Multibody kinematic optimisation vs body fat: A performance analysis

We have analysed the performance of mulitbody kinematic optimisation methods in reducing soft tissue artefacts for subject data of varying body fat percentages. Multibody kinematic optimisation methods are a critical aspect of movement analysis using musculoskeletal modelling software. By minimising soft tissue artefacts, they help in achieving higher fidelity joint kinematics and dynamics analyses. Prior studies have not examined the performance of multibody kinematic optimisation on subjects of varying body fat percentages. Herein, we: 1) have analysed the efficacy of three different multibody kinematic optimisation methods on varying body fat percentages, 2) implemented a novel weighting scheme to reduce error irrespective of body fat percentages. Residual error using gait data of 50 participants of varying body fat percentages was calculated through inverse kinematic analysis using OpenSim(c) musculoskeletal modelling software. The analysis was repeated using a time-based weighting scheme. The residual error of participants with higher body fat percentages was greater by 30% when compared to residual error of participants of lower body fat percentages. Additionally, time-based weighting scheme reduced residual error by 20% on average compared to constant-value weighting scheme. Our results indicate that multibody kinematic optimisation methods are adversely affected by higher body fat percentages and that time-based weighting can provide higher fidelity movement analysis irrespective of body fat percentages. Through our results we aim to develop tools which provide greater precision in obesity-related movement analysis. Such tools could also help address the disparities in rates of obesity associated with different ethnic or socioeconomic background.

Introduction rigid displacement of the cluster. [16] MKO methods, such as the global optimisation method (GOM) [15], Local Marker Estimation (LME) [17,18] and Kalman Smoothing 40 (KS) [19] reduce soft tissue artefacts by determining the optimal pose of a virtual model 41 in musculoskeletal modelling software. The optimal pose is determined by minimising 42 the error between virtual markers in the modelling software and experimental markers 43 worn by the patient. 44 Whilst MKO methods have been shown to reduce the influence of soft tissue 45 artefacts on calculated joint kinematics, validation has only been performed on either 46 simulated data, or on data from subjects with a body mass index (BMI) of less than 25,47 which is considered as a healthy weight for their height. [7,20]  of the MKO method used. [21][22][23] Obesity-specific marker sets, which use a 52 combination of traditional markers and digitally probed and placed markers, have been 53 shown to calculate the kinematics and dynamics of obese participants. [22,23] Research 54 has also been carried out to determine the effects of removing thigh and shin markers 55 on calculated kinematics [24,25] as thigh and shin markers are generally more prone to 56 soft tissue artefacts. [4,7,26] 57 This paper aims to test two hypotheses: a) The performance of MKO methods is 58 adversely affected by body fat and b) time-based weighting [27] could be an effective 59 way of reducing residual errors during the inverse kinematic step. The first hypothesis is 60 tested by comparing residual error in the inverse kinematic step using three MKO 61 methods on gait data collected using skin-based markers from people with varying body 62 fat percentages as shown in Fig 1. Additionally, by altering the weight of the thigh 63 marker at certain time instances through correlation of thigh-marker errors and total 64 errors, we aim to test whether time-based weighting can lead to reduction in total Workflow for analysing performance of MKO methods on varying body fat percentages. a) Collection of treadmill gait data using skin-mounted markers placed on participants of varying body fat. b) Generic OpenSim musculoskeletal model is scaled to subject specific dimensions using the gait data. c) Inverse kinematic analysis is performed using three different MKO methods. c1) Shows the difference in position of virtual marker (pink) and experimental marker (blue) for subjects of low body fat (blue arrow) and c2) shows the difference in virtual marker (pink marker) and experimental marker (blue marker) for subjects of high body fat (light blue arrow). d) Residual error obtained using three different MKO methods for subjects of low and high body fat is compared Workflow for analysing performance of time-based weighting for reducing residual error. a) Collection of treadmill gait data using skin-mounted markers placed on participants of varying body fat. b) Generic OpenSim musculoskeletal model is scaled to subject specific dimensions using the gait data. c) and d) Inverse kinematic analysis is performed using three different weighting schemes. e) Residual error obtained from the three different weighting scheme is compared  Additionally, body fat at the thigh and hip were measured using skin calipers.

78
Data was captured using Motive 2.4 and 10 Flex 3 cameras from OptiTrack.

79
Reflective markers were attached to each participant at specific anatomical landmarks 80 on the lower body as specified by lower body Helen Hayes biomechanical markerset as 81 shown in Fig 3. As part of the data collection protocol, participants were asked to stand 82 stationary in a 'T' pose followed by walking on the treadmill at speeds of 0.1m/s, 83 0.3m/s and 0.5m/s. These speed values were determined beforehand in order to reflect 84 standard clinical protocols. The raw data was filtered with a cut-off frequency of 10Hz; 85 missing data frames were filled using either cubic interpolation function or pattern fill

Residual error calculation
Residual errors from three MKO methods were calculated using musculoskeletal 89 modelling software OpenSim. [28,29] Residual error is calculated by taking the root 90 mean square difference between the trajectory of the experimental marker and 91 corresponding model marker in OpenSim. The three MKO methods analysed were the 92 global optimisation method(GOM), [15] the Local marker estimation method [17,30] 93 and the Kalman smoothing method. [19] Prior to the calculation of residual errors using 94 these three methods, OpenSim's generic musculoskeletal model, gait2354, was scaled to 95 subject-specific dimensions using either the static trial or the first frame of a dynamic 96 trial. [31] Guidelines were followed to ensure that the error from the scaling procedure To test the efficacy of time-based weighting scheme on reducing residual errors, a 101 progressive weighting strategy was integrated into OpenSim; this weighting scheme has 102 been previously applied to reduce errors resulting from occluded or missing 103 markers. [27] The weight of a marker is progressively decreased to zero to coincide with 104 the frame of interest, and progressively increased afterwards to its original value. The 105 frame of interest and the weighting factor by which the weight is increased or decreased 106 needs to be specified by the user. [27] The frame of interest was calculated by        Table 3. Median values of total residual error and residual error of thigh marker for Groups 5 and 6 Estimated marker trajectory for the thigh marker calculated using GOM, LME and 176 KS for a participant with low body fat is shown in Fig 10, to total residual error obtained using the default weighting scheme for the three walking 189 speed was 58%, 54% and 40% for participants with higher body fat percentages and 190 55%, 51% and 29% for participants with lower body fat percentages. When compared 191 to total residual error obtained when thigh markers were omitted, the reduction in error 192 at the three walking speeds were 41%, 35% and 29% for participants with high body fat 193 percentages and 29%, 29% and 14% for participants with low body fat percentages.A

194
Wilcoxon signed rank test indicated that these results were statistically significant 195 (p=0.0020,p=0.0020,p=0.0020,N=10) for high body fat percentages and low body fat 196 percentages (p=0.0059,p=0.0020,p=0.0020,N=10).  and on the kinematic and dynamic analyses of various movement studies. [32,33] 206 Musculoskeletal modeling, through its incorporation of MKO methods for compensating 207 soft tissue artefacts, has been increasingly used for movement analysis in both clinical 208 and research settings. [10,26] These optimisation methods [15,19,34,35] have been 209 predominantly validated on simulated movement data, with experimental data 210 restricted to subjects with healthy or ideal body fat percentages and BMI scores. [7] 211 Sensitivity analyses of musculoskeletal models to soft tissue artefacts have shown that 212 soft tissue artefacts can cause variation of up to 30% in estimating joint forces. the markers were used, the greatest error reduction in joint angles was obtained when 242 two markers were included in the thigh and shank. [24,25,37]  The main limitation of this study is that the analyses and results are based on can enhance our understanding and diagnosis of obesity related movement disorders.

293
The prevalence of such disorders may vary based on ethnicity and socioeconomic 294 conditions [39], therefore tools which address the effects of obesity on movement, may 295 help to promote more equitable healthcare.
We would like to thank the participants for their cooperation and involvement in the 298 study. We would also like to thank OptiTrak for their technical support provided during 299 the study.