Abstract
word count: 250 19 Text-only word count: 3519 20 Number of figures and tables: 2 figures and 1 table 21 22 23
2 Whole-body biomechanical load in running-based sports: the validity of 24 estimating ground reaction forces from segmental accelerations 25 26 Abstract 27 Objective: Unlike physiological loads, the biomechanical loads of training in running- based sport s are 28 still largely unexplored. This study, therefore, aimed to assess the validity of estimat ing ground 29 reaction forces (GRF), as a measure of external whole -body biomechanical load ing, from segmental 30 accelerations. 31 Methods: Fifteen team-sport athletes performed accelerations, decelerations, 90 ° cuts and straight 32 running at different speeds including sprinting. Full-body kinematics and GRF were recorded with a 33 three-dimensional motion capture system and a single force platform respectively. GRF profiles were 34 estimated as the sum of the product of all fifteen segmental masses and accelerations, or a reduced 35 number of segments. 36 Results: Errors for GRF profiles estimated from fifteen segmental accelerations were low (1 -2 N·kg -1 ) 37 for low-speed running, moderate (2 -3 N·kg -1 ) for accelerations, 90° cuts and moderate-speed running, 38 but very high (>4 N·kg -1 ) for decelerations and high-speed running. Similarly , impulse (2.3-11.1%), 39 impact peak (9.2-28.5%) and loading rate (20.1-42.8%) errors varied across tasks. Moreover, mean 40 errors increased from 3.26± 1.72 N·kg -1 to 6.76±3.62 N·kg -1 across tasks when the number of segments 41 was reduced. 42 Conclusions: Accuracy of estimated GRF profiles and loading characteristics was dependent on task , 43 and errors substantially increased when the number of segments was reduced. Using a direct 44 mechanical approach to estimate GRF from segmental accelerations is thus unlikely to be a valid 45 method to assess whole-body biomechanical loading across different dynamic and high -intensity 46 activities. Researchers and practitioners should, therefore, be very cautious when interpreting 47 accelerations from one or several segments, as these are unlikely to accurately represent external 48 whole-body biomechanical loads. 49
3 Keywords: Train ing load monitoring; Biomechanical loads ; Full-body segmental accelerations; 50 Loading characteristics; Segment reductions 51 52
4 Introduction 53 Training loads are monitored in sports as part of a process which aims to enhance performance, whilst 54 simultaneously reducing the risk of injur y. Although p hysiological loads have been investigated 55 extensively, biomechanical load measures are still limited and, therefore, largely unexplored 1 . Based 56 on the assumption that accelerations of the trunk are a good representation of whole -body centre of 57 mass (CoM) accelerations, trunk accelerometry derived load measures (e.g. New Body Load, Dynamic 58 Stress Load, PlayerLoad, Force Load) have been used to quantify and evaluate whole-body 59 biomechanical loads 2–6 . However, evidence relating accelerations of the trunk to established measures 60 of biomechanical loading is yet lacking. In fact, it has been shown that accelerations of individual 61 segments (including the trunk) cannot accurately represent whole-body biomechanica l loads 7–11 . 62 Ground reaction forces (GRFs) are a well-established measure of whole- body biomechanical loading . 63 GRFs have been used to optimise sprint performance 12,13 , improve running economy 14 and identify or 64 reduce potential injury risk factors 15,16 , and might thus be used to further understand the role of 65 external biomechanical forces in performance enhancement and injury prevention. Moreover, GRF 66 drives internal force production and contributes to internal stresses on e.g. muscles, tendons and bones 67 17,18 , which are currently difficult to measure in the field 1 . Since these structure- or tissue- specific 68 loads are the primary cause of e.g. overuse injur ies 19 , monitoring GRF in the field would be a first 69 step towards investigating internal biomechanical loads in more detail. However, valid methods for 70 accurately estimating GRF outside laboratory settings are currently unavailable. 71 Body-worn sensors, such as accelerometers, are commonly used in sport s to measure and monitor 72 numerous training load related metrics 20,21 . Given their widespread application to measure 73 accelerations of various body segments 22,23 , accelerometers might be used to estimate GRF , which can 74 be defined as the sum of the product of segmental mass and CoM
sensors, such as accelerometers, are commonly used in sport s to measure and monitor 72 numerous training load related metrics 20,21 . Given their widespread application to measure 73 accelerations of various body segments 22,23 , accelerometers might be used to estimate GRF , which can 74 be defined as the sum of the product of segmental mass and CoM accelerations of all body segments. 75 This alternative expression of Newton’s second law provides a way by which the contribution of 76 multiple segmental accelerations to the GRF can be systematically examined, especially since 77 accelerations of the trunk or other individual segment s have been shown to not be sufficient to 78 estimate GRF for several straight running and cutting activities 7–9,11,24 . Other studies have indeed 79
5 shown that for constant speed running, GRF can be estimated from seven 25 or eleven 26 segmental 80 accelerations measured with a laboratory based motion capture system. However, it is unknown 81 whether GRF for dynamic and high- intensity activities frequently undertaken in running -based sports 82 (e.g. rapidly accelerating, decelerating, cutting, sprinting) can be accurately estimated from segmental 83 accelerations and/or what the minimal required number of segments is. 84 If simultaneously measured segmental accelerations can be used to estimate GRF, this might 85 eventually allow GRF to be estimated in field settings and provide a meaningful measure of external 86 whole-body biomechanical loading. The aim of this study was, therefore, 1) to investigate whether 87 segmental accelerations measured in a laboratory setting can be used to estimate GRF for a variety of 88 dynamic and high- intensity tasks typically performed during running-based (team-) sports, and 2) to 89 determine the minimal number of segments required. 90 Methods 91 Participants. Fifteen team-sports athletes participated in this study (12 males and 3 females, 92 age 23±4 yrs, height 178±9 cm, body mass 73±10 kg). All participants were healthy and physically 93 active for at least three hours per week (sports participation 7±5 hrs per wk). This study was approved 94 by the Liverpool John Moores University ethics committee and participants provided informed 95 consent according to the ethics regulations. 96 Protocol. After a standardised warm -up, participants performed a range of dynamic and high-97 intensity running tasks including accelerations, decelerations, cutting, and steady running at constant 98 speeds ranging from 2 m ·s -1 to maximal sprinting (~7 m·s -1 , individual specific). Participants were 99 instructed to land with one whole foot on a single force platform embedded in the ground and 100 performed a minimum of five trials for each leg per task. For acceleration trials, participants were 101 instructed to accelerate from stand-still to their maximal sprinting speed (achieved in ~20 m), while 102 landing on the force platform for their second or third step of accelerating. For decelerations , 103 participants were instructed to decelerate
in the ground and 100 performed a minimum of five trials for each leg per task. For acceleration trials, participants were 101 instructed to accelerate from stand-still to their maximal sprinting speed (achieved in ~20 m), while 102 landing on the force platform for their second or third step of accelerating. For decelerations , 103 participants were instructed to decelerate as quickly as possible from maximal spr inting to immediate 104 stand-still, while landing on the force platform for their first or second step of decelerating. Cutting 105 trials were performed as a sharp change of direction on the force platform at a 90° angle from the 106
6 straight running direction. Steady (straight) running trials were performed at a constant low (2 -3 m·s - 107 1 ), moderate (4- 5 m·s -1 ) or high running speed (>6 m·s -1 ), including maximal sprinting. Running 108 speeds were measured with photocell timing gates (Brower Timing Systems, Draper, UT, USA) and 109 controlled by giving verbal feedback to speed up or slow down after each trial. Only trials within a ± 110 5% range of the target speed were included. 111 Kinematic and kinetic data collection. During the trials, full -body kinematic data were 112 collected using a seventy -six retro-reflective marker set attached to anatomical landmarks of the body 113 (appendix A ). Three-dimensional kinematic and kinetic data were synchronously recorded with ten 114 infrared cameras (Qqus 300+, Qualisys Inc., Gothenburg, Sweden) sampling at 250 Hz, in 115 combination with a single force platform (9287B, 90x60 cm, Kistler Holding AG, Winterthur, 116 Switzerland) embedded in the ground, sampling at 3000 Hz. Marker positions and ground reaction 117 forces (GRF) were recorded, synchronised and tracked using Qualisys Track Manager Software (QTM 118 version 2.16, Qualisys Inc., Gothenberg, Sweden). A static calibration was recorded at the start of each 119 session to determine the local coordinate systems, joint centres and segment dimensions for each 120 participant. From the marker data, a fifteen segment (head, trunk, pelvis, upper arms, forearms, hands, 121 thighs, shanks and feet) six-degree-of-freedom model was built, with segment mass and inertial 122 properties based on Dempster’s regression equations 27 and represented as geometric volumes 28 . 123 Kinematic and kinetic data were exported to Visual3D (C-motion, Germantown, MD, USA) and 124 Matlab (version R2017b, The MathWorks, Inc., Natick, MA, USA) for further processing and 125 analysis. 126 Data processing and analysis. Marker trajectories and force platform data were filtered with a 127 2 nd order Butterworth low-pass filter with 20 Hz and 50 Hz cut-off frequencies respectively. Trunk 128 defining marker trajectories were, however, filtered at 10 Hz based on a sensitivity analysis for 129 optimal GRF prediction (appendix B ). For each trial, touch-down
analysis. 126 Data processing and analysis. Marker trajectories and force platform data were filtered with a 127 2 nd order Butterworth low-pass filter with 20 Hz and 50 Hz cut-off frequencies respectively. Trunk 128 defining marker trajectories were, however, filtered at 10 Hz based on a sensitivity analysis for 129 optimal GRF prediction (appendix B ). For each trial, touch-down and take-off from the force platform 130 were identified by a 20 N threshold of the vertical GRF and resultant GRF was calculated from the 131 three individual force components (F x, Fy, Fz). The centre of mass (CoM) position for each segment 132 was used to define segmental movements from which accelerations were calculated as the double 133
7 differentiation (using three-point derivatives) of CoM motion along the three axes of the lab (x- y-z). 134 Resultant GRF curves were then estimated as the sum of the product of each segmental mass and CoM 135 acceleration in the three directions, according to equation 1. 136 GRF res,estimated=����a n,x∙m n� 1,2,3,…15 n=1 � 2 +���a n,y∙m n� 1,2,3,…15 n=1 � 2 +���a n,z∙m n� 1,2,3,…15 n=1 � 2 Eq. 1 In which a is the segmental acceleration, m the segmental mass and n the number of segments 137 included. To determine the number of segments required to accurately estimate resultant GRF, all 138 different segment combinations to estimate GRF from were examined. A total of 32,676 unique 139 combinations were analysed with a minimum of one and a maximum of fifteen segments. T o ensure a 140 constant total body mass, masses of the segments not included in a specific combination were equally 141 divided and added to the segment al masses that were part of that combination. 142 Measured and estimated GRF curves were normalised to each participant’s body mass. Accuracy of 143 estimated GRF profiles was evaluated by the absolute and relative curve root mean square errors 144 (RMSE). In addition, the accuracy of estimated GRF loading characteristics impulse (area under the 145 GRF curve), impact peak (force peak during the first 30% of stance) and loading rate (average GRF 146 gradient from touch- down to impact peak) was calculated and assessed. RMSE was rated as very low 147 (<1 N·kg -1 ), low (1- 2 N·kg -1 ), moderate (2- 3 N·kg -1 ), high (3- 4 N·kg -1 ) or very high (>4 N·kg -1 ). 148 RMSE values were analysed for all possible combinations of segments per task, as well as all trials 149 combined, to determine the best combination (i.e. lowest mean RMSE across trials) for each number 150 of segments. E stimated GRF loading characteristics errors were rated as very low (<5%), low (5-151 10%), moderate (10- 15%), high (15- 20%) or very high (>20%), which was based on mean ingful 152
of segments per task, as well as all trials 149 combined, to determine the best combination (i.e. lowest mean RMSE across trials) for each number 150 of segments. E stimated GRF loading characteristics errors were rated as very low (<5%), low (5-151 10%), moderate (10- 15%), high (15- 20%) or very high (>20%), which was based on mean ingful 152 performance or injury related differences in GRF 12,13,15 . Moreover, linear regression analyses were 153 performed between GRF loading characteristics (impulse, impact peak, loading rate) derived from the 154 estimated and measured GRF profiles. Regressions were performed per task, as well as for all trials 155 combined to examine the generalisability of GRF estimations across tasks, and rated as very weak 156 (R 2 <0.1), weak (R 2 =0.1-0.3), moderate (R 2 =0.3-0.5), strong (R 2 =0.5-0.7), very strong (R 2 =0.7-0.9) or 157 extremely strong (R 2 =0.9-1) 29 . Furthermore, Bland- Altman analyses 30 were performed across tasks to 158
8 explore mean difference s and 95% limits of agreement between the estimated and measured GRF 159 loading characteristics. 160 Results 161 Full body segmental accelerations. Accuracy of estimated GRF profiles from fifteen 162 segmental accelerations (full-body) varied across tasks (figure 1; table 1). Overall curve errors 163 (RMSE) were low for running at low speeds (2-3 m·s -1 ) and moderate for accelerations, 90° cuts and 164 moderate-speed (4-5 m·s -1 ) running. However, mean RMSE was very high for decelerations and high-165 speed running (>6 m·s -1 ). 166 The accuracy of estimated GRF loading characteristics varied between metrics and was dependent on 167 task (table 1). Impulses were accurately estimated with very low errors for 90° cuts and running at 168 constant low and moderate speeds, low errors for accelerations, and moderate errors for decelerations 169 and high- speed running. Similarly, impact peaks were estimated with low to moderate (9.2- 15%) 170 errors for all tasks, except accelerations, which had very high (28.5%) impact peak errors. Loading 171 rate errors however, were very high (20.1-42.8%) across all tasks. 172 Correlations and agreement between measured and estimated GRF loading characteristics across all 173 tasks varied. Impulses had extremely strong correlations, with a small bias and 95% confidence 174 interval of the limits of agreement (-0.04 to 0.45 N·s ·kg -1 ) (figure C .1 A and D; table 1). Despite the 175 very strong correlation and small bias for impact peaks however, there was a large variation of the 176 differences with limits of agreement ranging from -12.6 to 8.4 N·kg -1 (figure C .1 B and E). 177 Furthermore, measured and estimated loading rates had a strong correlation (R 2 = 0.68), but a large 178 bias and limits of agreement (-985 to 397 N·kg -1 ·s -1 ) (figure C .1 C and F). 179 Segment reductions. The best combinations of segments across all tasks for each given 180 number of segments are shown in table C .1. GRF estimated from a single segment was the best across 181 tasks from trunk accelerations, despite mean RMSE
Description
The study evaluates the accuracy of estimating ground reaction forces from segmental accelerations in sports.