Proceedings of the 12th IASTED International Conference Biomedical Engineering (BioMed 2016) February 15-16, 2016 Innsbruck, Austria REPLACING REDUNDANT STABILOMETRY PARAMETERS WITH RATIO AND MAXIMUM DEVIATION PARAMETERS Gergely Nagymate, Rita M. Kiss Budapest University of Technology and Economics, Dept of Mechatronics, Optics and Mechanical Engineering Informatics Műegyetem rakpart 3, Budapest 1111, Hungary nagymate@mogi.bme.hu, rikiss@mail.bme.hu ABSTRACT During the stabilometry foot center of pressure (COP) was measured on force plates or force distribution plates. From the sampled data several widely used parameters (95% confidence ellipse axes, area and, standard s of COP, path length and COP ranges) and newly defined parameters 95% confidence ellipse axis ratio, anteroposterior-mediolateral range ratio) can be calculated. Our study aims to reduce the number of the COP parameters by excluding those that contain redundant information using correlation analysis. We also replaced multiple parameters with summarizing parameters which based on the replaced parameters give a new and independent approach of the same phenomenon. To describe the extremes of the COP movement maximum from mean COP in forward and backward directions were calculated highlighting the random faults of postural control. The new and old uncorrelated parameters went under a variance analysis for the repeated measures. The new maximum parameters showed similar behavior with previous ellipse parameters between trial types. The ratio parameters also showed significant differences between the three trials. KEY WORDS stabilometry, center of pressure, balance, postural stability 1. Introduction Stabilometry is a widely used method to analyze human postural stability and the biomechanical effects of neural or physiological disorders. Due to the simplicity and short time of stabilometry measurements it can be useful in clinical diagnostic [1]. Stabilometry is usually based on the analysis of the time variant center of pressure (COP) coordinates of the feet during bipedal or single limb stance in eyes open or closed conditions. COP is usually measured on force plates or force distribution plates sampled on a certain frequency. From the sampled data several time-distance or frequency based parameter can be calculated. Among time-distance parameters the most widely used are COP path length, COP ranges and standard in anteroposterior (AP) and mediolateral (ML) directions, average COP velocity, sometimes separated in the previous directions. A 95% confidence ellipse is often fitted on the scattered COP points, where the area, and the length of its axes are the considered parameters. Lawson in a systematic review [2] summarizes several studies working with COP analysis on knee osteoarthritis patients and points out that the wide variety of parameters and their different combinations result difficulties comparing the results of different studies. Several researchers found that among the above parameters the COP path length [3] or on fixed measurement interval the equivalent average COP velocity [4, 5] offers the best testretest reliability. Most of these parameters describe postural stability on a statistical basis considering it as a static process. Several studies [6, 7] pointed out the nonstationarities of the postural control process. Therefore above the averaging and statistical parameters the description of the extremes of the COP motion should be considered as well. AP and ML ranges are these kinds of parameters, but the largest s that have both AP and ML components remain hidden with just these parameters. The parameters often undergo on statistical analyzes, where independency of parameters is beneficial. Our study aims to reduce the number of the COP parameters by excluding those that contain redundant information. We also replaced multiple parameters with summarizing parameters which based on the replaced parameters give a new and independent approach of the same phenomenon. New parameters were also defined to evaluate the largest s and maximum velocities of the COP. 2. Methods 2.1 Participants Participants involved in the study were 32 young healthy individuals (28 males, 4 females: 22.34±0.97 years, 74.34±4.23 kg, 171.12±8.51 cm). All participants were informed in writing about the risks and benefits of the study; each gave signed informed consent and was given the opportunity to withdraw from the study at any time. The study was approved by the Hungarian National Science and Research Ethics Committee (114/2004). 2.2 Experimental Procedure Data on postural stability was collected using a Zebris FDM-S multifunctional Force Distribution Measuring DOI: 10.2316/P.2016.832-022 140
plate (320 mm x 470 mm measuring surface with 1504 pcs. load cells) (ZEBRIS GmbH, Isny, Germany). Three trials were performed barefoot on every participant, each during 60 seconds long bipedal stance with eyes open and eyes closed (Fig 1). Furthermore single limb stance on the dominant leg was performed immediately successively. The participants were asked to stand at least 2 minutes before the trial started and possibly load both leg equally during the bipedal trials. Fig 2. COP AP+ (d 1 ) and AP- maximum (d 2 ) and s (α, β) 2.4 Statistical Analysis Fig 1. Bipedal stance on force distribution plate 2.3 Data Processing Measurement data was recorded using the Zebris WinPDMS processing software v1.2 (zebris GmbH, Isny, Germany). This software calculates only few of the most widely used COP parameters, namely the 95% Confidence ellipse axes, area and, path length, COP standard in AP and ML directions, however it offers a raw measurement data export function. The exported foot force distribution data were used to calculate the instantaneous COP for each timeframe in order to calculate further parameters for the analysis of COP. Our data processing program was developed in LabVIEW 2013 (National Instruments Inc., Austin, Texas). In addition to the parameters of the official software, our software calculates other widely used COP parameters (AP range, ML range, average path velocity) and new parameters that were not commonly used before: maximum from average COP and its in +AP and -AP directions maximum path velocity AP-ML range-ratio 95% confidence ellipse axis ratio While the calculation of most parameters is trivial, the COP +AP and -AP maximum and is described on Fig 2. To estimate the usability of the newly defined COP parameters statistical analysis was applied. To define which set of parameters can be replaced by fewer parameters correlation analysis was used on the resulting parameters of summed bipedal stance trials. The behavior of the new parameters in respect to the different stance trials was analyzed by a multivariate ANOVA. The ANOVA requires meeting the following assumptions. Continuity of the dependent variables which in our case is guaranteed by their nature. The independent variable must have at least 2 groups. In our study three groups for the three kinds of trials are given. Outliers and extremes should be excluded from the analysis to avoid distortion and gain higher statistical power for the analysis. Distribution of the dependent variable in the related groups should be approximately normally distributed. The variances of the differences between all combinations of related groups must be equal, also known as sphericity. The correlation matrix was also utilized to ensure the independency of the variables by excluding variables that are strongly correlated and replacing some with the newly defined parameters which s independency was also verified. Extremes and most outliers were manually removed from the dataset based on box plots considerably decreasing our sample size from 32 to 25, but still improving the statistical power thus reliability of the analysis. Several parameters in one participant s single leg trial were extremes, because he failed the trial and had to put down his leg, therefore he was also excluded from the study. Normal distribution of the dependent variables was analyzed by Shapiro-Wilk test of normality, which is more appropriate for sample size below 50 unlike the 141
Kolmogorov-Smirnov test of normality. Not normally distributed variables can be transformed to normal distribution, but this is not necessarily required because ANOVA is quite robust for the violation of this assumption. Sphericity was tested with Mauchly s test of sphericity. For the variables that do not meet the required sphericity Greenhouse-Geisser correction was applied when evaluating significance level of the within-subject effects. Overall significant difference between the means of the three stance types was calculated for each variable. To see where the specific differences occurred pairwise comparisons were performed between groups for each variable. During the statistical analysis the level of significance is 0.05 (α =0.05) 3. Results 3.1 Correlation Analysis The results of the correlation analysis are shown in Table 1. The 95% confidence ellipse area has naturally strong correlation with the length of its minor (0.896) and major axis (0.86). The area also has a strong correlation with the standard s in AP (0.769) and ML (0.821) directions and trajectory ranges in the same directions (AP: 0.686, ML: 0.788). There are strong correlations between the standard s and ranges of the COP in both AP (0.887) and ML directions (0.917). The confidence ellipse axes also correlate with the standard s and ranges (Minor axis ML SD: 0.885, Minor axis ML range: 0.896, Major axis AP SD: 0.937, Major axis AP range: 0.827). The average COP velocity is in a linear connection with the path length (correlation = 1.0). There is no significant correlation between the newly introduced parameters (Axis ratio, AP- ML range ratio, Max. in ±AP directions and its ) and the previously used parameters (correlation < 0.8). However, Axis ratio and AP-ML ratio show some redundancy (correlation = 0.744). Based on the correlation analysis results we excluded the redundant variables, therefore we applied the variance analysis only on the following parameters: Axis ratio, 95% confidence ellipse area, Path length, Maximum velocity, AP-ML range ratio, +AP, -AP, +AP, -AP, Confidence ellipse. Table 1. Covariance matrix of COP parameters Minor Axis Minor Axis Major Axis Area Angle Axis Ratio Path length Average velocity Maximum velocity AP standard ML standard AP range ML range AP-ML ratio AP+ Maximum AP+ Maximum AP- Maximum AP- Maximum 1.609.896.009 -.594 -.013 -.013.111.498.885.480.896 -.546.651.178.518 -.227 Major Axis.609 1.860.000.189 -.055 -.055.141.937.614.827.579.023.664.028.761.138 Area.896.860 1.009 -.259 -.073 -.073.151.769.821.686.788 -.288.692.087.681 -.077 Angle.009.000.009 1 -.052.054.054.129 -.022.034 -.022.021 -.091.166.223 -.168.177 Axis Ratio -.594.189 -.259 -.052 1 -.088 -.088.036.263 -.471.186 -.482.744 -.098 -.192.079.308 Path length -.013 -.055 -.073.054 -.088 1 1.000 -.019 -.009 -.064 -.004.021 -.053.059.157 -.047.188 Average velocity -.013 -.055 -.073.054 -.088 1.000 1 -.019 -.009 -.064 -.004.021 -.053.059.157 -.047.188 Maximum velocity.111.141.151.129.036 -.019 -.019 1.111.126.209.170.082.325 -.078.044 -.107 AP standard.498.937.769 -.022.263 -.009 -.009.111 1.366.887.393.212.608 -.048.746.265 ML standard.885.614.821.034 -.471 -.064 -.064.126.366 1.347.917 -.620.595.236.457 -.279 AP range.480.827.686 -.022.186 -.004 -.004.209.887.347 1.382.311.705 -.051.792.267 ML range.896.579.788.021 -.482.021.021.170.393.917.382 1 -.672.645.275.466 -.300 AP-ML ratio -.546.023 -.288 -.091.744 -.053 -.053.082.212 -.620.311 -.672 1 -.081 -.289.113.400 AP+ Maximum.651.664.692.166 -.098.059.059.325.608.595.705.645 -.081 1.142.350.014 AP+ Maximum.178.028.087.223 -.192.157.157 -.078 -.048.236 -.051.275 -.289.142 1.109 -.125 AP- Maximum.518.761.681 -.168.079 -.047 -.047.044.746.457.792.466.113.350.109 1.053 AP- Maximum -.227.138 -.077.177.308.188.188 -.107.265 -.279.267 -.300.400.014 -.125.053 1 Comment: Bold: strong correlation, Italic: newly introduced parameters 142
3.2 Variance Analysis Since the ANOVA requires independent variables among the strongly correlated variables the confidence ellipse area was kept, while the ellipse axes, COP ranges and standard s were excluded. Additionally axis ratio and AP-ML range ratio was used which are uncorrelated to the other variables. Instead of the average COP velocity maximum velocity was used. Detailed results of the variance analysis are presented in Table 2 and Table 3. Table 2. Univariate test results (difference significance levels among the three trials for each parameter) Parameter Significance Observed power Axis ratio <0.001 0.975 95% confidence ellipse area <0.001 1.000 Path length <0.001 1.000 Maximum velocity <0.001 1.000 AP-ML range ratio 0.001 0.960 +AP <0.001 1.000 -AP <0.001 1.000 +AP 0.425 0.192 -AP 0.112 0.444 Confidence ellipse 0.162 0.331 Comment: AP: anteroposterior, ML: mediolateral Table 3. Pairwise comparison on COP parameters Measure Mean Difference Std. Error Sig. Axis ratio EO Bipedal EC Bipedal -0.170 0.222 1.000 EO Bipedal EO Single leg 0.606 0.171 0.004 EC Bipedal EO Single leg 0.775 0.162 <0.001 95% confidence ellipse area EO Bipedal EC Bipedal 7.352 4.996 0.458 EO Bipedal EO Single leg -150.926 10.968 <0.001 EC Bipedal EO Single leg -158.278 11.688 <0.001 Path length EO Bipedal EC Bipedal -122.919 27.311 <0.001 EO Bipedal EO Single leg -1117.425 126.050 <0.001 EC Bipedal EO Single leg -994.507 112.910 <0.001 Maximum velocity EO Bipedal EC Bipedal -5.667 15.307 1.000 EO Bipedal EO Single leg -236.304 23.327 <0.001 EC Bipedal EO Single leg -230.637 24.792 <0.001 AP-ML range ratio EO Bipedal EC Bipedal -0.376 0.148 0.051 EO Bipedal EO Single leg 0.162 0.111 0.467 EC Bipedal EO Single leg 0.537 0.137 0.002 +AP maximum EO Bipedal EC Bipedal -1.054 0.984 0.881 EO Bipedal EO Single leg -10.504 1.250 <0.001 EC Bipedal EO Single leg -9.451 1.192 <0.001 -AP maximum EO Bipedal EC Bipedal 0.289 1.115 1.000 EO Bipedal EO Single leg -9.699 1.282 <0.001 EC Bipedal EO Single leg -9.988 1.465 <0.001 +AP maximum -AP maximum 95% Confidence ellipse Among the used variables described above the Shapiro-Wilk test of normality gave significant difference from normal distribution to axis ratio (0.024), maximum velocity (0.001) and AP-ML range ratio (0.024) for eyes open bipedal stance, ellipse area (0.017), path length (0.024) and maximum velocity (0.003) in eyes closed bipedal stance, furthermore axis ratio (0.018), ellipse (0.01) and path length (0.009) in eyes open single leg stance. The variance analysis is robust enough therefore these variables were not transformed to normal distribution. EO Bipedal EC Bipedal 9.810 7.970 0.687 EO Bipedal EO Single leg 8.983 8.816 0.952 EC Bipedal EO Single leg -0.826 7.950 1.000 EO Bipedal EC Bipedal -16.502 8.719 0.208 EO Bipedal EO Single leg -17.009 10.439 0.345 EC Bipedal EO Single leg -0.506 7.814 1.000 EO Bipedal EC Bipedal 1.015 7.558 1.000 EO Bipedal EO Single leg 13.911 9.738 0.494 EC Bipedal EO Single leg 12.896 5.928 0.116 Comment: EO: eyes open, EC: eyes closed, AP: anteroposterior, ML: mediolateral, Bold: significant difference 143
Mauchly s sphericity test indicated the violation of sphericity among stance types for ellipse area (<0.001), path length (<0.001), maximum velocity (0.022) and ellipse (0.008). For these variables the Greenhouse- Geisser correction was applied when evaluating differences among groups. The within subjects multivariate test showed overall significant differences between the three stance types (p<0.001, Observed power = 1.000). The univariate tests show the significant differences for means between stance types (Table 2). Basically most variables show significant differences between the three stance types except for the three type variables. Pairwise comparisons reveal among which stance types occurred the significant difference. The significance levels for pairwise comparisons are displayed in Table 3. 4. Discussion Correlation analysis revealed significant correlations between many of the previously used variables in describing COP motion, which makes them less suitable for variance analysis and introduces unnecessary redundancy in the COP analysis based evaluation processes. The standard s and COP ranges show strong correlations in the respective directions, but this happens only when there are no large excursions in the COP trajectory compared to the size of the confidence ellipse. This is due to the fact that the standard (as well as confidence ellipse) describes the whole average of the measurement, while the ranges describe the extremes of the COP trajectory. Usage of the ellipse axes together with the ellipse area contains redundant information. This study showed that it would be a better approach to describe the area of the ellipse along with its kurtosis defined by the major-minor axis ratio. AP-ML range ratio correlates with the ellipse axis ratio, but as before the range ratio describes the extremes, while axis ratio describes an overall statistic, thus both parameters may be used. The ±AP maximum s and corresponding s tells the sizes of the largest excursions in the anterior and posterior directions. These reveal the largest COP excursions relative to the average COP, which might stay hidden using only the AP and ML ranges. The comparisons of three trial types (bipedal stance with open eyes, bipedal stance with closed eyes and single leg stance) for the parameters used so far showed the same results that can be found in the literature [8]. Namely the values of the parameters rise but not significantly between EO and EC bipedal stance, whereas they rise significantly in single leg stance. Path length changed significantly between the two bipedal stance trials as well (Table 3). AP maximum s and 95% CE axis ratio presented the same behaviour. AP-ML ratio surprisingly not differs significantly between EO bipedal and single limb stance, but there is a significant in the two trials compared to the EC bipedal stance. This is due to that the COP overshoots in EO bipedal stance are equally small in both AP and ML directions while they are equally large during single leg stance, while in EO bipedal stance the sway and overshoots concentrated on the AP direction thus distorting the ratio. The limitation of this study is that it only studied healthy young individuals and has no estimation on the behaviour of the new parameters on people suffering from any physical or neurological disorder. Another limitation is that only the time-distance type parameters were analyzed but none of the frequency type COP parameters. 5. Conclusion Considering the extremes along with the overall statistics of the COP movement is important, because it shows the random failures of the human balancing, which otherwise just distort the statistics. The study showed that it would be beneficial to use the two approaches complementarily. Usage of parameters with redundant information (such as confidence ellipse axes length and area together) is unnecessary. From the previously used parameters we recommend using the 95% confidence ellipse area, the COP path length, and maximum COP velocity along with the newly defined confidence ellipse axis ratio and AP-ML range ratio to qualitatively describe the COP motion over time. Examination of the maximum COP s in forward and backward directions in respect to the average COP is also recommended to describe the random failures of postural control, which dissolves using only statistical parameters. References [1] Terekhov Y., Stabilometry as a diagnostic tool in clinical medicine, Can Med Assoc J. 1976 Oct 9;115(7):631-3. [2] Lawson T. et al., Laboratory-based measurement of standing balance in individuals with knee osteoarthritis: a systematic review. Clin Biomech (Bristol, Avon). 2015 May;30(4):330-42. [3] Takacs J. et al., Test re-test reliability of centre of pressure measures during standing balance in individuals with knee osteoarthritis, Gait Posture. 2014;40(1):270-3. [4] Salavati M. et al., Test retest reliabty of center of pressure measures of postural stability during quiet standing in a group with musculoskeletal disorders consisting of low back pain, anterior cruciate ligament injury and functional ankle instability, Gait Posture. 2009 Apr;29(3):460-4. [5] Lafond D. et al., Intrasession reliability of center of pressure measures of postural steadiness in healthy elderly people, Arch Phys Med Rehabil. 2004 Jun;85(6):896-901. [6] Carroll JP., Freedman W., Nonstationary properties of postural sway, J Biomech. 1993 Apr-May;26(4-5):409-16. [7] Loughlin PJ. et al. Nonstationarities of postural sway, IEEE Eng Med Biol Mag. 2003 Mar-Apr;22(2):69-75. [8] Masui T. et al.: Increasing postural sway in ruralcommunity-dwelling elderly persons with knee osteoarthritis, J Orthop Sci. 2006 Jul;11(4):353-8. 144