movement patterns. 3D analysis can provide a clear understanding of the mechanisms that lead to specific outcome measures on the SEBT. Currently, the only technology for 3D motion analysis during such tests is expensive marker-based motion analysis systems, which are impractical for use in clinical settings. In this study we validated the use of the Microsoft Kinect as a cost-effective and marker-less alternative to more complex and expensive gold-standard motion analysis systems. Ten healthy subjects performed the SEBT while their lower limb kinematics were measured concurrently using a traditional motion capture system and a single Kinect v2 sensor. Analyses revealed errors in lower limb kinematics of less than 5°, except for the knee frontal-plane angle (5.7°) in the posterior-lateral direction. Ensemble curve analyses supported these findings, showing minimal between-system differences in all directions. Additionally, we found that the Kinect displayed excellent agreement (ICC3,k = 0.99) and consistency (ICC2,k = 0.99) when assessing reach distances in all directions. These results indicate that this low-cost and easy to implement technology may provide to clinicians a simple tool to simultaneously assess reach distances while developing a clearer understanding of the lower extremity movement patterns associated with SEBT performance in healthy and injured populations.

1. Introduction Measures of dynamic and static balance are used as tools to assess lower extremity injury risk, as well as clinical improvements over the course of rehabilitative care, in physically active populations [1,2]. It has been hypothesized that dynamic balance tasks, such as the Star Excursion Balance Test (SEBT), are clinically meaningful since they impose multiple internal and task-related neuromuscular demands, while assessing the dynamics of multi-planar movements around the base of support, which are the hallmarks of common activities of daily living and sport-specific tasks [3,4]. Maintaining functional joint stability during dynamic movement is key to successful performance of dynamic tasks, and mitigates the risk of joint injury. Consequently, movement impairments, or alterations associated with injury risk or post-injury neuromuscular dysfunction, are commonly evaluated using dynamic postural control assessments like the SEBT [3]. The SEBT is widely used to assess injury risk in physically active

⁎

individuals [1]. It is also an effective tool for monitoring the progress of lower extremity musculoskeletal injury rehabilitation in this population [1]. Poor performance on the SEBT has been linked to neuromuscular dysfunction [5–7], patient-reported functional capacity [8], return-toplay status [9], and risk for lateral ankle sprain [10]. While these findings clearly highlight the clinical utility of the SEBT, they provide no information concerning the lower extremity strategies that produce the directional data derived from the test. Recently, researchers have proposed the potential importance of stance leg kinematics in conjunction with overall reach distance as a method to enhance understanding of the mechanisms resulting in the main reach outcomes obtained [11–13]. The combination of a common clinical assessment, the SEBT, with a more comprehensive laboratory-based assessment of lower extremity movement strategies, three-dimensional motion analysis (3DMA), allows investigation of the task outcome while providing information regarding the potential kinematic source of the of poor reach performance among pathological population. In injured

Corresponding author at: Department of Kinesiology and Sport Sciences University of Miami 1507 Levante Ave. Max Orovitz Building 114 Coral Gables, FL 33146, USA. E-mail address: [email protected] (J. Signorile).

http://dx.doi.org/10.1016/j.gaitpost.2017.09.010 Received 10 April 2017; Received in revised form 31 August 2017; Accepted 9 September 2017 0966-6362/ © 2017 Published by Elsevier B.V.

Gait & Posture 58 (2017) 421–427

M. Eltoukhy et al.

populations this may allow clinicians to more objectively assess dynamic balance performance with the goal of developing targeted rehabilitative intervention to improve patient-reported and objective lower extremity clinical outcomes. Several research groups have assessed the kinematics of the stance and reach legs using 3DMA during the SEBT in a number of clinical populations including individuals with lateral ankle sprain [11], chronic ankle instability [12], and a history of ACL reconstruction [14]. While the effects of these conditions on balance vary, it is clear that: (1) lower extremity injury can negatively affect SEBT performance; (2) the kinematic strategy utilized to complete the SEBT may be different after injury; and, (3) performance and lower extremity kinematics may be related to the type of injury [11,12,14]. This knowledge is clinically meaningful since it highlights potential controlling factors during injury risk evaluation, thereby providing clinicians with potential targets for intervention. Unfortunately, laboratory-based approaches that utilize cameras and reflective markers or bilateral sets of sophisticated wearable sensors are costly, time-consuming, and technically demanding. This undoubtedly reduces the likelihood of their use for these important clinical integrations. Given the extensive use of the SEBT and the importance of evaluating the cause of specific directionalized performance decrements, the addition of a low-cost and user-friendly technology to enable 3D kinematic analysis, with no additional time or effort by the clinician, would be a natural progression. Recently, the Microsoft Kinect for Xbox One [15], a commercially available video game accessory, was investigated as a cost-effective movement analysis tool for use within clinical settings for various applications such as gait analysis [16], jump landing [17], and rehabilitation [18]. Interfacing this objective, efficient, and cost-effective technology with an established clinical test, such as the SEBT, would enable clinicians to supplement simple reach distance outcomes with a more comprehensive understanding of the lower extremity movement strategies employed during this dynamic reaching task and represent a major advance in the clinical application of the SEBT and other simple clinical evaluations. Therefore, the purpose of this study was to assess the validity of the Microsoft Kinect against a gold-standard 3DMA system in assessing SEBT performance and 3D lower extremity kinematics among young, healthy individuals. We hypothesized that the Kinect would (1) display good to excellent consistency and agreement in the assessment of lower extremity stance leg kinematics when compared to the 3DMA system; and, (2) display excellent consistency and agreement in assessing SEBT reach distances when compared to manual measurements.

Fig. 1. The SEBT experimental setup including, the SEBT grid with the measuring tape, one Kinect sensor, and the 3DMA infrared cameras. (Reflective markers are attached to the subject’s lower extremity).

anterior (A), posteromedial (PM), and posterolateral (PL). Reach distances were measured from the convergence point of the tapes to the maximum reach distances in each direction [19]. The Kinect was located 2.5 m from the subject at a height of 0.75 m from the ground. This methodology is consistent with previous publications and allowed for full visualization of the participant throughout the SEBT in order to ensure consistent collection of all lower extremity segments [20]. All participants wore spandex shorts to facilitate marker placement and to reduce potential motion artifacts. The session began with the measurement of subjects’ heights, leg lengths, knee widths, and ankle widths. Reflective markers were placed on the lower extremity according to the Vicon Plug-in-Gait lower body model [21]. The toe landmark’s marker used to define the reaching task and determine the reach distance in each direction is defined in the 3DMA system as the second metatarsal head, on the mid-foot side of the equinus break between fore-foot and mid-foot; while it was manually identified in the Kinect system by visual inspection of the depth point cloud from the ankle joint using the subject’s foot length determined based on the method described in Winter et al. [22]. Prior to testing, subjects were familiarized with the laboratory setup, shown correct SEBT procedures, and provided four practice trials in each reach direction to reduce any learning effect and clarify any concerns about the testing procedure [23,24]. Following the familiarization period, spatial and temporal calibration and synchronization of the 3DMA and the Kinect occurred with the subject in a T-pose. As described in [25], subjects performed the SEBT while standing barefoot on their non-dominant leg, which was defined as the opposite self-reported kicking leg, while holding their hands on the hips. Subjects were instructed to maintain single-leg stance while touching with the contralateral leg as light as possible the furthest point along the specified direction with the most distal part of the reaching foot, and to then return to the original starting double-support position. Subjects performed this reaching procedure in all three directions at their confortable pace. Participants completed the anterior reach direction first, followed by the posteromedial and posterolateral directions, respectively. A total of three successful trials were recorded for each subject, a trial was repeated if the subject did not keep their hands on their hips, lifted the heel of the stance foot, or lost balance at any point during the test.

2. Methods This was a single session observational laboratory study in which participants completed the SEBT while being concurrently monitored by a traditional camera-based 3DMA system and a single Microsoft Kinect sensor (Kinect v2). 2.1. Study participants Ten healthy subjects (5 M, 5F; 26.8 ± 5.7 yr, 1.74 ± 0.08 m, 73.5 ± 10.8 kg) participated in this study. They reported no cardiovascular, neurological or musculoskeletal impairments. The study was approved by the University’s Human Subjects Internal Review Board, and each participant provided written informed consent prior to participation. 2.2. Experimental setup and procedures The experimental setup, shown in Fig. 1, included eight infrared cameras (SMART-DX 7000, BTS Bioengineering, Milano, Italy) and one Kinect v2 (Microsoft Corp., Redmond, WA). Tape measures were affixed to the laboratory floor along the corresponding SEBT directions,

2.3. Data analysis Data were collected concurrently from the 3DMA and Kinect systems at sampling rates of 100 Hz and 30 Hz, respectively. The global 422

Gait & Posture 58 (2017) 421–427

M. Eltoukhy et al.

2.4. Statistical analysis

coordinate systems of the Kinect and the 3DMA systems were spatially aligned according to the method proposed by Clark et al. [26], with the Y-axis aligned with the vertical direction, X-axis aligned with the mediolateral direction, and the Z-axis aligned with the anteroposterior direction. This was done using a box of known dimensions, with one face positioned with the laboratory X and Y axes. Then the depth values of the position to either edge of the box using the factory calibrated 640 × 480 resolution depth sensor data were used to determine if the Kinect is properly positioned. If necessary, the Kinect’s orientation was then adjusted manually and using the built-in servomotor to further improve the alignment of the reference frame. 3DMA data were extracted using Vicon Nexus software (VICON Motion Systems, Inc., Oxford, UK), while Kinect data were extracted by analyzing the Kinect’s depth information using the dynamic link library (DLL) with.NET framework and a customized MATLAB code (MathWorks, Massachusetts, USA) [27]. Kinect-based joint angles were calculated as the angle between two vectors using the global coordinate system [28]. At each point of time, the 3D positions of the joint centers (landmarks) in the space were used to determine the sagittal plane angles; for instance, hip was defined previously as the thigh relative to trunk and knee angle was defined as the shank relative to thigh. Then, the frontal plane angles were determined as per the frontal plane projection angle descried in [29,30]. Data collected from both systems were filtered using a second order zero phase lag low-pass Butterworth filter (cut-off frequency of 6 Hz) to minimize any fluctuations in the markers’ trajectories. Consistent with previously reported methodology utilizing the SEBT [11,12], kinematic outcome measures were assessed during the course of reaching in each direction. Sagittal, frontal, and transverse plane hip joint angles; sagittal, and frontal plane knee joint angles; and sagittal plane ankle joint angles were measured throughout the reaching task. The task was defined as the initiation of movement through the point of maximal reach distance. Additionally, the magnitudes of the lower limb joint angles at the point of maximum reach in each direction were recorded. Lastly, the toe markers’ trajectories measured by the Kinect along the anteroposterior (dAP) and mediolateral (dML) directions, were utilized to calculate the reach distance in the corresponding SEBT directions as described in Table 1. All calculations were performed using the original data sets without any up- or down-sampling to prevent any introduction of noise or loss of resolution of the data. The 3DMA and Kinect measured joints angles in the anterior, posteromedial, and posterolateral directions were graphed across the reach task with associated 90% confidence intervals (CI90). The time to 100% of the reaching task was normalized for graphing purposes only, due to the difference in the sampling rates between both systems. This was performed using a custom LabVIEW code by resampling the 3DMA measured joint angles data set to obtain an identical number of data points for any given reach task while maintaining the signal integrity in the time and frequency domains [31]. The reaching task in each direction was defined in both systems as the time between the two consecutive minima of the toe marker’s trajectory in the vertical direction.

Means and standard deviations were calculated for maximal reach distances in all directions, as well as lower extremity joint angles at the time of maximal reach distance in all directions during the SEBT for the Kinect and 3DMA system. Maximal reach distances and the lower extremity joint angles at the point of each maximal reach distance obtained from both systems were compared using paired samples t-tests to assess between-system differences. The average relative and absolute mean differences between systems were calculated for each outcome variable. Absolute mean difference reflects the absolute value of the difference between systems, which reflects error magnitude (Eq. (1)); while relative mean difference takes into account the directionality of the error between systems (Eq. (2)).

Diff. abs . = abs [3DMA − Kinect ]

(1)

Diff .rel. = 3DMA − Kinect

(2)

In the calculation of between-system errors, the Kinect value was subtracted from the 3DMA measure, which means that a positive relative mean difference would indicate that the Kinect measure was consistently less than the 3DMA measure. Consistency (ICC2,k) and absolute agreement (ICC3,k) between systems were assessed for each outcome variable using separate interclass correlation coefficients. Absolute agreement considers the within-subject agreement between systems without consideration of systematic error; while relative consistency takes this factor into account when estimating between-system consistency [32]. ICCs were interpreted as poor ( < 0.4), fair (0.4 − < 0.6), good (0.6 − < 0.75), and excellent (≥0.75) [33]. All statistical analyses were completed using the SPSS version 20.0 (IBM, Chicago Illinois). Bland-Altman plots were also generated using Microsoft Excel (version 2012, Microsoft Corp. Redmond, WA). Alpha levels were set a-priori as P ≤ 0.05. In addition to kinematic assessments at the time of maximal reach distance, we generated ensemble curves and associated 90% confidence intervals (CI90) for the hip, knee, and ankle joint kinematics to allow between-system comparisons across the entire reaching task. Significant differences between systems were established when the CI90 for each system did not overlap consecutively for at least 3% of the stance phase [34,35]. Ensemble curve analyses were completed using Microsoft Excel. 3. Results In the anterior reach direction, significant between-system differences were present for frontal plane hip (p = 0.02) and sagittal plane knee (p = 0.004) joint angles at the maximal reach distance (Table 2). When the entire task was considered, significant between system differences were present for hip transverse plane motion from toe-off to 15% completion of the reaching task (Fig. 2). Despite the between system differences, consistency (ICC2,k = 0.73–0.99) ranged from good to excellent; while absolute agreement was excellent (ICC3,k = 0.76–0.99) across all kinematic variables.

Table 1 The description of the reach distance calculation using the toe trajectories obtained by the Kinect. SEBT reach direction

Extracted reach distance calculation

Definition

A

Max ( dAPi : i = 1,…,nA)

The maximum absolute toe trajectory dAPi for points 1 to nA, where nA is the total number of points recorded in the positive anteroposterior (AP) direction during the trial. The maximum absolute toe trajectory DPMi for points 1 to nPM, where nPM is the total number of points recorded in the posteromedial (PM) direction during the trial. DPMi is calculated using the toe trajectories recorded in the anteroposterior (AP) and mediolateral (ML) directions. The maximum absolute toe trajectory DPLi for points 1 to nPL, where nPL is the total number of points recorded in the posterolateral (PL) direction during the trial. DPLi is calculated using the toe trajectories recorded in the anteroposterior (AP) and mediolateral (ML) directions.

PM

Max (DPMi: i = 1,…,nPM) where, DPMi =

PL

dAPi 2 + dMLi 2

Max (DPLi: i = 1,…,nPL) where, DPLi = dAPi 2 + dMLi 2

423

Gait & Posture 58 (2017) 421–427

M. Eltoukhy et al.

Table 2 Comparison between lower limb kinematics obtained by the Kinect and 3DMA. SEBT Direction

Joint

Task

3DMA Mean ± SD

Kinect Mean ± SD

p-value

Relative Mean Difference

Absolute Mean Difference

Agreement ICC (95%CI)

Consistency ICC (95%CI)

Anterior (A)

Hip

Reach Distance (cm) Flexion/ Extension Abduction/ Adduction Internal/ External Rotation Flexion/ Extension Abduction/ Adduction Dorsi-/ Plantar-flexion

64.99 ± 5.78

65.30 ± 6.08

0.33

−0.31 ± 0.89

0.80 ± 0.41

18.97 ± 10.66

17.78 ± 9.61

0.16

1.20 ± 2.30

2.31 ± 0.95

8.00 ± 4.68

5.73 ± 5.57

0.02*

2.26 ± 2.36

2.92 ± 1.32

9.75 ± 4.34

9.87 ± 2.06

0.91

−0.12 ± 3.12

2.86 ± 0.76

0.99† (0.98–1.00) 0.99† (0.93–1.00) 0.90† (0.32–0.98) 0.76† (−0.21–0.95)

0.99† (0.98–1.00) 0.99† (0.94–1.00) 0.94† (0.75–0.99) 0.73† (−0.19–0.94)

62.19 ± 8.92*

64.45 ± 8.66

0.004*

−2.26 ± 1.66

2.46 ± 1.29

6.05 ± 5.93

4.79 ± 3.62

0.20

1.26 ± 2.70

2.62 ± 1.17

38.51 ± 5.81

36.57 ± 6.21

0.06

1.95 ± 2.63

2.87 ± 1.37

0.98† (0.48–0.99) 0.91† (0.63–0.98) 0.93† (0.61–0.99)

0.99† (0.96–0.99 0.92† (0.64–0.98) 0.95† (0.78–0.99)

71.72 ± 10.22

70.42 ± 10.96

0.03*

1.29 ± 1.46

1.80 ± 0.59

73.23 ± 17.57

73.84 ± 20.64

0.69

−0.62 ± 4.41

4.09 ± 1.04

0.54 ± 6.09

3.41 ± 6.64

0.002*

−2.86 ± 1.92

3.19 ± 1.22

9.27 ± 7.98

7.13 ± 5.88

0.03*

2.14 ± 2.50

2.88 ± 1.43

0.99† (0.93–1.00) 0.99† (0.95–1.00) 0.93† (0.03–1.00) 0.95† (0.63–0.99)

0.99† (0.98–1.00) 0.99† (0.94–0.99) 0.98† (0.90–1.00) 0.97† (0.86–0.99)

74.37 ± 13.19

75.96 ± 15.37

0.19

−1.59 ± 3.31

3.29 ± 1.27

9.89 ± 4.33

10.43 ± 3.47

0.49

−0.55 ± 2.25

2.04 ± 0.83

33.59 ± 4.49

32.57 ± 3.79

0.28

1.02 ± 2.64

2.45 ± 1.18

96.03 ± 10.52

94.72 ± 10.97

0.05*

1.31 ± 1.73

2.07 ± 0.38

0.99† (0.94–1.00) 0.92† (0.64–0.98) 0.87† (0.49–0.97) 0.99† (0.94–1.00)

0.99† (0.94–1.00) 0.91† (0.61–0.98) 0.88† (0.46–0.97) 0.99† (0.97–1.00)

50.78 ± 17.41

55.73 ± 18.70

−4.95 ± 1.83

4.95 ± 1.83

3.70 ± 3.05

5.66 ± 3.79

0.05*

−1.96 ± 2.56

2.97 ± 0.99

7.63 ± 4.43

3.67 ± 3.63

0.09

3.96 ± 6.21

5.50 ± 4.71

0.98† (0.07–1.00) 0.78† (0.07–0.95) 0.00 (0.00–0.64)

0.99† (0.99–1.00) 0.84† (0.29–0.96) 0.00 (0.00–0.68)

48.72 ± 17.85

51.50 ± 18.18

< 0.001*

−2.78 ± 1.14

2.78 ± 1.14

24.52 ± 4.13

18.78 ± 3.62

< 0.001*

5.74 ± 1.79

5.74 ± 1.79

27.51 ± 6.75

35.30 ± 7.59

2.20 ± 2.51

2.73 ± 1.85

0.99† (0.37–1.00) 0.60† (0–0.92) 0.95† (0.62–0.99)

0.99† (0.99–1.00) 0.94† (0.75–0.99) 0.97† (0.86–0.99)

Knee

Ankle Posteromedial (PM)

Hip

Knee

Ankle

Posterolateral (PL)

Hip

Knee

Ankle

*

Reach Distance (cm) Flexion/ Extension Abduction/ Adduction Internal/ External Rotation Flexion/ Extension Abduction/ Adduction Dorsi-/ Plantar-flexion Reach Distance (cm) Flexion/ Extension Abduction/ Adduction Internal/ External Rotation Flexion/ Extension Abduction/ Adduction Dorsi-/ Plantar-flexion

< 0.001*

0.03*

Indicates a significant between system difference (p ≤ 0.05), † indicates a significant correlation (p ≤ 0.05).

absolute agreement (ICC3,k = 0.87–0.99) was excellent across all kinematic variables (Table 2). Significant between-system differences were present for reach distance in the PM (p = 0.03) and PL (p = 0.05) directions (Table 2). In both cases, the Kinect sensor systematically underestimated the reach distance as compared to the measurement generated from 3DMA. Despite the significant between-system differences, excellent agreement and excellent consistency between systems were observed in the anterior (ICC2,k = 0.99, ICC3,k = 0.99), PM (ICC2,k = 0.99, ICC3,k = 0.99), and PL (ICC2,k = 0.99, ICC3,k = 0.99) reach directions (Table 2).

In the PM reach direction, significant between-system differences were present for frontal plane hip (p = 0.002) and transverse plane hip (p = 0.03) joint angles at maximal reach distance (Table 2). When the entire task was considered, significant between-system differences were present for hip transverse plane motion from toe-off to 15% completion of the reaching task and termination phase (90–100% of the task) of the reaching task; while knee joint frontal plane motion differed from toeoff to 15% completion of the reaching task alone (Fig. 2). Despite between-system differences, consistency (ICC2,k = 0.88–0.99) and absolute agreement were excellent (ICC3,k = 0.87–0.99) across all kinematic variables. In the PL reach direction, significant between-system differences were present for the frontal plane hip (p = 0.05), sagittal plane hip (p < 0.001), frontal plane knee (p < 0.001), and sagittal plane knee (p < 0.001) joint angles at maximal reach distance (Table 2). When the entire task was considered, significant between-system differences were present for hip transverse plane motion at the initiation of the reaching task (0–10%); while knee joint frontal plane motion differed during the middle portion of the reaching task (35–36%) (Fig. 2). Between system consistency (ICC2,k = 0.00–0.99) ranged from poor to excellent while

4. Discussion Automated and quantifiable assessment of the SEBT using Kinect technology provides more comprehensive information during this commonly utilized dynamic balance assessment. This includes a greater understanding of the kinematic strategy utilized to execute the SEBT reach task as compared to traditional clinical assessment which is limited to measurement of maximal reach distance. These data not only include reach performance, they also reveal the lower extremity 424

Gait & Posture 58 (2017) 421–427

M. Eltoukhy et al.

Fig. 2. Ensemble curves of the lower extremity joint during the SEBT test including, A) hip flexion/extension, B) hip abduction/adduction, C) hip internal/external rotation, D) knee flexion/extension, E) knee abduction/adduction, and F) ankle dorsi-/plantar-flexion. Dotted vertical line shows the point of maximum reach during the SEBT. Black solid curves: mean Kinect values, black dotted curves: mean 3DMA values, and vertical small bars are the associated 90% confidence intervals (CI90).

425

Gait & Posture 58 (2017) 421–427

M. Eltoukhy et al.

well as the crossing of limbs required to successfully complete the task may help to explain this significant measurement limitation. Conversely, the sagittal plane ankle joint kinematics assessed by the Kinect in this study were considerably more consistent and in better agreement with traditional 3DMA than reported in previous studies utilizing this sensor [17]. This was expected and can be attributed to these ankle joint angles being assessed for the stance foot, which was stationary during the test. In contrast, for more dynamic situations, such as treadmill gait, the Kinect commonly exhibits poor tracking accuracy for the ankle joint [16]. In this investigation, we found that the Kinect sensor displayed excellent agreement (ICC3,k = 0.99, Table 2) and consistency (ICC2,k = 0.99, Table 2) when assessing reach distance in the anterior, PM, and PL reach directions; however, significant mean differences between systems were present despite the levels of agreement and consistency. This difference can be explained by the difference in the estimation of the toe landmark between the two systems (Table 2). The toe landmark in the Kinect was obtained by the use of a machine learning algorithm and the subject’s anthropometric measurement; while the 3DMA toe landmark was obtained based on a physical reflective marker placed on the subject’s skin. The magnitude of the relative difference between systems was less than 2.0 cm, which is less than the standard error of measurement associated with manual scoring of the SEBT [40]. While it is clear that the SEBT was initially designed as a non-instrumented and clinician-driven alternative to more costly and technically demanding balance assessments [41], the rapid development of motion analysis technology has created the opportunity for evolution of this test. There are several limitations that should be considered when evaluating the results of this investigation. The quantification of the SEBT reach distance was limited to automated motion analysis-based approaches in this study. While this approach has been shown to be valid in previous investigations, direct comparison of the Kinect-based assessment to the more traditional manually assessed SEBT would improve the strength of evidence regarding this new technique. Additionally, agreement between multiple Kinect sensors and test-retest reliability must be established prior to utilization of this technique in subsequent research or clinical activities. Lastly, the study utilized a young and healthy sample to ensure that all subjects could successfully complete the SEBT reaching task and that between-trial variability would be minimized since the goal of this study was to validate the camera system. Future investigations should consider incorporating individuals with acute or chronic lower extremity injury as the SEBT is most commonly applied in these populations.

kinematic strategies utilized to complete the task successfully. Given the negligible cost of the Kinect camera and the relative ease with which the SEBT scoring algorithm can be applied, this approach also bridges the gap between expensive, laboratory-based kinematic assessments and more cost-effective, clinic-based performance assessments. Based on our findings, the Kinect provides excellent consistency and agreement with traditional 3DMA for SEBT reach performance; and, with the exception of transverse plane hip motion in the PL direction, good to excellent consistency and agreement in the assessment of stance leg kinematics. While the SEBT was initially described as a non-instrumented, clinical assessment of dynamic balance, the integration of low-cost technology, such as the Kinect, may drastically improve the level of information provided in a clinical setting. Lower extremity kinematics during the reach tasks have been proposed to be a primary determinate of SEBT performance in healthy and previously injured individuals [9,12,14]. In this investigation, with the exception of transverse plane hip motion measured in the A (ICC2,k = 0.73) and PL (ICC2,k = 0.00, ICC3,k = 0.00) directions, the agreement and consistency between systems were excellent for all kinematic variables. According to the lower limb joint angles obtained by both systems and summarized in Table 2, results were considered to be acceptable using the criteria suggested by McGinley et al. [36], who stated that, in most clinical situations, errors of ≤ 2° are considered acceptable, > 2° and ≤ 5° are reasonable, and errors that are > 5° may mislead clinical interpretation. In our study, the only error found that was greater than 5° was for the knee frontal-plane angle (average error = 5.7°) in the PL direction. Interestingly, this finding was consistent when considering the relative and absolute difference between systems. Inherently, the absolute difference between systems tends to be larger based on the fact that it is a measure of absolute error magnitude and does not consider whether the Kinect sensor is over or under estimating a given kinematic variable. It is important to note that based on the relative mean differences presented in Table 2, the Kinect sensor did consistently under- and overestimate kinematic variables in all three reach directions. However, a pattern based on plane of motion or reach direction did not emerge with the exception of consistent but small magnitude overestimation of sagittal plane ankle and knee angles. Further supporting the feasibility of the Kinect as a clinical tool, ensemble curve analyses revealed minimal between-system differences, and in most cases these differences occurred near the initiation of the task and were of a similarly small magnitude (< 5 °, Fig. 2). While statistically significant kinematic differences between systems are important to consider, given the relatively small magnitude of these differences and the consistently excellent between system consistency and agreement, these statistical differences may be limited in the applicability of Kinect sensor. On the other hand, similar Kinect validation studies reported higher between-system error; for instance, Guess et al. [37] reported sagittal plane knee and hip root mean square error (RMSE) of 11° and 12° respectively, and hip abduction/adduction RMSE of 7°. Kharazi et al. [38] reported RMSE of the hip, knee, and ankle sagittal plane joint angles of 5.9, 6.3, and 23.3° respectively. Furthermore, Xu et al. [39], reported between-system differences in the knee and hip sagittal plane angles of 29.0° and 11.9° respectively. Regarding our findings for transverse plane hip kinematics in the PL direction, the lack of consistency and agreement may be attributed to subjects’ trunk posture and pelvic tilt when performing this task. During the PL reach, the subject was required to reach behind the stance leg while forward flexing and laterally rotating the trunk to successfully complete the task. This limitation of the Kinect sensor is especially apparent in the estimation of transverse plane hip kinematics during the PL reach task (Table 2). Based on pilot data collected by our study team during a number of dynamic tasks, accurately measuring hip joint kinematics during these patterns of movement may be a consistent issue associated with the depth mapping approach inherent with the Kinect sensor. Inability to correctly estimate the hip joint center due to obstruction of the depth-mapping camera with the participant’s trunk as

5. Conclusions The SEBT is a clinically-relevant test of dynamic balance that has been broadly applied in research and clinical settings for healthy individuals and those with a history of lower extremity injury. The results of the current study indicate that the Kinect sensor can be used to assess stance leg kinematics in a valid manner, with the exclusion of transverse plane hip motion during the PL reach distance. While more resource intensive than the original SEBT; integration of this low-cost and easy to implement technology may provide clinicians an easy way to complete SEBT screening and other movement-based screening on a large scale, while developing a clearer understanding of the lower extremity movement patterns that are associated with poor test performance in healthy and injured populations.

Conflicts of interest statement The authors confirm that there are no known conflicts of interest associated with this publication and there has been no significant financial support for this work that could have influenced the outcome. 426

Gait & Posture 58 (2017) 421–427

M. Eltoukhy et al.

[20] M. Eltoukhy, C. Kuenze, J. Oh, J. Signorile, Validation of static and dynamic balance assessment using Microsoft Kinect for young and elderly populations, IEEE journal of biomedical and health informatics (2017) Epub ahead of print. [21] Vicon Motion Systems Ltd. Available online: http://vicon.com/display/Nexus25/ Lower+body+modeling+with+Plug-in+Gait (Accessed June 13, 2015). [22] D.A. Winter, Biomechanics and Motor Control of Human Movement, 3rd ed., John Wiley & Sons, 2005. [23] J. Hertel, S.J. Miller, C.R. Denegar, Intratester and intertester reliability during the star excursion balance tests, J. Sport Rehabil. 9 (2000) 104–116. [24] R.H. Robinson, P.A. Gribble, Support for a reduction in the number of trials needed for the star excursion balance test, Arch. Phys. Med. Rehabil. 89 (2008) 364–370. [25] R. Robinson, P. Gribble, Kinematic predictors of performance on the star excursion balance test, J. Sport Rehabil. 17 (2008) 347–357. [26] R.A. Clark, Y.H. Pua, K. Fortin, C. Ritchie, K.E. Webster, L. Denehy, A.L. Bryant, Validity of the Microsoft Kinect for assessment of postural control, Gait Posture 36 (2012) 372–377. [27] D. Pagliari, L. Pinto, Calibration of Kinect for Xbox One and comparison between the two generations of Microsoft sensors, Sensors 15 (2015) 27569–27589. [28] A. Fern'ndez-Baena, A. Susín, X. Lligadas, Biomechanical validation of upper-body and lower-body joint movements of Kinect motion capture data for rehabilitation treatments, 4th International Conference on Intelligent Networking and Collaborative Systems (INCoS) (2012) 656–661. [29] R. Mizner, T. Chmielewski, J. Toepke, et al., Comparison of 2-dimensional measurement techniques for predicting knee angle and moment during a drop vertical jump, Clin. J. Sport Med. 22 (2012) 221–227. [30] A. Munro, L. Herrington, M. Carolan, Reliability of 2-dimensional video assessment of frontal-plane dynamic knee valgus during common athletic screening tasks, J Sport Rehab. 21 (2012) 7–11. [31] C. Thiebaut, S. Roques Time-scale, time-frequency analyses of irregularly sampled astronomical time series, EURASIP J. Appl. Signal Process. 15 (2005) 2486–2499. [32] C. Bunce Correlation, agreement: and Bland-Altman analysis: statistical analysis of method comparison studies, Am. J. Ophthal. 148 (2009) 4–6. [33] V. Cicchetti, Guidelines criteria, and rules of thumb for evaluating normed and standardized assessment instruments in psychology, Psychol. Assess. 6 (1994) 284–290. [34] P. McKeon, G. Paolini, C. Ingersoll, D. Kerrigan, E. Saliba, B. Bennett, J. Hertel, Effects of balance training on gait parameters in patients with chronic ankle instability: a randomized controlled trial, Clin. Rehabil. 23 (2009) 609–621. [35] C. Kuenze, J. Hertel, A. Weltman, D. Diduch, S. Saliba, J. Hart, Jogging biomechanics after exercise in individuals with ACL-reconstructed knees, Med. Sci. Sports Exerc. 46 (2014) 1067–1076. [36] J.L. McGinley, R. Baker, R. Wolfe, M.E. Morris, The reliability of three-dimensional kinematic gait measurements: a systematic review, Gait Posture 29 (2009) 360–369. [37] T. Guess, S. Razu, A. Jahandar, M. Skubic, Z. Huo, Comparison of 3D joint angles measured with the kinect 2.0 skeletal tracker versus a marker-based motion capture system, J. Appl. Biomech. 33 (2017) 176–181. [38] M. Kharazi, A. Memari, A. Shahrokhi, et al., Validity of microsoft kinectTM for measuring gait parameters, 22nd Iranian Conference on Biomedical Engineering (ICBME), Tehran, 2015, pp. 375–379. [39] X. Xu, R. McGorry, L. Chou, J. Lin, C. Chang, Accuracy of the Microsoft Kinect™ for measuring gait parameters during treadmill walking, Gait Posture 42 (2015) 145–151. [40] A.G. Munro, L.C. Herrington, Between-session reliability of the star excursion balance test, Phys. Ther. Sport 11 (2010) 128–132. [41] S.J. Kinzey, C.W. Armstrong, The reliability of the star-excursion test in assessing dynamic balance, J. Orthop. Sports Phys. Ther. 27 (1998) 356–360.

References [1] P.A. Gribble, J. Hertel, P. Plisky, Using the star excursion balance test to assess dynamic postural-control deficits and outcomes in lower extremity injury: a literature and systematic review, J. Athl. Train. 47 (2012) 339–357. [2] J.M. Dallinga, A. Benjaminse, K.A. Lemmink, Which screening tools can predict injury to the lower extremities in team sports? Sportsmedicine 42 (2012) 791–815. [3] P.A. Gribble, J. Hertel, P. Plisky, Using the star excursion balance test to assess dynamic postural-control deficits and outcomes in lower extremity injury: a literature and systematic review, J. Athl. Train. 47 (2012) 339–357. [4] P.J. Plisky, M.J. Rauh, T.W. Kaminski, F.B. Underwood, Star excursion balance test as a predictor of lower extremity injury in high school basketball players, J. Ortho. Sports Phys. Ther. 36 (2006) 911–919. [5] L. Herrington, J. Hatcher, A. Hatcher, M. McNicholas, A comparison of star excursion balance test reach distances between ACL deficient patients and asymptomatic controls, The Knee 16 (2009) 149–152. [6] L.C. Olmsted, C.R. Carcia, J. Hertel, S.J. Shultz, Efficacy of the star excursion balance tests in detecting reach deficits in subjects with chronic ankle instability, J. Athl. Train. 37 (2002) 501–506. [7] M. Terada, M.S. Harkey, A.M. Wells, B.G. Pietrosimone, P.A. Gribble, The influence of ankle dorsiflexion and self-reported patient outcomes on dynamic postural control in participants with chronic ankle instability, Gait Posture 40 (2014) 193–197. [8] M. Terada, M.S. Harkey, A.M. Wells, B.G. Pietrosimone, P.A. Gribble, The influence of ankle dorsiflexion and self-reported patient outcomes on dynamic postural control in participants with chronic ankle instability, Gait Posture 40 (2014) 193–197. [9] S. Clagg, M.V. Paterno, T.E. Hewett, L.C. Schmitt, Performance on the modified star excursion balance test at the time of return to sport following anterior cruciate ligament reconstruction, J. Orthop. Sports Phy. Ther. 45 (2015) 444–452. [10] P.A. Gribble, M. Terada, M.Q. Beard, K.B. Kosik, A.S. Lepley, R.S. McCann, A.C. Thomas, Prediction of lateral ankle sprains in football players based on clinical tests and body mass index, Am. J. Sports Med. 44 (2016) 460–467. [11] C. Doherty, C.M. Bleakley, J. Hertel, B. Caulfield, J. Ryan, E. Delahunt, Laboratory measures of postural control during the star excursion balance test after acute firsttime lateral ankle sprain, J. Athl. Train. 50 (2015) 651–664. [12] E. Delahunt, M. Chawke, J. Kelleher, K. Murphy, A. Prendiville, L. Sweeny, M. Patterson, Lower limb kinematics and dynamic postural stability in anterior cruciate ligament-reconstructed female athletes, J. Athl. Train. 48 (2013) 172–185. [13] P.A. Gribble, J. Hertel, C.R. Denegar, W.E. Buckley, The effects of fatigue and chronic ankle instability on dynamic postural control, J. Athl. Train. 39 (2004) 321–329. [14] R. Pionnier, N. Découfour, F. Barbier, C. Popineau, E. Simoneau-Buessinger, A new approach of the star excursion balance test to assess dynamic postural control in people complaining from chronic ankle instability, Gait Posture 45 (2016) 97–102. [15] D. Pagliari, L. Pinto, Calibration of kinect for Xbox One and comparison between the two generations of Microsoft sensors, Sensors 15 (2015) 27569–27589. [16] M. Eltoukhy, J. Oh, C. Kuenze, J.F. Signorile, Improved kinect-based spatiotemporal and kinematic treadmill gait assessment, Gait Posture 51 (2017) 77–83. [17] M. Eltoukhy, A. Kelly, C. Kim, J. Hyung-Pil, R. Campbell, C. Kuenze, Validation of the Microsoft Kinect® camera system for measurement of lower extremity jump landing and squatting kinematics, Sports Biomech. 15 (2016) 89–102. [18] N. Kitsunezaki, E. Adachi, T. Masuda, J. Mizusawa, KINECT applications for the physical rehabilitation, Medical Measurements and Applications Proceedings (MeMeA), 2013 IEEE International Symposium (2013) 294–299. [19] J. Hertel, R.A. Braham, S.A. Hale, L.C. Olmsted-Kramer, Simplifying the star excursion balance test: analyses of subjects with and without chronic ankle instability, J. Orthop. Sports Phys. Ther. 36 (2006) 131–137.

427