Validation of a hoof-mounted IMU sensor for detection of equine hoof motion on different grounds
Jenny Hagen1*, Franziska Theresa Wagner1, Joris Brouwer2, Ramon Bos2
1Institute of Veterinary Anatomy, Faculty of Veterinary Medicine, Leipzig University, Germany
2Joris Brouwer Consultancy – Research and product development, Doetinchem, The Netherlands
3Werkman Equilytics, Groningen, The Netherlands
*Corresponding author: Dr. habil. Jenny Hagen, An den Tierkliniken 43, 04103 Leipzig, Germany, email@example.com
Key words: horse, gait analysis, hoof-mounted sensor, optoelectronic examination, timing characteristics, step length
Objective: This study aimed to determine whether a hoof-mounted IMU system would provide similar timing characteristics and step lengths, in walk and trot on different grounds, compared to those of an established optoelectronic motion system.
Methods: The right hoof of one horse was equipped with an IMU sensor with an optical reference marker on top. Fifteen steps on firm ground, and five on penetrable ground, were performed at walk and trot. Level of agreement between the two measurement systems was determined by assessing the concordance correlation coefficients, accuracy, and precision of the duration of different gait events and step length.
Results: Timing characteristics and step length were in strong agreement between the two techniques for the majority of assessed parameters in walk and trot on both grounds. The agreement between techniques decreased for breakover and landing duration at trot on penetrable ground.
Conclusion: Disparity between the measurement techniques was related to difficulties in accurately defining single parts of the stance phase with marker-based optoelectronic analysis on penetrable ground. Detailed examinations of different parts of the stance phase are more accurately performed using hoof-mounted IMU sensors. Results emphasise the great potential of IMU sensors for equine motion analysis in daily practice.
The objective of the current study was to determine whether a hoof-mounted IMU system would provide similar timing characteristics and step lengths to those of a commercially available optoelectronic motion system in walk and trot. Recently, equine gait analyses for clinical and scientific purposes have been increasingly performed using non-optical, multiple component, motion sensor systems [1–3]. Sensors such as accelerometers, gyroscopes, magnetometers, global positioning systems, and temperature-correcting thermostats are combined and synchronised into a single sensor [4,5]. So-called inertial measurement units (IMU) are the key components of various systems available on the market for lameness detection and motion analysis [1,6]. These sensors enable measurements of multiple consecutive strides on different surfaces and under various conditions. Usually, their attachment is quick, easy to perform, and practical. Most IMU sensors are applied in veterinary practice for lameness detection, assessing motion symmetry of the head, wither, or pelvis with sensors attached to those body segments [3,7–9] [1,3]. However, equine-related fields such as farriery or training management are beginning to request the development of new, objective gait analysis systems [10,11].
Different parameters are of interest when evaluating gait quality related to trimming, shoeing, or performance [12–15]. With little additional effort a hoof-mounted sensor would provide information about detailed timing characteristics and spatial aspects describing the motion of the distal limb under different conditions. Parameters such as step duration (StepD), swing duration (SwingD), stance duration (StanceD), landing duration (LandD), midstance duration (MidD), and breakover duration (BoD) as well as step length (StepL) are relevant for examining the effect of trimming and shoeing on efficiency, cleanness, and soundness of equine gait [12,14–16]. However, to our knowledge the accuracy, precision, and agreement of a hoof-mounted IMU system, to evaluate detailed timing characteristics of gait quality in horses, has never been validated.
Moorman et al. (2012) compared acceleration of the hoof during the swing phase in the x-, y-, z- axes using an IMU with a reflective marker for optoelectronic analysis on top, attached to the lateral hoof wall . Overall, results for the linear and angular variables determined by the IMU correlated well with those of the 3-D optical kinematic system in walk and trot. Other groups have validated the accuracy and precision of detecting hoof-on and hoof-off gait events using limb-mounted IMU sensors in walk and trot [4,10]. Horses were equipped with metacarpal/metatarsal IMU sensors and reﬂective motion capture markers and guided over a force plate to calculate stance duration in milliseconds. Accuracy of timing parameters such as hoof-on/off, toe-on/off, heel-on/off was calculated as the difference in milliseconds between the IMU, or motion capture data, and the data from the force plate . Both groups concluded that IMU sensors can be used to determine temporal kinematic stride variables at walk and trot justifying its use in gait and performance analysis [4,10].
The aim of the current study was to assess the comparability of optoelectronic and IMU-based motion systems, attached to the equine hoof, in terms of timing characteristics of hoof motion and step length, in both walk and trot and on firm and penetrable ground. The following hypotheses have been examined:
- There is strong agreement between the motion analysis systems for less detailed parameters (StepD, SwingD, StanceD, and StepL)
- The agreement between the systems is weaker for detailed, fast motion events such as LandD and BoD.
- The agreement between the systems is higher in slow gait (walk) compared to fast gait (trot).
- The agreement between the systems is higher on firm ground than on penetrable ground.
2.1 Data acquisition
The right front hoof of a sound horse was equipped with an IMU sensor with the optical reference marker placed on top. The horse was previously checked by two independent veterinarians and one farrier for lameness and gait irregularities. At the time of measurement, the horse was sound and showed no lameness. It had been trimmed and shoed regularly, with the last shoeing 5 days earlier. The same experienced handler always guided the horse on a straight line, without much interference, when walking on firm ground (n = 15) and in trot (n = 15). The walkway was approximately 25 m long (figure 1). The area in which the step of the right front limb was detectable for the high-speed camera (camera 1)a had a length of 4.2 m and a width of 1.2 m. The horse started each walk at a defined position 10 m away from the examination window, with the right limb placed in front. Each walk ended 10 m past the examination window. A white screen was installed behind the examination window and the entire area was lit with two spotlights. The start of the walkway and the whole examination area were recorded by two extra cameras (camera 2 + 3)b. For each trail, right hoof motion was simultaneously recorded with the hoof-mounted IMU system. Measurements were repeated on penetrable ground in walk (n = 5) and trot (n = 5) with the same setup. The examinations were ethically approved.
2.1.1 Optoelectronic technique
The optoelectronic motion analysis was performed on 1000 fps videos captured with a high-speed camera (camera 1), perpendicular to the walkway at a distance of 3.5 m from the midline. Camera 2 and 3 were only used for documentation.
Prior to any testing, the right distal limb was equipped with markers for optoelectronic tracking. Three markers were used for calibration (figure 2). These were glued (21 cm apart) to a straight thin adhesive foil and taped laterally on a gaiter. This was affixed to the metacarpus of the horse. The hoof was equipped with markers laterally, at the coronary band, at the midline of the dorsal hoof wall, as well as lateral and dorsal at the IMU sensor. The marker lateral at the sensor was of interest for this validation. After adapting the horse to the setup and equipment, it was placed with the right limb in front at the start line. All cameras and the IMU sensor were started simultaneously at a signal, and the horse was guided through the examination window. The walk was valid, when a complete step of the right limb was recorded with camera 1.
2.1.2 IMU sensor technique
Each IMU sensorc consisted of a low-speed accelerometer, a high-speed 3D accelerometer, and a gyroscope. All measuring frequencies exceeded 1000 Hz. The low- and high-speed accelerometers had 1 mg and 100 mg resolutions respectively and ranges of +/- 32 g and +/- 200g. The gyroscope had a resolution of 0.12 deg/s and a range of 4000 deg/s. The IMU sensors were tightly fixed with strong-holding, self-adhesive Velcro taped to the midline of the dorsal hoof wall, underneath the coronary band (figure 3).
Prior to each walk, sensors were started simultaneously with all cameras using a trigger. Steps of each front foot were recorded until the horse reached the end of the walkway. The sensors were stopped manually. Data were automatically uploaded and saved on a tablet and a corresponding server.
2.2 Data analysis
For analysis, the step of the right hoof placed inside the examination window was identified by counting all previous steps in each video of camera 3. A valid measurement contained a whole step: from initial contact at the beginning of the first stance phase over swing phase to next stance phase until breakover. This specific step was taken to compare the values obtained with the optoelectronic and sensor-based technique.
2.2.1 Optoelectronic analysis
Video files of the measurements recorded with camera 1 (n = 40) were imported to the motion analysis program TRACKER®d. Contrast and brightness was optimally adjusted and the videos were calibrated to correct for length, projection obliquity, and distortion. Subsequently, the start and end of the step of interst were determined by marking the frames of initial contact of hoof on ground and the final contact prior to the second stance phase. The 21 cm scale was then calibrated and the coordinate system defined. The origin point (x/y = 0) as well as the x-axis should go through the lateral yellow marker of the sensor at mid-stance when the limb is vertical and there is no horizontal velocity. Auto-tracker function has been used for marker tracking with manual correction if necessary. A tracker frame around the yellow marker placed laterally on the sensor was defined. Properties for analysis were set to an evolution rate of 5% and an acceptance level of 5 for marker tracking. For each assessed frame the motion the position of the marker in the y-axis (vertical plane) and x-axis (horizontal plane) was calculated and displayed in a timing table and two-dimensional graph. From these data, parameters of interest were calculated by defining each motion event manually in the graph for each video.
2.2.2 IMU sensor analysis (BLACK®)
The BLACK® system uses an algorithm, developed in Matlab R2016b and compiled into C-code using the Matlab Coder Toolbox, to calculate all parameters of interest automatically for multiple strides. Parameters and internal data of the algorithm were extracted to make comparisons possible on an individual stride level.
2.2.3 Parameters of interest
Parameters of interest to describe timing and spatial aspects of each walk were: Step duration (StepD), Step length (StepL), Swing duration (SwingD), Stance duration (StanceD), Landing duration (LandD), Midstance duration (MidD), and Breakover duration (BoD) (figure 4).
StepD was defined as the duration in milliseconds (ms) from first initial contact during landing of the right hoof until the second initial contact. StepL was calculated in centimetres as the distance from the first to the second midstance of the right hoof. SwingD (ms) describes the time the hoof is without ground contact, beginning at first breakover and lasting until the second initial contact of the right hoof. StanceD (ms) is the time period from initial contact during landing until breakover of the right hoof. LandD (ms) is the time needed from initial contact until complete stabilisation of the hoof at the ground surface. MidD (ms) is the phase after full stabilisation of the limb at the ground until heel-off. BoD (ms) is the time period from heel-off until toe-off at the final part of the stance phase.
2.2.4 Statistical analysis
Statistical analysis was carried out with SPSS (v13.4). Mean, standard deviation (SD), maxima, and minima were calculated for StepD, SwingD, StanceD, LandD, MidD, BoD, and StepL. Concordance correlation coefficient (CCC), which can be decomposed into precision, location shift, and scale shift is a suitable method to compare two quantitative measurement techniques in the same individual . The method evaluates the agreement between two values from the same sample by measuring the variation from the 45° line through the origin (the concordance line). Lin’s CCC is considered the best measure of agreement for assessing the same continuous variable while evaluating reproducibility (Lin 1989, McBride 2005). CCC was chosen to fully assess the agreement and reproducibility between an optoelectronic motion analysis system and a non-optical motion sensor system in terms of timing parameters of the hoof and StepL in motion. According to Koch and Spoerl (2006) the following formula is used for calculation of the CCC .
Koch and Spoerl (2006) suggest the following descriptive scale for values of the concordance correlation coefficient for continuous variables (table 1) :
In addition, accuracy and precision of the values obtained with both techniques are calculated. Accuracy is a measure of statistical bias describing systematic error, calculated as the mean difference in milliseconds between the two examination techniques . Positive accuracy indicates an overestimation of the parameter calculated by the IMU (i.e. a longer duration or length than the optoelectronic system), whereas negative accuracy indicates an underestimation of that parameter. Precision of a measurement is the degree to which repeated measurements under the same conditions show the same findings. It is calculated as the standard deviation of the accuracy . Accuracy and precision are deemed better if closer to zero.
A total of 16 trails were needed to obtain 15 valid steps at walk on firm ground. After video processing and evaluation of step quality and the IMU data, n = 11 measurements were included into statistical analysis for walk on firm ground. For trot on firm ground, 26 walks were necessary to obtain 15 valid steps from which 14 were used for further analysis. On penetrable ground, 11 walks were needed to get 5 valid steps. All were included in further calculations. In trot on penetrable ground 22 walks were needed to record 5 valid steps, which were then used for further statistics. The speed of single measurements in the different gaits and on different grounds were quite stable with 1,41 m/s ± 0,08 m/s in walk on firm ground, 2,91 m/s ± 0,10 m/s in trot on firm ground, 1,3 m/s ± 0,18 m/s in walk on penetrable ground, and 3,05 m/s ± 0,2 m/s in trot on penetrable ground.
Assessed time graphs showing the vertical motion (z-axis, height) of the hoof during swing and stance phase are very similar (figure 5).
CCC values calculated for the different parameters under the specific conditions as well as accuracy and precision of all measurements are presented in Table 2. The agreement between the two motion analysis systems for less detailed timing characteristics, such as StepD, SwingD, StanceD, and StepL, was very strong with coefficients > 0,900. The strong agreement was independent of the ground surface and gait. Detailed timing parameters describing the stance phase, such as LandD, MidD, and BoD, also showed high CCC values. In particular, MidD had a coefficient > 0,960, except for trot on penetrable ground (CCC = 0,456). Fast motion events of the gait, such as LandD and BoD showed weaker agreement between examination techniques. The CCC for LandD showed only a moderate agreement in trot on penetrable ground (CCC = 0,559). Under all other conditions it ranged between 0,841 and 0,986. The lowest agreement between gait analysis techniques occurred for BoD with CCC values between 0,229 and 0,880. It became obvious that the agreement for BoD between the analysis techniques was better on firm ground. There was no difference in the agreement of the measurements in slow gait (walk) compared to fast gait (trot) on firm ground. However, faster gait combined with motion on penetrable ground showed the weakest coefficients for detailed parameters of the stance phase. For LandD, MidD, and BoD CCCs between 0,229 and 0,559 were demonstrated.
Accuracy and precision of the different examination techniques were best for measurements done in trot on firm ground. Accuracy was also closer to zero for values obtained in walk on firm ground compared to motion on penetrable ground. However, precision in walk on firm ground was lower for StepD, StanceD, and MidD. Indeed, the lowest precision of the timing parameters was present in walk on penetrable ground. Precision improved again in trot on penetrable ground. However, in walk on penetrable ground accuracy was lower for StanceD (12,13 ms) and BreakD (11,77 ms). In trot on penetrable ground accuracy was lowest for MidD (-24,38) and BreakD (18,41 ms).
4.1 Discussion of the findings
In the present study an established optoelectronic technique has been compared to a novel hoof-mounted IMU senor system to examine the motion of the equine hoof during swing and stance phase in walk and trot on two different grounds. The aim of the study was to assess whether the sensor-based system would provide similar data to those of an established marker-based motion analysis system. The agreement between the results of the motion analysis techniques has been described with the CCC, accuracy, and precision of the timing parameters and step length. Previous validation tests and studies describing hoof-mounted IMU sensors for equine motion analysis did not evaluate different parts of the stance in detail. Clinically, the stance phase, in particular landing (impact) and breakover (push-off), is more important in the pathogenesis of injury than is the swing phase [1,17]. The reliability of the IMU technique for motion analysis has been improved in recent years, as is seen in reduced ringing artefacts or the use of combined high and low g 3D-accelerometers and gyroscopes [4,6,17]. High correlation between the measurement techniques is evidence that the IMU technique has been improved.
The comparison of the timing characteristics representing general motion of the hoof during swing and stance phase showed a high agreement between the examination techniques. CCC values > 0,900 for StepD, SwingD, StanceD, and StepL confirm hypothesis 1. Since more detailed information on the motion of the hoof during the stance phase is of interest in equine orthopedics and gait analysis [12,15,19], LandD, MidD, and BoD were calculated separately for each technique. In comparison to CCCs for MidD (> 0,960 for all conditions except for trot on soft ground), slightly weaker agreement between the systems was found for faster events such as LandD and BoD. These findings support hypothesis 2, stating that the agreement between the motion analysis systems is weaker during fast motion events of the gait. Since landing and breakover were defined manually in the marker-based analysis, subjective failures have to be expected. The faster the motion of the hoof the more difficult it is to define the exact starting and ending point of these events in the optoelectronic measurements. In particular, the exact moment when the heels lift off the ground is hard to see in the high-speed motion videos and associated graphs. Zooming in the image can help but in turn leads to poor quality and reduced accuracy, further reducing accuracy of event management. In the marker-based time-curves it was possible to detect fast motion events such as landing and breakover. Still, a decrease of marker motion upon landing was easier to define than the motion increase at the start of breakover. In contrast, the IMU-system was sensitive to the first tilt of the dorsal hoof wall, initiating breakover shortly before heel-off. This explains the slight disagreement of BoD between the systems. Yet, CCC of BreakD ranged between 0,807 (trot) and 0,880 (walk) on firm ground. With regard to the influence of the different gaits on the agreement of the timing characteristics, hypothesis 3 cannot be confirmed for examinations on firm ground. The agreement between the values of both gait analysis techniques was very strong, whether measured in walk or trot. These gaits are the most common for subjective and objective lameness examination or evaluation of the effect of trimming or shoeing on equine gait [1,3,12]. Since motion on penetrable footing is relevant for most athletic disciplines [20,21], such as dressage or show jumping, the current study validated both systems on soft bedding, as used for training and competition. Such bedding provides a higher degree of elasticity leading to less sinking into the footing [21,22]. This effect becomes quite obvious for the values assessed in walk. Opposite to hypothesis 4, CCC values for walk on penetrable ground are similar to data assessed in walk on firm ground. A strong agreement between both examination techniques occurred in walk on penetrable ground, except for breakover (CCC = 0,660). Compared to faster gaits, the hoof sinks less into the ground in walk, so that the swing phase and the different parts of stance phase are still well detectable in the optoelectronic measurements. Only breakover is difficult to define, since the tip of the hoof rotates into the ground shortly before early heel-off. Hence, heel-off is detectable later in the marker-based time-curves compared to the IMU-sensor data, which remain unaffected by this failure due to adapted algorithms. This effect increases in trot on penetrable ground supporting hypothesis 4. Due to higher impact on the ground surface in faster gaits , the hoof sinks deeper into the penetrable footing, causing a decrease in accuracy for defining the different parts of the stance phase with the optoelectronic technique. Initial ground contact is hard to see if the hoof sinks into the footing during landing. In addition, the hoof is partially covered with footing material making it difficult to see the start of breakover in the videos and graphs of optoelectronic motion analysis. If landing and breakover are poorly defined, accuracy of the midstance duration decreases accordingly. In contrast, the IMU sensors accurately determine landing using adaptive thresholds on acceleration signals and their time derivatives (jerk signals) independent of the ground firmness. These facts are represented by the weaker CCC values for LandD, MidD, and BoD in trot on penetrable ground. The disagreement between the systems is associated with the differences of data analysis related to the described problems with the optoelectronic measurements. The highest accuracy and precision in trot on firm ground are explainable because penetrable ground provides less stability than firm ground. In addition, in slow gaits there is more time for variations in limb placement, breakover, and swing phase compared to faster gaits, such as trot [24,25]. These facts cause higher inter-stride variability, decreasing precision. Less accuracy of some parameters on penetrable ground was related to greater differences between the means of the variables assessed with the different systems. Usually, the higher the agreement between the systems the greater was the accuracy of the values.
4.2 Comparison of both techniques
The two systems compared in the current study are very different motion analysis techniques in terms of data acquisition and analysis, but also with regard to practicability and usability in the equine practice. The requirements to create high-quality high-speed videos for optoelectronic marker tracking are very high . Time, cost, and staff expenses to build the whole setup are limiting the use of optoelectronic measurements in daily practice [27,28]. Using a hoof-mounted IMU sensor for motion analysis requires no more than a correct fixation of the sensor at the dorsal hoof wall and a plane, straight ground surface for walking the horse. Fixation of the sensors has been improved compared to previous approaches and the fit is tight and strong enough to measure horses at fast speeds. Equipping the horse with four sensors, conducting the measurement, and receiving the fully analysed results takes only a few minutes. In addition, using optoelectronic motion analysis techniques requires the horse to walk in a defined examination window, making it difficult to measure multiple consecutive strides in one walk. This is a big disadvantage compared to sensor-based examination. The IMU-system used in this study averages the data of as many strides as wanted and automatically selects valid strides for further analysis. The most reliable output is produced with an average of 20 strides or more. Since visibility of the markers for the high-speed camera is crucial to perform optoelectronic measurements, examinations on penetrable ground surfaces are limited. Markers covered with dirt or inappropriate detection of the different gait events makes it difficult to obtain accurate values for detailed hoof motion analysis in different parts of the stance phase. Sensor-based measurements are not prone to this failure. The automatic, algorithm-based event management of the different gait phases and automatic visualisation of the data assessed with the IMU-system used here minimises failures related to subjective, manual definition of the different gait events. In addition, the automatic data analysis and visualisation is very fast and requires no effort, compared to marker tracking of the optoelectronic measurements. Results can be directly shown and discussed with owners, colleagues or trainers. Automatic documentation of follow-up examinations is possible to compare the influence of different orthopaedic treatments, shoeing, or training on the equine gait quality.
It can be concluded that the agreement of the timing characteristics and the step length assessed with an optoelectronic system and an IMU-based sensor technique in walk and trot on different grounds is very strong for the majority of assessed parameters. The weaker agreement of BreakD and LandD between the motion analysis techniques in trot on penetrable ground was related to the difficulties accurately defining the parts of the stance phase with the marker-based optoelectronic analysis. Detailed examinations of different parts of the stance phase are more accurately performed with hoof-mounted IMU sensors. In general, the sensor-based examinations are more practical, faster, and easier to perform compared with optoelectronic examinations. The potential of hoof-mounted IMU sensors for examination of equine motion in daily diagnostics or evaluation of trimming or shoeing seems to be high.
a Chronos 1.4, KRO00580, Metesco Netherlands, 2958 SX Ridderkerk, The Netherlands
b Casio Exilim EX-FH25, CASIO Europe GmbH, Norderstedt, Germany
c Werkman BLACK®, Werkman Equilytics, Groningen, The Netherlands
d Tracker® Cabrillo Community College District, Aptos Campus, 6500 Soquel Drive Aptos CA 95003, USA
Table 1: Descriptive scale for interpretation of the concordance correlation coefficient for continuous variables according to Koch and Spoerl (2006)
Table 2: Means, standard deviations (SD), concordance correlation coefficients (CCC), accuracy and precision of timing characteristics and step length on firm and penetrable ground in walk and trot assessed with optoelectronic (OE) and IMU sensor-based (IMU) motion analysis systems
Figure 1: Walkway examination window set-up of motion analysis techniques and position of the horse for optoelectronic and sensor-based motion analysis performed in the current study.
Figure 2: Attachment and position of the calibration and reference markers.
Figure 3: Attachment and position of the IMU sensors.
Figure 4: Definition and description of the timing characteristics and step length calculated in this study.
Figure 5: Time graphs showing the motion of the right hoof calculated by tracking the reference marker (orange) and the values assessed with the IMU sensor (blue) in the y-axis (vertical plane) and x-axis (horizontal plane).
Many thanks to Christel Werkman, Hero Werkman, Peter Werkman, Mehmet Yazgonul, Kristian…
 Keegan KG. The Lameness Locator (wireless inertial sensors for detection of lameness in horses). Proceedings of the 15th Congress of the European Society Veterinary Orthopaedics and Traumatology 2010;15:215–7.
 McCracken MJ, Kramer J, Keegan KG, Lopes M, Wilson DA, Reed SK et al. Comparison of an inertial sensor system of lameness quantification with subjective lameness evaluation. Equine Veterinary Journal 2012;44(6):652–6.
 Pfau T, Robilliard JJ, Weller R, Jespers K, Eliashar E, Wilson AM. Assessment of mild hindlimb lameness during over ground locomotion using linear discriminant analysis of inertial sensor data. Equine Vet J 2007;39(5):407–13.
 Olsen E, Pfau T, Ritz C. Functional limits of agreement applied as a novel method comparison tool for accuracy and precision of inertial measurement unit derived displacement of the distal limb in horses. J Biomech 2013;46(13):2320–5.
 Tong K, Granat MH. A practical gait analysis system using gyroscopes. Med Eng Phys 1999;21(2):87–94.
 Pfau T, Witte TH, Wilson AM. A method for deriving displacement data during cyclical movement using an inertial sensor. J Exp Biol 2005;208(Pt 13):2503–14.
 Keegan KG, MacAllister, C. G. W, D. A. G, C. A., Kramer J et al. Comparison of an inertial sensor system with a stationary force plate for evaluation of horses with bilateral forelimb lameness. American journal of veterinary research 2012;73(3):368–74.
 Keegan KG, Wilson DA, Kramer J, Reed SK, Yonezawa Y, Maki H et al. Comparison of a body-mounted inertial sensor system-based method with subjective evaluation for detection of lameness in horses. American journal of veterinary research 2013;74(1):17–24.
 Kelmer G, Keegan KG, Kramer J, Wilson DA, Pai FP, Singh P. Computer-assisted kinematic evaluation of induced compensatory movements resembling lameness in horses trotting on a treadmill. American journal of veterinary research 2005;66(4):646–55.
 Bragança FM, Bosch S, Voskamp JP, Marin-Perianu M, van der Zwaag BJ, Vernooij JCM et al. Validation of distal limb mounted inertial measurement unit sensors for stride detection in Warmblood horses at walk and trot. Equine Veterinary Journal 2017;49(4):545–51.
 Dallap Schaer BL, Ryan CT, Boston RC, Nunamaker DM. The horse-racetrack interface: A preliminary study on the effect of shoeing on impact trauma using a novel wireless data acquisition system. Equine Vet J 2006;38(7):664–70.
 Clayton HM, Sigafoos, R., Curle, R. D. Effect of three Shoe Types on the Duration of Breakover in sound Trotting Horses. J Equine Vet Sci 1991;11(2):129–32.
 Roepstroff L, Johnston C, Drevemo S. The effect of shoeing on kinetics and kinematics during the stance phase. Equine Vet J Suppl 1999;30:279–85.
 van Heel MCV, van Weeren PR, Back W. Shoeing sound Warmblood horses with a rolled toe optimises hoof-unrollment and lowers peak loading during breakover. Equine Vet J 2006;38(3).
 Weishaupt MA, Waldern NM, Amport C, Ramseier LC, Wiestner T. Effects of shoeing on intra- and inter-limb coordination and movement consistency in Icelandic horses at walk, tölt and trot. Vet J 2013;198 Suppl 1:13.
 Waldern NM, Wiestner T, Ramseier LC, Amport C, Weishaupt MA. Effects of shoeing on limb movement and ground reaction forces in Icelandic horses at walk, tölt and trot. Vet J 2013;198 Suppl 1:8.
 Moorman VJ, Reiser RF, McIlwraith CW, Kawcak CE. Validation of an equine inertial measurement unit system in clinically normal horses during walking and trotting. Am J Vet Res 2012;73(8):1160–70.
 Koch R, Spörl E. Statistische Verfahren zum Vergleich zweier Messmethoden und zur Kalibrierung: Konkordanz-, Korrelations- und Regressionsanalyse am Beispiel der Augeninnendruckmessung. Klinische Monatsblatter fur Augenheilkunde 2007;224(1):52–7.
 Eliashar E. An evidence-based assessment of the biomechanical effects of the common shoeing and farriery techniques. Veterinary Clinics of North America: Equine Practice 2007;23(2):425–42.
 Murray RC, Walters, Juli M., Snart H, Dyson SJ, Parkin TDH. How do features of dressage arenas influence training surface properties which are potentially associated with lameness? Vet J 2010;186(2):172–9.
 Northrop AJ, Dagg L-A, Martin JH, Brigden CV, Owen AG, Blundell EL et al. The effect of two preparation procedures on an equine arena surface in relation to motion of the hoof and metacarpophalangeal joint. Vet J 2013;198 Suppl 1:42.
 Chateau H, Camus M, Holden-Douilly L, Falala S, Ravary B, Vergari C et al. Kinetics of the forelimb in horses circling on different ground surfaces at the trot. Vet J 2013;198:e20-e26.
 Back W, Schamhardt HC, Barneveld A. Are kinematics of the walk related to the locomotion of a warmblood horse at the trot? Vet Q 1996;18(sup2):79–84.
 Nauwelaerts S, Hobbs SJ, Back W. A horse’s locomotor signature: COP path determined by the individual limb. PloS one 2017;12(2):e0167477.
 Roach JM, Pfau T, Bryars J, Unt V, Channon SB, Weller R. Sagittal distal limb kinematics inside the hoof capsule captured using high-speed fluoroscopy in walking and trotting horses. Vet J 2014;202(1):94–8.
 Barrey E. Methods, Applications and Limitations of Gait Analysis in Horses. Vet J 1999;157:7–22.
 Chateau H, Degueurce C, Denoix J-M. Evaluation of three-dimensional kinematics of the distal portion of the forelimb in horses walking in a straight line. Am J Vet Res 2004;65(4):447–55.
 Chateau H, Degueurce C, Denoix J-M. Three-dimensional kinematics of the distal forelimb in horses trotting on a treadmill and effects of elevation of heel and toe. Equine Vet J 2006;38(2):164–9.
 Weller R, Pfau T, BABBAGE D, BRITTIN E, May SA, Wilson AM. Reliability of conformational measurements in the horse using a three-dimensional motion analysis system. Equine Veterinary Journal 2006;38(7):610–5.