Lameness is one of the most common reasons that horses are examined by veterinarians and may account for > $1 billion/y in expenses.1 Subtle lameness may be difficult to detect, and a lameness examination, as performed by both experienced and inexperienced veterinarians, is subjective with high interobserver and intraobserver variability.2,3 One proposed reason for the high amount of variability is the various aspects of gait or body position (ie, head position, foot flight, and limb movement) that are evaluated by a veterinarian.4 The most commonly used lameness grading scale in the United States is the American Association of Equine Practitioners scale,5 which has strict criteria for each grade but also allows for a large amount of variability within each grade.4 In a recent study,6 investigators reported that there was better interobserver agreement with an American Association of Equine Practitioners lameness score > 1.5 (93.1%) than for a score < 1.5 (61.9%). Thus, because the subjective lameness examination cannot be judged as a true criterion-referenced standard of lameness diagnosis,6 there is a need for more quantitative motion analysis systems to describe motion and lameness of horses.
Currently, the most accepted methods for quantitative equine gait analyses are force platforms and 3-D optical kinematics. Both of these systems have variables that are well correlated with lameness, including the kinetic variables peak vertical force and impulse7 and the kinematic variables extension of the metacarpophalangeal (or metatarsophalangeal) joint (fetlock joint), changes in protraction and retraction of the distal aspects of limbs, temporal changes in stride parameters, and alterations in maximum hoof height.8,9 However, these systems are best used in research settings because they are available at limited locations, are expensive, require special training for use and interpretation of results, and typically only measure 1 or 2 gait cycles at a time4 because there is usually a small camera field of view and only 1 or 2 consecutive force platforms. Considering that subtle lameness may not be detected at every stride or at every velocity, force platform and 3-D optical kinematics may not be able to evaluate the exact strides at which lameness is apparent.4
A force-measuring treadmill has been developed10 that allows for multiple strides to be evaluated with regard to optical kinematics and vertical forces. Its ability to measure only vertical forces10 is 1 limitation of this treadmill. However, it also has limited availability, and because it is a custom-made, expensive piece of equipment, it is not likely that it will be of use in clinical settings. Additionally, it is often difficult to separate the forces associated with each hoof when the treadmill is used, especially when the detail for detection of subtle lameness is necessary. A force-measuring shoe has been developed that has application for clinical settings,11 but there is some concern that the weight of the shoe may impact movement of the distal aspect of a limb, which could influence the evaluation of normal gait and lameness.4 Also, force-measuring shoes currently are machined to fit a particular horse.12
Several nonoptical motion sensor systems have been developed for use in analysis of equine gaits.13–16 These motion sensor systems have multiple components, including accelerometers, gyroscopes, magnetometers, and global positioning system data logging systems. Those composed of an accelerometer, gyroscope, magnetometer, and temperature-correcting thermostat are often referred to as inertial sensor or IMU systems.15 The most widely used commercially available system uses gyroscopes and accelerometers to identify asymmetric head or pelvic movement to identify a lame limb.13 Therefore, a hoof-mounted IMU system would provide additional information about alterations in mechanics of the distal aspect of a limb that a body-mounted sensor would be unable to provide.
The IMU technology may provide another avenue of equine motion analysis because of the sensor's small size, portability, and ability to measure multiple consecutive strides in field settings. One human IMU system requires the sensors to be attached to each other by wires, which could be cumbersome if used on the distal aspect of the limbs of horses. This human IMU system has been used to measure equine trunk movement,15,17 but the sensors can only measure accelerations of ± 10 g, which may not be high enough for the distal aspect of limbs. Investigators in 1 study15 measured accelerations of ± 5 g when the inertial sensor was mounted on the most dorsal aspect of the shoulders (withers) of a horse traveling at a fast canter. Accelerometers used on the hoof of a horse can typically measure accelerations in the range of ± 9,800 m/s2.15,18 Thus, IMU systems designed for use in the study of human locomotion are not adequate for use in examining locomotion of the distal aspect of the limbs and hooves of horses at high speeds. A newly developed equine IMU system differs from other motion analysis systems in that it uses multiple sensors (placed on the head and the distal aspects of up to 4 limbs), which are wireless. In addition, this equine IMU system was developed for potential use on the distal aspects of the limbs of a horse at speeds up to a gallop. This IMU system also differs from other lameness detection systems that use gyroscopes and accelerometers in that the full 3-D kinematics (linear and angular positions, velocities, and accelerations) are accessible for analysis. However, the accuracy of this system has not been established by the manufacturer.
A limited number of studies15,17,19 have been conducted with IMU systems in horses. A validation study19 of a human IMU system investigating equine trunk movement found that the error of the IMU sensor, compared with results for an optical kinematic system, was < 7% of the total range of motion in all 3 orthogonal directions during walking, trotting, and cantering. A larger number of studies have been conducted to determine the accuracy of IMU systems for use in evaluating specific human movements. Several studies20–22 revealed a high degree of correlation between IMU and optical systems, with most correlations > 0.90. In other studies,21–23 the RMSE has been used to assess the accuracy of IMU systems. Root mean squared errors ranging from 0.7° to 25.6° have been reported for orientation data. In 1 study22 conducted to evaluate movement of the upper arm, there was a high overall accuracy of the IMU system with positional RMSEs < 1 cm and orientation RMSEs between 2.5° and 5°. On the basis of these previous comparisons in combination with the knowledge that the acceleration at hoof impact is of higher frequency than that of the assessed human motions, any new IMU system must be validated for use in horses.
Therefore, the objective of the study reported here was to compare the accuracy of an equine IMU system with that of a 3-D optical kinematic system, which is the criterion-referenced standard for motion measurement, and to validate the IMU system by examining equine locomotion of the hooves of the right forelimb and hind limb of nonlame horses during walking and trotting. We hypothesized that the IMU system, rigidly attached to a hoof and to optical markers, would provide data as accurate and precise as those reported by other research groups for a 3-D optical kinematic system.
Materials and Methods
Horses—Five clinically normal Quarter Horses (age, 2 to 3 years) with no obvious lameness while walking or trotting were used for the study. All hooves of each horse were trimmed and balanced before the study began. All horses were acclimated to the laboratory facility where data were collected. All procedures were approved by the Institutional Animal Care and Use Committee of Colorado State University.
IMUs and retroreflective markers—Each node of the IMU system measured 63 × 63 × 25 mm with a weight of approximately 80 g (Figure 1). The IMU system was composed of three 3 degree of freedom accelerometers, a 3 degree of freedom rate gyroscope, a 3 degree of freedom magnetometer, and a thermostat; samples were recorded at 500 Hz. The 3 accelerometers were used for each orthogonal axis and had operating ranges of ± 1.7 g, ± 18 g, and ± 125 g, which were selected for the highest possible resolution at slow gaits and moderate- and high-impact gaits. Data were collected simultaneously from all accelerometers for all axes. The signal-processing algorithms then selected the accelerometer for each axis that provided the most accurate estimate of the acceleration, taking into account their maximum range and sensitivity. The gyroscope range could be programmed for rates of rotation up to ± 10,000°/s.
For each trial, 5 IMU nodes were used (1 on the horse's head over the occipital protrusion at the back of the skull [poll] and 1 on each limb). The IMU nodes were attached to the lateral hoof wall of all 4 hooves with acrylic glue.a Three 1.5-cm-diameter, spherical, retroreflective markers were placed on top of the IMU nodes located on the hooves of the right limbs of each horse by use of a custom bracket, which created a triad that moved rigidly with the node. The bracket and triad measured 11 × 12.4 × 0.5 cm and were covered by nonreflective white tape. The weight of the bracket and triad was 102.6 g, so the combined node-triad unit weighed approximately 182.6 g. The bracket was attached to the sensor by fabric hook-and-loop fasteners and further stabilized by two 1/8-inch pieces of umbilical tape, which were incorporated into the glue used to attach the sensor to the hoof wall. The markers were located approximately 8 to 10 cm apart on the triad.
Cameras and collection of kinematic data—Eight infrared camerasb operating at 200 Hz were used to collect 3-D optical kinematic data. The cameras were placed in a semicircular configuration on the right side of each horse for all trials. Four cameras were placed on tripods near the ground, and 4 were placed on elevated overhead positions. Calibration of the optical kinematic systemc for the trials yielded coordinate resolution to within 2 mm.
Trial design and synchronization—For data collection, all horses were led at a walk and a trot over a rubberized runway covering an asphalt surface measuring 1.2 × 24.8 m with an optical capture volume of 3.7 × 1.3 × 2.4 m. The capture volume was located near the middle of the long axis of the runway, which allowed each horse to be at a constant speed at the location of the capture volume. The capture volume contained 2 force platforms imbedded in the runway; these force platforms were not used for the present study. Horses were led at a walk and a trot at a consistent and comfortable speed for each horse. Each horse dictated its own optimal velocity, and the velocity of each trial was determined by 5 infrared timing gatesd spaced at intervals of 1.5 m, which were linked to the optical kinematics computer and triggered by a horse immediately before it entered the capture volume.
Five complete strides were collected from the right forelimbs and hind limbs for each horse during walking and trotting. An anomaly detected by the IMU magnetometer in the data of the stride prior to the stride of interest as the horse approached the imbedded force platforms was used to synchronize the 2 systems and ensure that common strides were compared. This anomaly was a consistent, reproducible peak in the magnetic field, with a strength many magnitudes higher than the earth's magnetic field. Because the magnetometers were not used during this phase of the signal processing, the data output from the system was not affected. Typically, only 1 swing phase/runway pass was analyzed, although occasionally data for both a forelimb and hind limb were collected on the same runway pass. The acceleration curves for the Z direction (acceleration in the proximal-distal vertical direction), velocity curves for the X direction (velocity in the cranial-caudal forward direction), and angular orientation curve for Θ (orientation around the hoof medial-lateral axis) were matched between the IMU and optical systems so that the swing phases of the 2 systems were synchronized with regard to time and started and ended at the same times. The beginning of the swing phase examined in the present study included the beginning of hoof rotation (ie, when the heel started to rotate around the toe before leaving the ground). The end of the swing phase was immediately before hoof contact with the ground (the accelerometers within the inertial sensor revealed a ringing artifact when a hoof contacted the ground).
Optical data filtering and reference frames—Optical data were low-pass filtered at 12 Hz with a recursive fourth-order Butterworth filter. To yield linear and angular kinematics consistent with the IMU system, a local optical frame was necessary. The local origin of each triad was determined by creating a virtual point located at the mean of the cranial and caudal markers placed near the local origin of the IMU node (Figure 1). The local optical reference frame was constructed around the local origin with the markers of the triad. The local optical reference frame was aligned with that of the IMU (Figure 2). The origin of the global optical reference frame was translated to the local origin position at the start of the swing phase with the cranial-caudal axis (x-axis) rotated through the local origin at the end of the swing phase. This kept the z-axis vertical and the y-axis medial-lateral with the horse in motion. The linear position, velocity, and acceleration of the local reference frame origin within the global reference frame were compared with the IMU linear kinematics. Hoof orientation was examined by use of Cardan angles (rotation around the y-axis [Θ], x′-axis [Θ], and z″-axis [Φ]), with rotation of the local optical reference frame about the y-axis determined first, followed by rotation around the new x-axis (x′), and lastly rotation around the new z-axis (z″). This was consistent with the hoof orientations calculated by the equine IMU system.
IMU data processing—The IMU processing began with node synchronization so that all data had a common time base. Because there is little movement when a hoof is on the ground during the stance phase, all values were set to zero. The beginning of the stance phase was detected by identifying a period of high accelerations (hoof impact) followed by low accelerations. The end of the stance phase—beginning of the swing phase was identified by changes in orientation and increased accelerations in the IMUs.e
Raw data for the IMU were collected in a local reference frame of the hoof, with x-, y-, and z-axes identical to those of the optical system. Orientation of an IMU node with respect to each horse was determined when the node was not rotating or moving (ie, during the stance phase). Then, by use of measurements from the magnetometers of the earth's magnetic field and from the accelerometers of the earth's gravitational field, orientation of the sensor was established.e
Custom softwaref was used to determine linear velocities and positions via single and double integration of the linear acceleration data, respectively. The rate gyroscopes provided angular velocities, which were integrated to provide orientations and differentiated to determine angular accelerations.
All IMU data were collected and processed separately by the IMU manufacturer.g Therefore, both parties were not aware of the origin of the data until after comparisons were made. After the IMU data were processed, they were submitted to the authors for synchronization and comparison with the 3-D optical data.
Variables examined—Linear and angular variables were examined in all 3 dimensions. Maximum, minimum, and mean values were extracted for each linear and angular variable (Appendix). In the Z direction (proximal-distal), the position-versus-time curve had 2 peaks, so 2 maxima were extracted for this variable.
Statistical analysis—Linear and angular positions, velocities, and accelerations were compared between systems. Maximum, minimum, and mean values were extracted and compared by use of commercial softwareh via a paired t test or Wilcoxon signed rank tests and via Pearson correlation coefficients, with significance set at values of P < 0.05. All horses, gaits, and hooves were pooled for statistical analysis (total of 70 trials). To determine the appropriate paired test of difference, tests of normality were performed on the differences of the 2 systems; histograms and normal plots were also created from these differences. A paired t test was performed when data appeared to be normally distributed, and the Wilcoxon signed rank test was used for nonnormally distributed data.
Root mean squared errors were calculated for each hoof of each horse at each gait across the entire swing phase to evaluate the overall error between the 2 systems. Because the 3-D optical kinematics system recorded data at 200 Hz and the inertial sensor recorded data at 500 Hz, the 2 systems were compared at a common frequency of 100 Hz. Mean RMSE and SD were calculated for the 5 horses for each gait and hoof (eg, forelimb hoof during trotting). The mean RMSE was compared with the range of values collected for each swing phase. A mean range was then calculated for the 5 horses for each gait and hoof, and the RMSE was then calculated as a percentage of the range. Finally, time-normalized and mean curves were created and overlaid for the 2 systems to visually explore the profiles of each variable.
Results
Twenty-five strides, including both stance and swing phases, were collected for each hoof at each gait. The mean velocity of all trials for all horses during trotting was 2.75 m/s, with a range of 2.4 to 3.2 m/s. The mean velocity of all trials for all horses during walking was 1.3 m/s, with a range of 1.1 to 1.5 m/s. Individual horses typically moved at a consistent speed throughout the trials for each gait; the velocities of all trials were within 10% of the individual mean for each gait, and 53 of 70 (75%) were within 5% of the mean.
Because data for the stance phase were set to zero for IMU processing, only swing phase data were analyzed. Fifteen trials for the right forelimb hoof during trotting, 14 trials for the right hind limb hoof during trotting, 22 trials for the right forelimb hoof during walking, and 19 trials for the right hind limb hoof during walking yielded common data sets that were complete for both linear and angular kinematics from both systems.
Overall, results for the linear and angular variables determined by the IMU correlated well with results for the 3-D optical kinematic system (Table 1). In the cranial-caudal direction (X direction), 6 of 9 extracted values were highly correlated (r > 0.8), and 2 of 9 were moderately correlated (r > 0.5). The minimum position, which was close to zero for both systems (the translated origin), was the only variable that was weakly correlated (r < 0.25). Six of 9 extracted x-axis values were significantly different between the 2 systems, as determined via the paired test of differences. Only 3 extracted values (mean position, minimum acceleration, and mean acceleration) were not significantly different between the 2 systems.
Pearson correlation coefficients with their associated P values and results for paired test of differences (t test or Wilcoxon signed rank test) for the X direction (cranial-caudal), Y direction (medial-lateral), and Z direction (proximal-distal) between the 3-D optical kinematics and IMU systems in horses during walking and trotting.
Variable | Value | X direction | Y direction | Z direction | ||||||
---|---|---|---|---|---|---|---|---|---|---|
r | P value | P valuefor t test or Wilcoxon signed rank test* | r | P value | P value for t test or Wilcoxon signed rank test* | r | P value | P valuefor t test or Wilcoxon signed rank test* | ||
Position | Maximum 1 | 0.927 | < 0.001 | < 0.001† | 0.832 | < 0.001 | 0.247† | 0.884 | < 0.001 | < 0.001 |
Maximum 2 | NA | NA | NA | NA | NA | NA | 0.508 | < 0.001 | 0.255 | |
Minimum | 0.120 | 0.321 | < 0.001 | 0.838 | < 0.001 | 0.218† | 0.084 | 0.492 | 0.107 | |
Mean | 0.928 | < 0.001 | 0.896 | 0.784 | < 0.001 | 0.825† | 0.634 | < 0.001 | 0.014† | |
Velocity | Maximum | 0.960 | < 0.001 | < 0.001 | 0.911 | < 0.001 | 0.005† | 0.873 | < 0.001 | 0.938 |
Minimum | −0.556 | < 0.001 | < 0.001† | 0.867 | < 0.001 | < 0.001† | 0.850 | < 0.001 | 0.413† | |
Mean | 0.961 | < 0.001 | 0.003 | −0.868 | < 0.001 | 0.036† | −0.334 | 0.005 | < 0.001† | |
Acceleration | Maximum | 0.894 | < 0.001 | < 0.001 | 0.644 | < 0.001 | < 0.001 | 0.757 | < 0.001 | < 0.001 |
Minimum | 0.943 | < 0.001 | 0.296† | 0.906 | < 0.001 | < 0.001 | 0.825 | < 0.001 | 0.922† | |
Mean | 0.669 | < 0.001 | 0.979 | 0.899 | < 0.001 | 0.938† | 0.481 | < 0.001 | 0.001† | |
Angular orientation | Maximum | 0.880 | < 0.001 | < 0.001† | 0.314 | 0.008 | < 0.001† | 0.850 | < 0.001 | 0.450† |
Minimum | 0.456 | < 0.001 | 0.601† | 0.693 | < 0.001 | 0.052† | 0.423 | < 0.001 | < 0.001† | |
Mean | 0.909 | < 0.001 | 0.0169† | 0.450 | < 0.001 | < 0.001† | 0.724 | < 0.001 | 0.002† | |
Angular velocity | Maximum | 0.969 | < 0.001 | < 0.001† | 0.408 | < 0.001 | < 0.001† | 0.816 | < 0.001 | < 0.001 |
Minimum | 0.908 | < 0.001 | < 0.001 | 0.770 | < 0.001 | < 0.001 | 0.719 | < 0.001 | < 0.001† | |
Mean | 0.582 | < 0.001 | 0.219† | 0.396 | < 0.001 | < 0.001† | 0.485 | < 0.001 | 0.314† | |
Angular acceleration | Maximum | 0.664 | < 0.001 | < 0.001 | 0.697 | < 0.001 | < 0.001 | 0.774 | < 0.001 | < 0.001 |
Minimum | 0.764 | < 0.001 | < 0.001 | 0.801 | < 0.001 | < 0.001 | 0.826 | < 0.001 | < 0.001 | |
Mean | 0.705 | < 0.001 | 0.054† | 0.496 | < 0.001 | 0.315† | 0.670 | < 0.001 | < 0.001† |
Correlations were considered high (r > 0.8), moderate (r > 0.5 to ≤ 0.8), or mild (r > 0.25 to ≤ 0.5).
Values differed significantly (P < 0.05) between the 2 kinematic systems.
A paired t test was performed.
In the medial-lateral direction (Y direction), 7 of 9 extracted values were highly correlated, and the remaining 2 were moderately correlated. Five of 9 y-axis values were significantly different between the 2 systems. The 4 extracted values that were not significantly different between the 2 systems were maximum, minimum, and mean position and mean acceleration.
In the proximal-distal direction (Z direction), 4 of 10 extracted values were highly correlated, 3 were moderately correlated, 2 were mildly correlated (r > 0.25), and 1 (ie, minimum position) was weakly correlated. Five of the 10 values were significantly different between the 2 systems. The 5 values that were not significantly different between the 2 systems were minimum position, second maximum position, maximum and minimum velocities, and minimum acceleration. Similar to results for the minimum position of the X direction, the minimum position of the Z direction was close to zero for both systems, although in the Z direction, it could occur at the beginning or end of the swing phase.
For the angular variables, Θ (ie, rotation around the medial-lateral axis) appeared to have the highest correlation between the 2 systems. Four of 9 extracted values were highly correlated, 4 of 9 were moderately correlated, and 1 of 9 was mildly correlated. Six of 9 values were significantly different between the 2 systems. The 3 values that were not significantly different between the 2 systems were minimum angular orientation, mean angular velocity, and mean angular acceleration.
For Θ, 1 of 9 values was highly correlated, 3 were moderately correlated, and 5 were mildly correlated. Seven of 9 values were significantly different between the 2 systems. The 2 values that were not significantly different between the 2 systems were minimum angular orientation and mean angular acceleration.
For ψ, 3 of 9 values were highly correlated, 4 were moderately correlated, and 2 were mildly correlated. Seven values were significantly different between the 2 systems. Only 2 values (ie, maximum angular orientation and mean velocity) were not significantly different between the 2 systems.
Root mean squared errors, as a percentage of the mean of the range for each variable, were similar between gaits (Table 2). Linear displacement in the cranial-caudal direction and Θ (ie, rotation around the medial-lateral axis) appeared to have the least error (1.06% to 2.89% and 2.60% to 3.46%, respectively). The RMSE percentages for position and acceleration in the Y-direction (medial-lateral) typically were higher than those for both the X direction (cranial-caudal) and Z direction (proximal-distal). The range of RMSE percentages for position, velocity, and acceleration in the X direction were 1.06% to 2.89%, 3.84% to 6.11%, and 7.18% to 11.79%, respectively. The range of RMSE percentages for position, velocity, and acceleration in the Y direction were 12.42% to 19.23%, 9.39% to 18.72%, and 20.69% to 26.96%, respectively. The range of RMSE percentages for position, velocity, and acceleration in the Z direction were 10.13% to 17.70%, 8.38% to 11.41%, and 12.54% to 20.16%, respectively. For the angular variables, the RMSE percentages were higher for Θ and Φ than for Θ. The range of RMSE percentages for position, velocity, and acceleration of Θ were 2.60% to 3.46%, 4.34% to 6.68%, and 9.50% to 22.27%, respectively. The range of RMSE percentages for position, velocity, and acceleration of Θ were 16.16% to 46.06%, 15.16% to 23.58%, and 22.61% to 28.95%, respectively. The range of RMSE percentages for position, velocity, and acceleration of Φ were 17.22% to 31.74%, 19.49% to 22.51%, and 22.62% to 33.31%, respectively.
Mean ± SD RMSE, mean range, and RMSE as a percentage of the mean range for each gait and hoof for the X direction (cranial-caudal), Y direction (medial-lateral), and Z direction (proximal-distal) during the swing phase for horses during walking and trotting.
Variable | Speed and limb | X direction | Y direction | Z direction | ||||||
---|---|---|---|---|---|---|---|---|---|---|
RMSE mean ± SD | Range of mean | RMSE as % of range | RMSE mean ± SD | Range of mean | RMSE as % of range | RMSE mean ± SD | Range of mean | RMSE as % of range | ||
Position (m) | Trotting and forelimb | 0.029 ± 0.015 | 1.88 | 1.57 | 0.009 ± 0.005 | 0.04 | 19.23 | 0.008 ± 0.006 | 0.08 | 10.13 |
Trotting and hind limb | 0.025 ± 0.015 | 1.89 | 1.33 | 0.012 ± 0.007 | 0.10 | 12.42 | 0.012 ± 0.010 | 0.07 | 17.70 | |
Walking and forelimb | 0.044 ± 0.039 | 1.53 | 2.89 | 0.009 ± 0.007 | 0.05 | 18.57 | 0.006 ± 0.003 | 0.05 | 10.40 | |
Walking and hind limb | 0.016 ± 0.009 | 1.53 | 1.06 | 0.007 ± 0.003 | 0.04 | 16.69 | 0.008 ± 0.005 | 0.06 | 12.91 | |
Velocity (m/s) | Trotting and fore limb | 0.410 ± 0.194 | 6.71 | 6.11 | 0.193 ± 0.062 | 1.03 | 18.72 | 0.211 ± 0.088 | 2.11 | 9.96 |
Trotting and hind limb | 0.256 ± 0.070 | 6.45 | 3.97 | 0.169 ± 0.075 | 1.80 | 9.39 | 0.200 ± 0.123 | 1.97 | 10.14 | |
Walking and forelimb | 0.259 ± 0.109 | 4.37 | 5.94 | 0.122 ± 0.073 | 0.76 | 16.03 | 0.093 ± 0.037 | 1.11 | 8.38 | |
Walking and hind limb | 0.190 ± 0.093 | 4.94 | 3.84 | 0.110 ± 0.037 | 0.77 | 14.29 | 0.115 ± 0.060 | 1.01 | 11.41 | |
Acceleration (m/s/s) | Trotting and forelimb | 17.44 ± 8.97 | 197.59 | 8.83 | 12.74 ± 5.39 | 47.25 | 26.96 | 20.55 ± 8.68 | 101.95 | 20.16 |
Trotting and hind limb | 11.80 ± 3.53 | 100.14 | 11.79 | 8.04 ± 2.13 | 38.86 | 20.69 | 9.38 ± 3.16 | 74.83 | 12.54 | |
Walking and forelimb | 7.11 ± 1.74 | 99.11 | 7.18 | 3.94 ± 2.08 | 18.68 | 21.09 | 4.49 ± 1.64 | 32.89 | 13.67 | |
Walking and hind limb | 5.85 ± 2.32 | 78.44 | 7.46 | 4.03 ± 1.41 | 18.97 | 21.25 | 4.30 ± 1.93 | 26.65 | 16.14 | |
Angular orientation (°) | Trotting and forelimb | 2.92 ± 1.28 | 108.98 | 2.68 | 3.79 ± 1.43 | 22.94 | 16.51 | 3.36 ± 1.21 | 19.54 | 17.22 |
Trotting and hind limb | 2.52 ± 1.25 | 96.72 | 2.60 | 3.44 ± 2.21 | 21.30 | 16.16 | 5.20 ± 2.48 | 21.19 | 24.55 | |
Walking and forelimb | 3.42 ± 1.78 | 98.75 | 3.46 | 5.77 ± 3.08 | 13.90 | 41.49 | 4.01 ± 2.34 | 17.44 | 23.03 | |
Walking and hind limb | 2.80 ± 1.90 | 88.75 | 3.16 | 5.03 ± 2.48 | 10.92 | 46.06 | 4.77 ± 2.23 | 15.05 | 31.74 | |
Angular velocity (°/s) | Trotting and forelimb | 121.80 ± 43.40 | 2,531.22 | 4.81 | 121.20 ± 30.84 | 799.40 | 15.16 | 129.70 ± 51.22 | 609.63 | 21.28 |
Trotting and hind limb | 136.80 ± 62.89 | 2,046.47 | 6.68 | 97.18 ± 70.71 | 580.57 | 16.74 | 112.27 ± 51.03 | 512.38 | 21.91 | |
Walking and forelimb | 83.31 ± 39.21 | 1,921.19 | 4.34 | 96.90 ± 49.15 | 410.87 | 23.58 | 85.23 ± 47.82 | 437.23 | 19.49 | |
Walking and hind limb | 74.10 ± 47.81 | 1,560.09 | 4.75 | 97.24 ± 56.81 | 420.19 | 23.14 | 82.26 ± 40.87 | 365.53 | 22.51 | |
Angular acceleration (°/s/s) | Trotting and forelimb | 13,297 ± 4,975 | 77,451 | 17.17 | 13,150 ± 4,039 | 540,202 | 24.34 | 11,383 ± 3,419 | 34,172 | 33.31 |
Trotting and hind limb | 15,403 ± 8,142 | 69,169 | 22.27 | 8,097 ± 5,824 | 27,964 | 28.95 | 7,606 ± 3,445 | 28,364 | 26.82 | |
Walking and forelimb | 6,142 ± 2,602 | 64,652 | 9.50 | 5,479 ± 2,437 | 21,843 | 25.08 | 4,836 ± 2,699 | 21,383 | 22.62 | |
Walking and hind limb | 5,295 ± 3,028 | 50,820 | 10.42 | 5,657 ± 3,735 | 25,017 | 22.61 | 4,052 ± 2,423 | 17,853 | 22.70 |
The appearance of the time-normalized and mean overlay plots was extremely similar between the 2 systems, both gaits, and both limbs (Figures 3–8). For the linear variables, positional and velocity data in all 3 directions had similar patterns. The accelerations in all 3 directions had a similar appearance, but the IMU curves contained more fluctuations and higher frequencies. In addition, the IMU curves had larger magnitude peaks and troughs for accelerations in the Y and Z directions in the hoof of the right forelimb during trotting.
For the orientation variables for the IMU system, Θ and Φ were fairly similar in shape, compared with the optical system, whereas Θ typically had higher values. For angular velocity, Θ had similar patterns between the 2 systems. Angular velocities for the IMU system for Θ and Φ typically had higher magnitude peaks near the beginning of the swing phase at both gaits. For all 3 angular accelerations, there were similarities in appearance of the curves, but there were more fluctuations in the IMU, compared with the optical curves (Figure 8).
Discussion
In the present study, we compared a novel equine IMU system with a commercially available 3-D optical system. The objective of the study was to determine whether the equine IMU system would provide similar data to those of the criterion-referenced standard of kinematics via the commercially available 3-D optical system. The swing phase was examined because during IMU processing, data in the stance phase were set to zero. In addition, there is a larger range of excursion in the linear and angular variables examined during the swing phase. Clinically, the stance phase, especially initial hoof impact and planting of the hoof, is more important in the pathogenesis of injury than is the swing phase.18 The ringing artifact from the accelerometer in the IMU system at the end of the swing phase precluded the ability to examine hoof impact and planting of the hoof. The source of the ringing artifact is unclear and has not been previously observed by the manufacturer.e It is speculated that this ringing may be caused by a combination of factors, including rigid attachment of the IMU node to the hoof (compared with a boot attachment method), a relatively hard surface (compared with dirt), and the additional mass of the marker triad. Additional research to evaluate surface and attachment method would need to be performed to determine the source of the ringing. Even though hoof impact and planting of the hoof could not be examined, breakover at the end of the stance phase could be evaluated, as indicated in the present study, and this may be an additional phase of the stride important for evaluation. Considering that the IMU system has a high-frequency event at toe-off, which was not evident in data for the optical system, the IMU system may be preferable to the optical system to examine this phase of the stride.
Temporal components of the stride and duty factor (ie, percentage of the stride when the limb is weight bearing) of each limb are variables that could be measured by the IMU system and could be important in lameness evaluation. Because the swing phase comparisons were closely matched on the basis of time, these components were not statistically evaluated between the 2 systems. From cursory examination of the data, the swing phase times appeared to be extremely similar between the 2 systems.
In general, the IMU system was correlated fairly well with the 3-D optical kinematic system, as indicated by the high number of moderate to high correlations (Table 1). There were 25 of 55 (45%) extracted values with high correlations (r > 0.8) and 18 of 55 (33%) values with moderate correlations (r > 0.5). As expected, several variables were not highly correlated. These variables included minimums in the X and Z direction, which were clustered around zero and were not expected to be correlated. The IMU and 3-D optical systems did not always provide exactly the same values for the linear and angular variables examined, which was indicated by the paired tests of differences and RMSEs. However, the overall shape of each variable versus percentage time curve was extremely similar.
The same handler assisted for all horses, trials, and gaits. This handler allowed each horse to move at a comfortable speed chosen by each horse. This approach was used to mimic a clinical setting. In addition, because all comparisons between the IMU and optical systems were made on the basis of individual trial and the mean was subsequently calculated for each horse and gait, differences in individual trial velocities did not hinder comparison between the 2 systems. Also, in determining correlations, it is useful to have a range of data, which was accomplished by combining walking and trotting data and enhanced by the small amount of variability in gait velocity.
A few studies in humans have included Pearson correlation coefficients for comparison of IMU to 3-D optical systems. A study21 on human vertebral column posture in the X and Y directions revealed correlations > 0.77 for thoracic movements and > 0.97 for lumbar movements. Another study22 in which upper limb movements in humans was evaluated revealed correlations > 0.96 for linear position in the X, Y, and Z directions and > 0.94 for orientation. A third study20 conducted to evaluate the stride, step, and stance duration in humans during running at various velocities revealed correlations > 0.76, with the majority (10/12 variables) of correlations > 0.90. Overall, these studies, in which several IMU units were compared with optical systems, revealed higher correlations in both linear positional and angular orientation values than were determined for the equine IMU system used in the present study. Overall, the equine IMU system used in the present study only had 10 of 55 (18%) values of r > 0.90, and when examining linear positions and angular orientations, there were only 3 of 19 variables with r > 0.90 (Table 1). A small number of the variables examined would not be expected to have high correlations because they are expected to cluster around zero. From this correlation data, it is not clear whether the equine IMU system at this stage of development would perform adequately for use in clinical settings.
The IMU system collected linear data with accelerations, and these variables were integrated to calculate velocities and positions, whereas the 3-D optical system began with positional data and differentiated these variables to calculate velocities and accelerations. Because of this difference in calculation of variables, examining linear positions and accelerations allowed the evaluation of both ends of the calculation process. The RMSEs for both position and acceleration in the X direction (cranial-caudal) were smaller than for the Y and Z directions, and overall, the RMSE percentages for position were slightly to moderately better than the RMSE percentages for acceleration. Given that the IMU system starts with linear acceleration, it is likely that some of the errors in the acceleration values originated from the 3-D optical system. Because the calculation of acceleration involves 2 differentiations from the initial positional data, any errors when determining velocities are compounded in the calculation of acceleration. Also, the 2 systems differ in frequency content, which may also lead to the differences in acceleration. Because the 3-D optical system collected data at a lower frequency (200 Hz) and was then low-pass filtered, this system yielded a smoother profile and lower magnitude maximum and minimum velocities and accelerations (Figures 4 and 5). Although this may reduce the accuracy of the linear acceleration data within the optical system, it does provide for a more stable output. The higher-frequency content of the IMU system, although more sensitive to high-frequency fluctuations, may make it more difficult to use acceleration data clinically. However, further evaluation of the higher-frequency IMU system for both detection and quantification of lameness, whether clinical or experimentally induced, is warranted.
The angular variables were calculated from linear positions in the 3-D optical system and integrated and differentiated from angular velocities in the IMU system. Analysis of the graphs revealed that Θ and Φ have the same general shape, but there was high variability leading to higher RMSE percentages for the Θ and Φ variables, compared with the variability for the Θ variables. It is also expected that Θ and Φ would have higher error than Θ because the order of calculation of these 3 variables starts with Θ rotation, then Θ, and finally Φ. Any errors for Θ would also be added to errors in Θ, and both of those errors would be compounded in Φ. In the human IMU literature, higher error rates have been reported in both Θ and Φ than in Θ.24
In general, the variables that performed best in the RMSE evaluations were those that had larger ranges (specifically the position for the X direction and Θ, generally in the sagittal plane). This is a finding similar to that in a report22 of a human IMU system in which there was less accuracy in movements with small ranges of motion (< 2° or < 0.5 cm). When comparing mean RMSEs among variables with the same limb and gait, there were few consistent results. However, when the mean RMSE was converted into a percentage of its range, the variables with a larger range had lower errors. It is possible that if a larger range of motion were performed for the Y, Z, Θ, and Φ variables, the performance of these variables would improve.
A large range of RMSEs for both position and orientation have been reported for a variety of movements in humans20–23 as well as movement of the trunk in horses.19Analysis of data for the study reported here indicated positional RMSEs within 5 cm in all 3 orthogonal directions, with larger RMSEs in the X direction, where there is a larger range of motion (Table 2). Most RMSEs in the Y and Z directions were < 1 cm. These Y and Z values corresponded fairly well (positional errors < 1 cm) with those in a previous report.22 In another study,19 investigators reported errors as a percentage of the range, with positional values < 6.5% during walking and < 4.3% during trotting for all 3 directions. In general, RMSE positional values in the present study were higher (< 18.6% during walking and < 19.3% during trotting). Investigators in that study19 also found a trend in that the Y direction had higher RMSEs than did the X or Z directions, which is similar to results of the present study. For orientation data, investigators in another study23 found a large variability in RMSEs (depending on movement), with mean RMSEs ranging from 0.7° to 25.6°. In the present study, we found similar but higher results, with orientation RMSEs ranging from 2.6° to 46.06°.
The ultimate clinical goal of a motion-sensing system, such as an IMU system, would be accurate detection and evaluation of lameness. Asymmetry between the lame and nonlame limbs in vertical displacement of lameness during the swing phase has been reported in a small number of horses.25 Thus, accuracy in the Z direction (proximal-distal) is necessary to appreciate those asymmetries. This equine IMU system performs moderately well in the Z direction (moderate to high correlations, but moderately high RMSEs). Further comparison of this IMU system with the optical system would be needed to determine whether the IMU system provides sufficient accuracy in the Z direction. In addition, comparing the vertical displacements measured by the IMU system in a lame versus a nonlame limb may provide additional information about the ability of this system to identify asymmetry.
In addition, accelerations of the hoof in the X and Z directions have been measured previously for use in evaluation of ground surfaces because the accelerations at hoof impact may be important in the development of injury.26 In the present study, acceleration in the X and Z directions had moderate to high correlations and had similar appearance in the percentage-versus-time graphs (Figure 8) but had RMSEs (percentage of range) that were fairly high (7.18% to 11.79% and 12.54% to 20.16%, for X and Z, respectively). However, considering that there was a ringing artifact of unknown origin for the accelerometers at hoof impact, it is unclear whether this equine IMU system is adequate for use in examination of the impact phase of the stride. Given that hoof impact is an important phase of the stride for evaluation, the lack of ability of this equine IMU system to evaluate this phase of the stride would severely hinder its clinical usefulness. Determination of the source and elimination of the ringing artifact would be necessary to determine whether the IMU system would be useful in the examination of hoof impact.
Measurement of 3-D rotations of joints has been reported in clinically normal horses and may also be important in evaluating lameness.27 Thus, if IMU systems are to be used, it is important that they provide accurate measurement of dynamic orientation. The equine IMU system examined in the present study appeared to provide adequate accuracy for the Θ rotation but may not be adequate in the Θ and Φ rotations (larger RMSEs and lower correlations). The limited accuracy in measurement of orientation in all 3 rotations may reduce the clinical usefulness of this equine IMU system; however, this hypothesis would need to be tested in horses with experimentally induced lameness.
Overall, the IMU system attached rigidly to a hoof provided similar data with patterns similar to those for the criterion-referenced standard 3-D optical system during the swing phase of walking and trotting horses. Further studies would need to be performed to examine faster gaits (canter and gallop). Although the 2 systems did not provide identical data, overall, there was moderate to high correlation between most variables determined by the 2 systems during the swing phase of the stride. Considering that there was more error between the 2 systems when the variable went through a small range of motion, it would be important to compare these 2 systems during the stance phase of the stride, when the hoof has small ranges of motion. In addition, if the IMU nodes are attached by a nonrigid method to a hoof, such as with elastic bandage material or a boot, there may be different results because of motion of the node relative to the hoof. It would be important to examine attachment methods other than acrylic glue because a quick and easy method of IMU node attachment would be desirable for clinical use of this system. Additionally, because there was a ringing artifact during a key phase of the stride (hoof impact), further investigation of the source of the artifact would be necessary before this IMU system could be recommended for clinical use. One marketing point of this product would undoubtedly be its clinical use for lameness detection; thus, it would be necessary to determine the diagnostic utility of this IMU system for various degrees of lameness in horses with experimentally induced lameness or in clinical cases.
ABBREVIATIONS
IMU | Inertial measurement unit |
RMSE | Root mean squared error |
Superfast glue, Vettec Hoof Care Products, Oxnard, Calif.
Volant, Peak Performance Technologies Inc, Centennial, Colo.
Vicon-Motus 9.2, Vicon Motion Systems Inc, Centennial, Colo.
MEK 92-PAD photoelectric control, Mekontrol Inc, Northboro, Mass.
Davies M, EquuSys Inc, Sudbury, Mass: Personal communication, 2011.
MATLAB, Mathworks, Natick, Mass.
Equusense Ultra, EquuSys Inc, Sudbury, Mass.
SAS, version 9.2, SAS Institute Inc, Cary, NC.
References
1 USDA. National economic cost of equine lameness, colic, and equine protozoal myeloencephalitis in the United States. Information sheet. Fort Collins, Colo: USDA APHIS Veterinary Services National Health Monitoring System, 2001.
2 Fuller CJ, Bladon BM, Driver AJ, et al. The intra- and inter-assessor reliability of measurement of functional outcome by lameness scoring in horses. Vet J 2006; 171: 281–286.
3 Hewetson M, Christley RM, Hunt ID, et al. Investigations of the reliability of observational gait analysis for the assessment of lameness in horses. Vet Rec 2006; 158: 852–857.
4 Keegan KG. Evidence-based lameness detection and quantification. Vet Clin North Am Equine Pract 2007; 23: 403–423.
5 American Association of Equine Practitioners. Guide for veterinary service and judging of equestrian events. 4th ed. Lexington, Ky: American Association of Equine Practitioners, 1991;19.
6 Keegan KG, Dent EV, Wilson DA, et al. Repeatability of subjective evaluation of lameness in horses. Equine Vet J 2010; 42: 92–97.
7 Ishihara A, Bertone AL, Rajala-Schultz PJ. Association between subjective lameness grade and kinetic gait parameters in horses with experimentally induced forelimb lameness. Am J Vet Res 2005; 66: 1805–1815.
8 Buchner HHF, Savelberg HHCM, Schamhardt HC, et al. Limb movement adaptations in horses with experimentally induced fore- or hindlimb lameness. Equine Vet J 1996; 28: 63–70.
9 Keegan KG, Wilson DA, Smith BK, et al. Changes in kinematic variables observed during pressure-induced forelimb lameness in adult horses trotting on a treadmill. Am J Vet Res 2000; 61: 612–619.
10 Weishaupt MA, Hogg HP, Wiestner T, et al. Instrumented treadmill for measuring vertical ground reaction forces in horses. Am J Vet Res 2002; 63: 520–527.
11 Robin D, Chateau H, Pacquet L, et al. Use of a 3D dynamometric horseshoe to assess the effects of an all-weather waxed track and a crushed sand track at high speed trot: preliminary study. Equine Vet J 2009; 41: 253–256.
12 Chateau H, Robin D, Simonelli T, et al. Design and validation of a dynamometric horseshoe for the measurement of three-dimensional ground reaction force on a moving horse. J Biomech 2009; 42: 336–340.
13 Keegan KG, Yonezawa Y, Pai PF, et al. Evaluation of a sensor-based system of motion analysis for detection and quantification of forelimb and hind limb lameness in horses. Am J Vet Res 2004; 65: 665–670.
14 Parsons KJ, Pfau T, Wilson AM. High-speed gallop locomotion in the Thoroughbred racehorse. I. The effect of incline on stride parameters. J Exp Biol 2008; 211: 935–944.
15 Pfau T, Robilliard JJ, Weller R, et al. Assessment of mild hindlimb lameness during over ground locomotion using linear discriminant analysis of inertial sensor data. Equine Vet J 2007; 39: 407–413.
16 Witte TH, Hirst CV, Wilson AM. Effect of speed on stride parameters in racehorses at gallop in field conditions. J Exp Biol 2006; 209: 4389–4397.
17 Pfau T, Ferrari M, Parsons K, et al. A hidden Markov model-based stride segmentation technique applied to equine inertial sensor trunk movement data. J Biomech 2008; 41: 216–220.
18 Thomason JJ, Peterson ML. Biomechanical and mechanical investigations of the hoof-track interface in racing horses. Vet Clin North Am Equine Pract 2008; 24: 53–77.
19 Pfau T, Witte TH, Wilson AM. A method for deriving displacement data during cyclical movement using an inertial sensor. J Exp Biol 2005; 208: 2503–2514.
20 Lee JB, Mellifont RB, Burkett BJ. The use of a single inertial sensor to identify stride, step, and stance durations of running gait. J Sci Med Sport 2010; 13: 270–273.
21 Wong WY, Wong MS. Trunk posture monitoring with inertial sensors. Eur Spine J 2008; 17: 743–753.
22 Zhou H, Stone T, Hu H, et al. Use of multiple wearable inertial sensors in upper limb motion tracking. Med Eng Phys 2008; 30: 123–133.
23 Godwin A, Agnew M, Stevenson J. Accuracy of inertial motion sensors in static, quasistatic, and complex dynamic motion. J Biomech Eng 2009; 131: 114501.
24 O'Donovan KJ, Kamnick R, O'Keeffe DT, et al. An inertial and magnetic sensor based technique for joint angle measurement. J Biomech 2007; 40: 2604–2611.
25 Pourcelot P, Audigie F, Degueurce C, et al. Kinematic symmetry index: a method for quantifying the horse locomotion symmetry using kinematic data. Vet Res 1997; 28: 525–538.
26 Setterbo JJ, Garcia TC, Campbell IP, et al. Hoof accelerations and ground reaction forces of Thoroughbred racehorses measured on dirt, synthetic, and turf track surfaces. Am J Vet Res 2009; 70: 1220–1229.
27 Clayton HM, Sha DH, Stick JA, et al. 3D kinematics of the interphalangeal joints in the forelimb of walking and trotting horses. Vet Comp Orthop Traumatol 2007; 20: 1–7.
Appendix
Definitions of maximum and minimum linear and angular variables used for comparison of IMU and 3-D optical capture systems for analysis of horses during walking and trotting.
Variable | Maximum | Minimum |
---|---|---|
X direction position | Displacement of the hoof cranially (ie, swing length) | Most caudal position relative to start of the swing phase |
X direction velocity | Peak cranial velocity of the hoof | Lowest cranial velocity of the hoof (caudal velocity if negative) |
X direction acceleration | Peak cranial acceleration of the hoof | Lowest acceleration of the hoof in the cranial direction (caudal direction if negative) |
Y direction position | Peak displacement of the hoof medially relative to the start of the swing phase | Most lateral displacement of the hoof relative to the start of the swing phase |
Y direction velocity | Peak velocity of the hoof medially | Lowest velocity of the hoof medially (laterally if negative) |
Y direction acceleration | Peak acceleration of the hoof medially | Lowest acceleration of the hoof medially (laterally if negative) |
Z direction position | Maximum 1: First proximally vertical peak displacement of the hoof relative to start of the swing phase | Lowest displacement of the hoof vertically relative to start of the swing phase (below start of the swing phase if negative) |
Maximum 2: Second proximally vertical peak displacement of the hoof relative to start of the swing phase | ||
Z direction velocity | Peak proximally vertical velocity of the hoof | Lowest proximally vertical velocity of the hoof (distally if negative) |
Z direction acceleration | Peak proximally vertical acceleration of the hoof | Lowest proximally vertical acceleration of the hoof (distally if negative) |
Orientation for Θ | Peak counterclockwise angle of the hoof about the y-axis (ie, toe down) | Lowest counterclockwise angle of the hoof about the y-axis (toe up if negative) |
Angular velocity for Θ | Peak counterclockwise velocity of the hoof about the y-axis (ie, toe down) | Lowest counterclockwise velocity of the hoof about the y-axis (toe up if negative) |
Angular acceleration for Θ | Peak counterclockwise acceleration of the hoof about the y-axis (ie, toe down) | Lowest counterclockwise acceleration of the hoof about the y-axis (toe up if negative) |
Orientation for Θ | Peak counterclockwise angle of the hoof about the x′-axis (ie, medial edge elevated relative to lateral edge of hoof) | Lowest counterclockwise angle of the hoof about the x′-axis (lateral edge elevated relative to medial edge of hoof if negative) |
Angular velocity for Θ | Peak counterclockwise velocity of the hoof about the x′-axis (ie, medial edge elevated) | Lowest counterclockwise velocity of the hoof about the x′-axis (lateral edge elevated if negative) |
Angular acceleration for Θ | Peak counterclockwise acceleration of the hoof about the x′-axis (ie, medial edge elevated) | Lowest counterclockwise acceleration of the hoof about the x′-axis (lateral edge elevated if negative) |
Orientation for Φ | Peak counterclockwise angle of the hoof about the z″-axis (ie, toe in) | Lowest counterclockwise angle of the hoof about the z″-axis (toe out if negative) |
Angular velocity for Φ | Peak counterclockwise velocity of the hoof about the z″-axis (ie, toe in) | Lowest counterclockwise velocity of the hoof about the z″-axis (toe out if negative) |
Angular acceleration for Φ | Peak counterclockwise acceleration of the hoof about the z″-axis (ie, toe in) | Lowest counterclockwise acceleration of the hoof about the z″-axis (toe out if negative) |