A public dataset of running biomechanics and the effects of running speed on lower extremity kinematics and kinetics

Participants

The study analyzed a convenience sample of 28 subjects who were recruited through posted flyers, advertisement on the BMClab Internet home-page (http://demotu.org), and social media. The inclusion criteria included being a regular runner with a weekly mileage greater than 20 km, a minimum average running pace of 1 km in 5 min during 10-km races, and familiarity and comfort with running on a treadmill. The exclusion criteria were the presence of any neurological or musculoskeletal disorder that compromises its locomotion or the use of any assistive devices. The data related to demographics, running-training characteristics, previous injuries, and other relevant information were made available in the public dataset (see also the Table S1).

Equipment

The running kinematics were collected via a 3D motion-capture system with 12 cameras having 4 Mb of resolution and the Cortex 6.0 software (Raptor-4, Motion Analysis, Santa Rosa, CA, USA). The GRF data were collected via an instrumented, dual-belt treadmill (FIT, Bertec, Columbus, OH, USA).

The cameras were distributed around the laboratory such that they aimed at the instrumented treadmill’s motion-capture volume (Fig. 1). The cameras were mounted in a metallic truss setup structure with a length of 11.5 m, a width of 9.3 m, and a height of 2.8 m. This structure allowed positioning some cameras with varying elevations; however, most were placed atop the truss setup to optimize capturing the markers during the running trials (Fig. 1). The instrumented treadmill was mounted over a pit, with the treadmill surface at the same level as the laboratory floor (Fig. 1).

The Cortex 6.0 software (Motion Analysis, Santa Rosa, CA, USA) was used to (1) calibrate the motion-capture volume; (2) capture and identify the reflective markers; and (3) prepare the data and export it to the c3d file format. To provide an unbiased, raw dataset having marker trajectories and force signals, no further processing (e.g., filtering, gap filling) was performed on the data.

The motion-capture volume consisted of an area 3.1 m long, 2.3 m wide, and 1.2 m high, and this volume was calibrated daily. The system was deemed properly calibrated only if the length of the calibration wand, which was measured by the capture system, was within 0.10 mm of the true wand length. The rates of acquisition of the kinematics and kinetics data were set at 150 Hz and 300 Hz, respectively.

The laboratory-coordinate system used for the study was the same as that proposed by the International Society of Biomechanics (Wu & Cavanagh, 1995) and, as shown in Fig. 1, contained the following:

• X-axis in the direction of gait progression and positive pointing forward.

• Y-axis in the vertical direction and positive pointing upward.

• Z-axis in the medial-lateral direction and positive pointing to the right.

To record the strength and flexibility measures of selected muscle groups and joints, a hand-held dynamometer (HHD) (range: 0–1,330 N; accuracy: ±1%; resolution: 1 N; Nicholas MMT, Lafayette Instruments, Lafayette, IN, USA) and a magnetic-angle locator (Model 700; Johnson Level & Tool Mfg. Co., Inc., Mequon, WI, USA) were used, respectively.

Protocol

The data-collection protocol involved the following procedures:

• 1.

  Interview. Upon arrival, the participant was introduced to the laboratory and given a brief explanation of the experimental procedures. Then, the participant was asked to provide written informed consent and undergo a brief interview regarding eligibility criteria, demographic data, and running habits.

• 2.

  Preliminary measurements. Body height and mass were measured, and shoe-fitting was conducted to determine the appropriate shoe size. All participants wore neutral laboratory-controlled shoes (Nike Dual Fusion X).

• 3.

  Marker placement. The study used 48 technical and anatomical reflective markers (see details in Table 1 and Fig. 2). Clusters with four technical markers, placed in a rigid shell, were used on the thigh and shank segments. Their design was based on Cappozzo et al. (1997). These shells were securely fastened to the segments using a combination of elastic and Velcro straps.

• 4.

  Standing calibration trial. A template was used to align the subject’s feet in a standardized position such that the long axes of the feet were parallel to the X-axis of the laboratory-coordinate system (Fig. 2). Then, the markers’ 3D coordinates were recorded for 1 s.

• 5.

  Removal. After the calibration trial, the anatomical markers were removed except for those considered both anatomical and technical markers (T/A in Table 1).

• 6.

  The force plates were zeroed, the subject was asked to step onto the treadmill, and the following protocol was followed:

  • a.

    The subject walked at 1.2 m/s for 1 min to become familiar with the treadmill.

  • b.

    Next, the subject was asked to stay on the left belt of the treadmill, the belt speed was incrementally increased to 2.5 m/s, and after a 3-min accommodation period at this velocity, the data were recorded for 30 s. This procedure was repeated at speeds of 3.5 m/s and 4.5 m/s, always in the same sequence.

  • c.

    After the running trials, the treadmill speed was again set to 1.2 m/s for a 1-min cool-down period prior to being stopped.

• 7.

  Measuring the flexibility of the iliotibial band using the angle locator during the Ober’s test procedure. In brief, the test is performed with the subject lying on his/her side. The examiner then passively moves the tested leg (leg on top) into hip flexion, abduction, and extension and lowers the limb into adduction until it stops, limited by soft tissue stiffness. Further details about this test procedure can be found in Fukuchi et al. (2014).

• 8.

  Measuring the flexibility of the hip flexors using the angle locator during the Thomas’ test procedure. In brief, the test is performed while the subject lies supine with the hip joint positioned over the edge of the examination table and flexes the contralateral limb (hip and knee), bringing the thigh to the chest and holding it while the contralateral leg is suspended by the resistance imposed by the soft tissue to withhold the limb’s weight. Further details about this test procedure can be found in Fukuchi et al. (2014).

• 9.

  Three trials of maximal isometric voluntary contraction (MIVC) of the hip abductors, extensors, and internal and external rotator muscles were measured. The procedures used to take these measurements were described previously (Fukuchi et al., 2014).

The definition of the anatomical-segment coordinate system used to determine the 3D position and orientation of the lower extremity and pelvis segments was a combination of anatomical-frame conventions proposed previously (Cappozzo et al., 1995; Fukuchi et al., 2014). A model template file (RBDSmodelV3D.mdh) for the Visual 3D software (C-Motion Inc., Germantown, MD, USA) is available at Figshare. This .mdh file is an ASCII file containing the definitions of all landmarks, segments, and segment properties adopted by the present study.

Data processing and analysis

Raw marker-trajectory data and GRF data were filtered using a fourth-order, low-pass Butterworth filter with the same cut-off frequency of 10 Hz (Kristianslund, Krosshaug & Van den Bogert, 2012). The foot strike and toe off were determined when the vertical GRF crossed a 20-N threshold level. The foot strike patterns were determined using the strike index, which was calculated as the ratio of the center of pressure (COP) position in relation to the heel position, at foot strike, and the length of the foot. The measurements were taken, however, during instrumented treadmill running instead of on an overground condition as originally proposed by Cavanagh & Lafortune (1980).

In addition, 3D hip, knee, and ankle angles were calculated using Cardan angles, with the distal segment expressed relative to the proximal segment and adopting the following convention: the first rotation described occurred in the medial-lateral axis (Z-axis, perpendicular to the sagittal plane), which defines the flexion-extension movement; the third rotation described was around the longitudinal axis (Y-axis, perpendicular to the transverse plane), which defines the internal/external rotations; and the second rotation described was around an axis perpendicular to the previous two axes, which in the anatomic position represents the anterior-posterior axis (X-axis, perpendicular to the frontal plane), where abduction/adduction occurs. This convention is defined simply as the Z − X − Y convention and is frequently used to describe lower extremity rotations (Cappozzo et al., 1995). The net internal joint torques were represented in the joint-coordinate system (Schache & Baker, 2007) and were calculated using a standard inverse-dynamics approach. Joint powers were calculated as the scalar product of joint torques and joint angular velocities. The joint kinetic and the GRF variables were normalized by the subject’s body mass.

Joint angles, joint torques, joint powers, and GRF were normalized to the gait cycle over 101 time points. Then, these curves were averaged across trials, resulting in one curve comprising each subject’s average pattern. The number of footfalls varied with subject and speed, but the minimum number was always greater than 30. Next, to enable statistical comparison, discrete variables were calculated for each curve. Global maximum and minimum values for the joint angles and joint torques curves, GRF impulses, and joint work were considered for further analysis. The GRF impulses and joint work were calculated as the area under the GRF-time and joint power-time curves, respectively. The stride length and cadence were also calculated as the spatiotemporal gait parameters. These variables have been examined in previous studies related to running biomechanics in the context of injury etiology, running performance and aging (Fukuchi & Duarte, 2008; Fukuchi et al., 2011; Fukuchi et al., 2014; Fukuchi et al., 2016; Hall et al., 2013). The Visual 3D software program (C-motion Inc., Germantown, MD, USA) was used to filter the marker and GRF data and to calculate joint kinematics and kinetics. Finally, these time-normalized data were exported as ASCII text files. Then, the discrete variables, GRF impulses, and joint work were calculated using in-house algorithms written in Matlab 9.0 2016a (Mathworks Inc., Natick, MA, USA). A file written for the Visual 3D software program (RBDSpipelineV3D.v3s) is available at Figshare. This file is in text format and contains a series of pipeline Visual 3D commands that were used to process the raw c3d files, which are also available at Figshare. In addition to the raw c3d files, the time-normalized kinematic and kinetic data for each subject are available as ASCII files at Figshare (see the ‘Results’ for details). An exemplary Matlab code is also available in the Supplemental Information.

Statistical analysis of the processed data

The normality and homogeneity of variances assumptions of the dependent variables were tested using the Bartlett’s test. Either one-way ANOVAs or Kruskal-Wallis tests were conducted to examine the effect of running speed on gait-biomechanics variables when the dependent variables did or did not address the assumptions, respectively, at a significance level of 0.05. Whenever a main effect was observed, post-hoc analysis was conducted using t-tests with Bonferroni adjustments to minimize type I statistical errors. A multinomial logistic regression analysis was performed to determine the effect of running speed (the predictor) on foot-strike patterns (the categorical response variable). The statistical calculations were performed in Matlab 9.0 2016a (Mathworks Inc., Natick, MA, USA) using the Statistics toolbox.

Raw data

The raw data are stored in both c3d and text file formats. The c3d file format is a flexible binary file containing all the unprocessed data from a captured trial. This file format is supported by the main motion-capture manufacturers and other biomechanics software programs (e.g., Visual 3D). Although the Cortex software program, which was used to collect data, does offer the capability of processing and analyzing data, the raw files available in the present dataset contain only 3D, raw marker coordinates and transformed force signals (i.e., forces (N) transformed into the laboratory coordinate system).

The c3d files contain both the raw marker trajectories and force signals (both raw analog signals (V) and transformed signals (N)) in a single file. In contrast, separate text files were created for markers and forces signals, as these signals were sampled at 150 Hz for kinematics and 300 Hz for kinetics data. In addition, there is one c3d and one ASCII text file containing only marker trajectories corresponding to one second of the standing, anatomical calibration data of each subject. Finally, the average time-normalized kinematics (joint angles) and kinetics (joint torques, joint powers, and GRF) data for each subject are available in the ASCII file. Hence, as the running trials were performed on an instrumented treadmill at three distinct gait speeds (2.5 m/s, 3.5 m/s, and 4.5 m/s), there are four c3d files and eight text files per subject. Table 2 describes the file-naming convention used for the raw dataset.

The abbreviation RBDS in the file names stands for “Running Biomechanics Dataset” and xxx refers to the subject’s identification number (e.g., the first subject is 001). The c3d files can be analyzed using the Visual 3D software program and the supplied model template file (.mdh) along with the pipeline command files (.v3s) or other software, including Mokka (http://biomechanical-toolkit.github.io/mokka/). The ASCII files with marker trajectories contain 97 columns, with the first column containing the recording time (in seconds) and the remaining 96 columns being the X, Y, and Z coordinates (in millimeters) of the markers in the laboratory-coordinate system. The number of columns varied between the static and running trials (145 vs. 97, respectively) since the markers deemed solely anatomical were removed before the running trials (see the ‘Methods’). The headers of the marker files contain the markers’ labels (except for the first column, which is elapsed time) and are consistent with the “label” column in Table 1. In turn, the columns of the forces ASCII files are comprised of the forces, center of pressure, and free moment about the vertical axis measured by the instrumented treadmill and transformed on to the laboratory-coordinate system. Each force file has the following header: Time [s], Fx [N], Fy [N], Fz [N], COPx [mm], COPy [mm], COPz [mm], and Ty [Nm], followed by data in 9,000 rows and eight columns with six-digit numerical precision.

Metadata

One file named RBDSinfo (in both .txt and .xlsx formats) is supported with metadata, demographics, running characteristics, foot-strike patterns, muscle-strength and flexibility measurements for each subject. Below is the coding for the metadata. The first word identifies the name of the column in the header.

• 1.

  Subject: number of subjects (from one to 28).

• 2.

  Filename: names of files, including format extensions. Table 2 provides descriptions of the files.

• 3.

  Age: subject’s age in years.

• 4.

  Height: subject’s height in centimeters, measured with a calibrated stadiometer.

• 5.

  Mass: subject’s body mass in kilograms, measured with a calibrated scale.

• 6.

  Gender: subject’s gender (M or F).

• 7.

  Dominance: answer to the question “What leg would you use if you had to kick a ball (R or L)?”.

• 8.

  Level: answer to multiple-choice question about self-assessed level of running performance (only recreational; recreational competitive; competitive; elite).

• 9.

  Experience: number of months of regular running practice (at least three times/week).

• 10.

  SessionsPerWk: number of running training sessions per week.

• 11.

  Treadmill: number of treadmill running training sessions per week.

• 12.

  Asphalt: number of running training sessions on an asphalt surface per week.

• 13.

  Grass: number of running training sessions on a grass surface per week.

• 14.

  Trail: number of trail running training sessions per week.

• 15.

  Sand: number of running training sessions on sand per week.

• 16.

  Concrete: number of running training sessions on a concrete surface per week.

• 17.

  SurfaceAlt: number of running training sessions per week on other surfaces not listed previously.

• 18.

  RunGrp: whether the subjects participated in a running training group, as self-declared (Yes or No).

• 19.

  Volume: weekly running training volume (kilometers/week).

• 20.

  Pace: average running pace in the shortest long-distance running race (minutes/ kilometer).

• 21.

  RaceDist: running race distance participated in recently, as self-declared (in kilometers).

• 22.

  Injury: answer to the question “Have you experienced any injury or pain that has significantly affected your running practice?” (Yes or No).

• 23.

  InjuryLoc: anatomical location of the most recent injury.

• 24.

  DiagnosticMed: answer to the question “Was this injury medically diagnosed?” (Yes or No).

• 25.

  Diagnostic: diagnosis of running-related injury, as self-declared.

• 26.

  InjuryOnDate: approximate date of onset of injury symptoms, as self-declared (dd/mm/yyyy).

• 27.

  ShoeSize: size of running shoes, as self-declared (US standard).

• 28.

  ShoeBrand: preferred running shoe manufacturer, as self-declared.

• 29.

  ShoeModel: model of running shoes, as self-declared.

• 30.

  ShoePairs: current number of pairs of running shoes, as self-declared.

• 31.

  ShoeChange: answer to the following multiple choice question “How often do you replace your running shoes?” (less than 6 months; between 7 months and 1 year; between 1 and 2 years; more than 2 years).

• 32.

  ShoeComfort: comfort rating of their current running shoes, as self-declared on a 10-point rating scale. An average rating score was calculated if they had more than one pair of shoes.

• 33.

  ShoeInsert: type of foot insert (if any) worn in their running shoes (off-the-shelf insoles; orthotics; taping; none).

• 34.

  RFSI25: right foot-strike pattern (rearfoot, midfoot, or forefoot) while running at 2.5 m/s (see description in ‘Methods’).

• 35.

  RFSI35: right foot-strike pattern (rearfoot, midfoot, or forefoot) while running at 3.5 m/s (see description in ‘Methods’).

• 36.

  RFSI45: right foot-strike pattern (rearfoot, midfoot, or forefoot) while running at 4.5 m/s (see description in ‘Methods’).

• 37.

  LFSI25: left foot-strike pattern (rearfoot, midfoot, or forefoot) while running at 2.5 m/s (see description in ‘Methods’).

• 38.

  LFSI35: left foot-strike pattern (rearfoot, midfoot, or forefoot) while running at 3.5 m/s (see description in ‘Methods’).

• 39.

  LFSI45: left foot-strike pattern (rearfoot, midfoot, or forefoot) while running at 4.5 m/s (see description in ‘Methods’).

• 40.

  RThomas: angle of the right thigh relative to the horizontal during the Thomas’ test measured with a magnetic angle locator, in degrees (see description in ‘Methods’). Positive and negative values represent the thigh below and above a line parallel to the therapeutic bench.

• 41.

  LThomas: angle of the left thigh relative to the horizontal during the Thomas’ test measured with a magnetic angle locator, in degrees (see description in ‘Methods’). Positive and negative values represent the thigh below and above a line parallel to the therapeutic bench.

• 42.

  ROber: angle of the right thigh relative to the horizontal during the Ober’s test measured with a magnetic angle locator, in degrees (see description in ‘Methods’). Positive and negative values represent the thigh below and above a line parallel to the therapeutic bench.

• 43.

  LOber: angle of the left thigh relative to the horizontal during the Ober’s test measured with a magnetic angle locator, in degrees (see description in ‘Methods’). Positive and negative values represent the thigh below and above a line parallel to the therapeutic bench.

• 44.

  RHIPABD: average maximal isometric voluntary contraction (MIVC) of the right hip abductors measured by a hand-held dynamometer (HHD) in kilograms (see Fukuchi et al., 2014).

• 45.

  LHIPABD: MIVC of the left hip abductors measured by an HHD in kilograms.

• 46.

  RHIPEXT: MIVC of the right hip extensors measured by an HHD in kilograms.

• 47.

  LHIPEXT: MIVC of the left hip extensors measured by an HHD in kilograms.

• 48.

  RHIPER: MIVC of the right hip external rotators measured by an HHD in kilograms.

• 49.

  LHIPER: MIVC of the left hip external rotators measured by an HHD in kilograms.

• 50.

  RHIPIR: MIVC of the right hip internal rotators measured by an HHD in kilograms.

• 51.

  LHIPIR: MIVC of the left hip internal rotators measured by an HHD in kilograms.

Processed data

The processed data comprise average 3D time-normalized joint angles (hip, knee, and ankle), joint torques, and GRFs along with the joint powers at the sagittal plane. These data have six-digit precision (except the percentage column, which is an integer number), and they are stored in a single tab-separated ASCII text file. The processed data were stored such that their columns consisted of the following variables: Gait cycle (101 × 1) (percentage), 3D joint angles (101 × 9) (degrees), 3D joint moments (101 × 9) (Nm/kg), 3D GRFS (101 × 3) (N/kg), and scalar joint powers (101 × 3) (W/kg). The numbers within parentheses represent the dimensions of the matrix of data (number of rows and columns), considering only one gait speed and one lower limb. Table 3 displays the arrangement of the first 25 columns of processed data stored in the ASCII files. Since there are three gait speeds (2.5 m/s, 3.5 m/s, and 4.5 m/s) and two lower limbs (right and left), the resultant matrix has 101 rows and 145 columns (144 columns of biomechanics data plus one column of gait-cycle percentage data).

Effect of gait speed on running biomechanics

To study the effects of gait speed on running biomechanics, we compared the kinematics and kinetics running patterns of the subjects across gait speeds. Note that the subjects had an accommodation period at each running speed (see Protocol in the ‘Methods’) before the data were recorded. Figures 3 and 4 show the average pattern (across subjects) of the lower-extremity 3D joint angles and the joint torques curves, respectively. Figure 5 shows the average GRF curves in the medial-lateral, anterior-posterior, and vertical directions along with the lower extremity joint power curves at the sagittal plane. Overall, an increase could be observed in the magnitude of both kinematics (joint angles) and kinetic (torques, GRFs, and powers) variables following an increase in gait speed.

For a more specific, quantitative examination of the effects of gait speed, Table 4 shows the results of the descriptive and inferential statistical analyses. Six of the 24 variables did not meet the assumptions for ANOVA and they were compared using Kruskal-Wallis tests. Figure 6 shows repeated-measures plots with the distribution of the subjects’ data across running speeds for all variables that had significant effects. The results of the post-hoc analyses are indicated whenever a significant difference was found in the pairwise comparisons. The running-speed conditions of 2.5 m/s, 3.5 m/s, and 4.5 m/s are defined hereafter as V1, V2, and V3, respectively.