Methods. We used Motognosis Labs in combination with Microsoft Kinect V2 and Software. Development Kit (SDK) Version 2.0. This poster is based on data ...
Evaluation of kinematic parameters of potential clinical use extracted from Microsoft Kinect V2 motor assessments Karen Otte1 , Bastian Kayser1 , Sebastian Mansow-Model1, Alexander U. Brandt3,5, Julius Verrel2, Friedemann Paul3,4,5, Tanja Schmitz-Hübsch3,4 1) Motognosis UG (haftungsbeschränkt), Berlin, Germany 2) Center for Lifespan Psychology, Max Planck Institute for Human Development, Berlin, Germany 3) NeuroCure Clinical Research Center and Clinical and Experimental Multiple Sclerosis Research Center, Charité - Universitätsmedizin Berlin, Berlin, Germany 4) Department of Neurology, Charité – Universitätsmedizin Berlin, Berlin, Germany 5) Experimental and Clinical Research Center, Max Delbrück Center for Molecular Medicine and Charité - Universitätsmedizin Berlin, Berlin, Germany
Background
Table 2: Measurement Repeatability for 50 HC, 41 MS and 10 PD Measurements
Repeat. HC Mean (SD) Mean (SD) Mean (SD) SEM ICC HC MS PD [%] (1,1)
Instrumental motor assessment has been proposed to increase objectivity but most current systems are time and cost intensive. After we established spatial accuracy of Kinect V2 anatomical landmarks, we now explore the accuracy and repeatability of derived kinematic parameters for possible clinical use.
Repeat. MS SEM ICC [%] (1,1)
Repeat. PD SEM ICC [%] (1,1)
POCO
Methods We used Motognosis Labs in combination with Microsoft Kinect V2 and Software Development Kit (SDK) Version 2.0. This poster is based on data from four different pilot studies. In total, 50 healthy controls (HC, 17 male / 33 female, age 42±12 years), 41 patients with Multiple Sclerosis (MS, 19 male / 22 female, age 50±6 years) and 10 patients with Parkinson‘s Disease (PD, 9 male / 1 female, age 66±13 years). All subjects were recorded with the same standardized measurement protocol including 5 different motor tasks (See Table 1). A subgroup of 19 HCs was simultaneously recorded with a gold standard reference system (Vicon, MX13+, Nexus 2.1), to evaluate accuracy and repeatability.
Figure 1: Motognosis Labs and Study Setup
DR 3D [°] MSV 3D [°/s] SAS Up Time [s] Down Time [s] SMSW Speed [m/s] Hip V [cm] SCSW Speed [m/s] Hip ML [cm] SLW Speed [m/s] Hip ML [cm]
1.82 (0.18) 3.04 (2.77) 1.97 (0.05) 29.74 0.34 (0.08) 0.68 (0.49) 0.45 (0.09) 27.29
0.78 6.66* 0.85* 21.84 0.86 13.1* 0.94* 22.94
0.86 0.91
1.55 (0.00) 2.06 (0.08) 2.08 (0.60) 1.71 (0.01) 2.10 (0.03) 2.27 (0.12)
4.95 7.07
0.80 11.85 0.67 9.68
0.90 35.89 0.88 13.10
0.53 0.80
1.99 (0.01) 1.37 (0.03) 1.41 (0.01) 4.72 0.91 (0.09) 2.51 (0.17) 1.28 (0.09) 31.06
0.87 10.96 0.63 122.4
0.84 3.64 0.78 14.54
0.97 0.89
1.28 (0.00) 0.98 (0.00) 0.98 (0.01) 5.26 1.44 (0.02) 3.20 (0.51) 5.51 (0.08) 22.88
0.85 5.76 0.36 26.11
0.94 7.43 0.98 207.3
0.93 0.08
0.28 (0.01) 0.25 (0.00) 0.26 (0.00) 12.47 1.87 (2.96) 2.37 (2.36) 3.93 (9.45) 181.5
0.87 12.03 0.01 132.0
0.93 10.79 0.12 257.7
0.89 0
Abbrev: ML – medio-lateral, V – vertical, DR – deflection range, MSV – mean angular sway velocity, LOA – Limits of Agreement, SEM – Standard Error of Measurement * based on data from [2] using Kinect V1
The five measurement tasks included standing up and sitting down (SAS), standing with open and closed eyes for 20s each (POCO) and walking in different conditions (comfortable speed (SCSW), maximum speed (SMSW) and line walk (SLW)). In total, 45 clinical parameters were calculated to describe the movement behavior of a subject. A small subset of parameters is shown here, including the mean walking speed in m/s for all walk conditions, as well as the left-right hip deviation (Hip ML) and up-down hip deviation (Hip V) in cm [1]. The stand up and sit down assessment is described by the time needed to stand up (Up Time) and sit down (Down Time) in seconds. The sway behavior of a subject is described by the angular 3D deflection range of the sway (DR 3D) and the angular 3D mean sway velocity (MSV 3D) at hip height [2].
As shown, gait speed is slowed in comfortable and maximum speed condition and time for postural transition is increased in both, MS and PD. This was not statistically tested due to differences in group size and participant characteristics of the presented population. In contrast, body sway in static balance test is increased specifically in MS patients. Whether increased hip ML is specific for PD or related to differences in step size needs further study. Except for hip ML movement in walk tests, all kinematic parameters can be accurately tested in both, patients and controls, regarding their high repeatability in immediate retest.
Figure 2: Group Comparison of HC (black) MS (orange) and PD (blue)
SMSW
SCSW
SLW
POCO
SAS
SAS
Results Regarding the comparison of Kinect and Vicon (Table 1), most kinematic parameters showed good to excellent absolute agreement (ICC(3,1)) and consistency (Pearson’s r). Repeatability (ICC(1,1)) was acceptable for most parameters and of similar magnitude for both systems. Relative SEM was acceptable (< 20%) for most of the investigated parameters.
Table 1: Accuracy and Repeatability of Kinect against Vicon (19 HC) Measurements Assessment / Mean (SD) Mean (SD) Feature Kinect Vicon POCO DR 3D [°] MSV 3D [°/s] SAS Up Time [s] Down Time [s] SMSW Speed [m/s] Hip V [cm] SCSW Speed [m/s] Hip ML [cm] SLW Speed [m/s] Hip ML [cm]
Accuracy Repeatability Repeatability Kinect vs Vicon Kinect Vicon LOA ICC Pear- SEM ICC SEM ICC (3,1) son’s r [%] (1,1) [%] (1,1)
1.60 (0.73) 0.30 (0.10)
1.50 (0.60) 0.5 0.30 (0.09) 0.05
0.92 0.96
0.95 19.09 0.97 10.78
0.66 23.84 0.65 13.40
0.56 0.66
1.31 (0.25) 1.53 (0.23)
1.38 (0.25) 0.09 1.59 (0.23) 0.14
0.95 0.92
0.98 0.95
0.72 0.55
8.59 9.10
0.77 0.62
2.10 (0.21) 0.81 (0.42)
2.08 (0.21) 0.02 0.88 (0.52) 0.30
0.99 0.90
1.00 3.72 0.93 21.43
0.86 3.65 0.83 17.76
0.87 0.91
9.89 9.91
1.29 (0.16) 1.40 (0.45)
1.28 (0.15) 0.02 1.19 (0.48) 0.54
1.00 0.75
1.00 5.38 0.82 25.03
0.81 5.29 0.40 28.75
0.80 0.49
0.24 (0.05) 2.04 (0.78)
0.24 (0.05) 0.00 1.87 (0.71) 0.40
1.00 0.94
1.00 5.78 0.97 17.74
0.92 5.83 0.78 20.25
0.92 0.71
Abbrev: ML – medio-lateral, V – vertical, DR – deflection range, MSV – mean angular sway velocity, LOA – Limits of Agreement, SEM – Standard Error of Measurement
Conclusion We found excellent agreement between Kinect and Vicon gold standard as well as retest reliability for a variety of kinematic parameters extracted from different motor tasks of clinical interest. This supports Kinect V2 as potentially suitable for clinical use and coincides with the findings in [3-4]. However, our results are based on groups that are not age/gender matched and contain only few patients with Parkinson’s disease. Further investigations are required to analyze the influence of symptom severity on measured movement performances. Acknowledgement: We would like to thank Alessandro Carpanese for measuring the patients with Parkinson’s disease, as well as Anuschka Grobelny and Theresa Krüger for measurement of the patients with multiple sclerosis and a subset of the healthy controls.
Reference: 1) Behrens et al, Journal of NeuroEngineering and Rehabilitation, 2014 DOI: 10.1186/1743-0003-11-89 2) Behrens et al, Multiple Sclerosis Journal. 2016 DOI: 10.1177/1352458515625807 3) Clark et al, Gait & Posture, 2015 DOI: http://dx.doi.org/10.1016/j.gaitpost.2015.03.005 4) Mentiplay et al, Journal of Biomechanics, 2015 DOI: http://dx.doi.org/10.1016/j.jbiomech.2015.05.021
Disclaimer: Motognosis is a start-up company from the Charité’s laboratory with commercial interest in the described technology. All other authors declare no potential conflict of interest.
