Force and pressure platforms (FPP).

Force, pressure, and weight—along with their derivatives—are the main kinetic attributes measured by floor-sensitive plates. These measures are taken either when an animal is walking over the plates (dynamic measurements) or when standing on them (static measurements). Measuring plates were the most widely used technology (48 studies, 58.5%) in bovine biomechanical analysis. In the current review, these technologies are classified into three main groups: force platforms (FP), pressure mapping systems (PMS), and weight distribution platforms (WDP). More information on these three main kinetic technologies can be found in Table 4.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Overview of studies included in the review that used force and pressure platforms (FPP).

https://doi.org/10.1371/journal.pone.0266287.t004

Force platforms (FP). Force platforms use one of several different types of transducers (aka load cells) which are installed under a cover plate to measure ground reaction forces (GRF) applied over the platform’s top surface. There were 18 bovine gait analysis studies that have used different force platform types, and these vary in the type and number of force transducers. The most common FP among the studies were 1D (single-axis) plates that can only measure the vertical forces. Alternatively, 4 studies included 3D force plates that can provide information on the three-dimensional components of force (vertical, transverse and longitudinal). One study also used a thin electromechanical film (Emfit) to measure the force generated as a result of a change in the film thickness [31]. Additionally, a treadmill-integrated force measuring system equipped with 18 force transducers (9 on each side) was employed in one study to measure GRF while cows were walking at a controlled speed [27].

Most studies implemented a multiple-force-plate layout, usually a double setup with parallel orientation, to capture bi-lateral and/or individual footfall measurements of force. Three studies, however, used a single force plate system, which allowed for the evaluation of only one side of the animal (i.e., the fore and hind limbs of one side) during a measurement. The length of the FP varied between the studies that have reported platform dimensions, typically ranging from 0.9 to 2 m. An exception to this was the application of a 3-meter-long force platform system in a study that placed 5 smaller plates (each 0.6 m long) in a row to measure multiple consecutive footfalls [41]. Force data recording rate also had considerable variation among the studies that reported it, ranging from 50 Hz to 2000 Hz. However, the usual recording frequency for most of the studies ranged from 100 to 250 Hz.

Force platform variables: Dynamic measurements recorded while animals walk over the plates are the primary way of using force platforms. However, measurements recorded while the animal is standing have also been used in two studies; one study used a force plate system comprised of four plates installed in a hoof-trimming chute [48], and the other used a combined system of force and pressure plates [21]. Peak and average ground reaction forces (GRF) were the most important and frequent variables obtained from FP. The GRF measurements have been primarily measured vertically, except in the case of 3D platforms, which can supplement the vertical inputs with the transvers and the longitudinal inputs. In addition to GRF, other derivatives of force with respect to time (e.g., GRFω and impulse) have been measured. Centre of pressure (COP) and duty-factor (ratio of stance time and stride time) are other variables that have been measured using 3D force platforms. COP was measured to identify stance phases [51] or to detect left and right limbs [41]. Stance time is the most frequent temporal variable that has been measured by most of the FP. More temporal variables such as swing time, and stride time were provided when a five-platform layout was used [41].

The studies that used Reaction Force Detection (RFD) systems [16]—Stepmetrix and the redesigned 3D version of the RFD system—usually presented their measurements as Limb Movement Variables (LMV), which are typically a collection of the following variables: peak GRF, average GRF, stance time, impulse, and GRFω. The Stepmetrix machine, as a lameness detection system, provides an automated locomotion scoring system (Stepmetrix locomotion scoring, SLS) using the LMVs. The SLS ranges from 1 to 100, and has been only reported in one study [26]. Asymmetry measurements of some aforementioned variables such as GRF, stance time, impulse and GRFω have been also calculated by two studies [44, 54].

Pressure mapping systems (PMS). Pressure mapping systems consist of a continuous network of sensors which measure multiple points of contact simultaneously, allowing for a detailed pressure map to be produced. They have been employed in 19 studies, 3 of which used a combined system of a pressure plate laid over a force plate. They typically come in the form of walkway plates and mats that allow animals to walk over or stand on them. However, one study developed an insole pressure system by attaching a sensor-equipped shoe (a leather claw shoe) to the cow’s left hind claw to measure kinetic outputs for both standing and walking conditions [62].

Pressure mapping systems had considerable variation in dimensions, ranging from 24.6 × 24.6 cm (I-Scan system to assess static pressure distribution of only the left hind foot) to 61 × 488 cm (Gaitwise system to capture more gait cycle steps for lameness detection purposes). The sensor resolution of a pressure mapping system is another aspect that varied among the studies that reported it, ranging from 1.4 to 3.2 sensing elements (pressure sensors) per cm². Recording rate of data (frequency) also varied greatly among the studies, ranging from 40 to 300 frames per second (Hz).

Pressure mapping system variables: Kinetic outputs measured by PMS are typically obtained through dynamic assessments, with the exception of 4 studies that performed static measurements [18, 21, 32, 57]. Vertical pressure, vertical GRF, impulse, contact area, and stance time are common variables measured among the studies using pressure mapping systems. One study also calculated center of pressure to investigate hoof pressure patterns during stance phase (39). Additionally, studies that have used the Gaitwise system [38, 46, 52, 59], provided mainly temporal (step time, stance time, and stride time) and spatial (step width, step length, stride length, tracking-up distance, and abduction) variables and their asymmetries. Relative force was the only force/pressure variable measured by the Gaitwise system.

Weight distribution platforms (WDP). Four-scale weight distribution platforms consist of four independent recording units (one for each limb) that measure the weight distribution between limbs while the animal is standing. These platforms have been used in 15 studies, in which they were installed either inside or outside automatic milking systems. Studies incorporating WDP mainly aimed to detect leg problems, as lame cows reduce weight-bearing on the affected limb, consequently showing asymmetry in weight distribution across contralateral limbs (i.e., legs on the left vs. right side of the cow). Each recording unit typically contained 1 or 2 transducers (load cells) measuring the load applied on them. However, two studies used a weighing platform that had 4 load cells within each recording unit [42, 50]. Dimensions of recording units ranged from 31 × 31 cm to 56 × 91 cm. The sizes of front and rear units may differ, as the rear recording units installed in the milking robots were usually larger than the front units. The duration of time that a cow stands on the weighing platform for data recording varied among studies, typically ranging from 2 to 5 minutes in 1 to 4 separate measurements. The reported recording rate (frequency) varied from 1.1 to 14 recordings per seconds (Hz).

Weigh distribution platform variables: Mean and standard deviation of the weight distributed between all legs or between a pair of legs were the most frequent variables that have been measured during the recording time in which animals stood on the weight distribution platform. Leg weight ratio (LWR) and weight difference (Δweight) were the other important variables associated with contralateral legs. The number and frequency of kicks and step behavior have been also calculated using weight distribution data.

Vision-based technologies (VB).

Vision-based technologies involve acquiring and analyzing the motion and posture (i.e., kinematic parameters) of a walking animal using videos or sequential images. These technologies have evolved considerably from manual annotation of images and videos to automatic motion/posture optical trackers, and computer vision and machine learning algorithms. The beginning of kinematic gait analysis in cattle using vision-based technology goes back to the study of Herlin and Drevemo [64], who measured angular patterns and hoof trajectories. Since 2000, there were 28 (34% of studies) published papers in bovine motion analysis that have utilized vision-based technology in different methods to measure locomotion variables while animals are walking through a corridor. Table 5 shows more information on vision-based technologies. The most frequent methods are video analysis (VA) and image processing (IP) techniques.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Overview of studies included in the review that used vision-based technologies (VB).

https://doi.org/10.1371/journal.pone.0266287.t005

Video analysis (VA) techniques were used in 13 studies, eleven of which involved digitizing and tracking anatomical landmarks of a walking cow using motion analysis software to extract kinematic data either directly from a video (8 studies) or from consecutive still images obtained from a video (3 studies). Labelling and digitizing the anatomical landmarks can be done either manually, which requires manual localization of several points of interest, or using a marker-based system attaching reflective markers to palpable anatomical landmarks to measure kinematic data more accurately. Seven studies have employed a marker-based VA method using different types of passive markers (retroreflective markers) including reflective tape, color cardboard and reflective plastic balls. Additionally, two studies used the VA method to look at the pattern of ground contact of the claws of animals walking on a treadmill [27, 72]. All VA studies employed only a single digital camera, which is usually positioned perpendicular to the plane of movement of the cows to record sagittal plane kinematics data. Therefore, they were only limited to analysis of two-dimensional (2D) kinematic data. Filming a walking cow has also been employed by a few studies to assist other technologies, i.e., force and pressure platforms and accelerometers, as a video synchronization in the determination of stride events (e.g., hoof strike, and toe off). These studies are not included under VB technologies in this review since they were not primarily used to produce gait variables.

Image processing (IP) is the other method of VB technologies that was used in 13 studies to analyze the motion and posture of a walking cow through an algorithm. After extracting still images of several desired frames from a video, some operations will be applied to the images to prepare them for the cow motion/posture recognition. Background subtraction is one of the major tasks in image processing procedure, which allows an image’s foreground, i.e., a moving cow, to be extracted from background for further processing. Then the motion/posture of the desired regions will be analyzed either manually or using machine learning algorithms. Studies that have employed IP techniques usually targeted one anatomical region. Back posture as one of the main lameness indicators is the most popular single target in the studies that utilized image processing techniques, followed by hoof and leg movement. Similar to the videography analysis method, the image processing studies also used a side-view camera, unless in the case of using 3D camera, i.e., depth-sensing camera, to produce depth images of the cow’s back posture that is positioned overhead.

The object/target detection system is a novel technique in the field of vision-based technology and computer vision that allows us to automatically identify and locate multiple objects/targets in a video. This technique, used recently in one bovine gait analysis study, works by applying deep learning algorithms to detect the leg, head, and back regions of the cow [90]. However, only leg data has been analyzed and published. Biplane high-speed fluoroscopic kinematography (HFK) was another new technique to analyze bone movement with high accuracy in a 3D approach that has been utilized in a bovine study to evaluate the range of motion of interphalangeal joints [89].

Type of cameras varies among all the VB studies from a digital video camera to 3D depth sensors and dome cameras with various recording rates ranging from 15 to 60 fps. A high recording rate of 500 fps was set for the HFK study [89] as well as for two studies that walked the animals on a treadmill for video analysis [27, 72]. In most of the studies it has also been stated that camera placement was set up in a position to allow researchers to evaluate at least two consecutive strides.

Vision-based technology variables. Vision-based technologies can provide distance (spatial) and time (temporal) parameters, angular range of motion (ROM) of joints, displacement, and velocities, all of which are classified under kinematic variables. Since video analysis techniques allow researchers to quantify motion and posture of multiple anatomical regions at the same time–i.e., several points (usually major joints) on fore and rear limbs, back posture, and head position–a more complete set of kinematic variables can be obtained. Stride length and stride time are the most frequent variables that have been measured using VA techniques, followed by maximum stride height, hoof velocity, stance time, swing time, triple support, hoof overlap and ROM of fetlock joint.

On the other hand, most of the studies that used image processing methods generally focused on one specific anatomical area leading to a particular variable. There are 7 IP studies that have focused only on the back posture and provided particular variables such as curvature angles (curvature angle of back around shoulders, curvature angle of back around hip joints, overall back curvature angle, curvature distance of shoulder and hip), back posture measure (BPM) and inverse radius. In addition to back curvature, one study [84] also looked at the hook bones in order to track the hind limb symmetry movements. There were two other studies that only looked at the hoof displacements and measured hoof location trackway, i.e., hoof overlap [71, 73].

Leg swing/movement has been another target for the kinematic studies using vision-based technologies. One study evaluated ROM of leg’s touch and release angles while another analyzed leg swing to generate six features referring to the gait asymmetry, speed, tracking up, stance time, and stride length [49, 87]. Also, a different work adopted an object detection system to detect leg movement and measured the relative step size characteristic vector [90].

Accelerometers.

Gait analysis in cattle has also been done using accelerometers which are attached to the cow’s leg to measure acceleration data. The output from accelerometers in bovine studies are generally classified as either behavior measures, including step activity, lying, and standing behaviors or as gait measures, including kinematics and kinetics of a gait cycle and symmetry between the legs. The current review focuses only on the gait measurement aspects of using accelerometers. There were 12 studies that have utilized the wearable sensors (i.e., 3D accelerometers) to measure gait variables. The details of these studies are shown in Table 6.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Overview of studies included in the review that used accelerometers.

https://doi.org/10.1371/journal.pone.0266287.t006

The typical location for attaching the accelerometer is above the fetlock joint at the metatarsal/metacarpal level of a limb; however, 2 studies placed accelerometer at thoracic vertebrae level as well. The main objective of using an accelerometer is to detect a lame cow or to find a gait abnormality in intervention studies. The number of accelerometers per cows depends on the research set-up and the purposes. Studies that used only one accelerometer per cow, i.e., attached to one of the hind limbs or the affected (lame) limb, looked at the consistency/inconsistency of a variable within an animal or the differences between animals. In cases where more than one accelerometer per animal was used, accelerometers were worn on 2 limbs (hind limbs only or one side of the cow) or all 4 limbs to study differences in acceleration data between the limbs. The sampling rate of the accelerometer varies among the studies, ranging from 10 to 400 Hz. It stands to reason that the studies which employed accelerometers with high sampling rates (more than 100 Hz) could provide more details on events and moments of a gait cycle.

Accelerometer variables. Measuring asymmetries and differences of acceleration data across the legs is one of the main parameters in the studies which employed more than one accelerometer per cow. Acceleration data has also been used to extrapolate kinematic and kinetic variables. Studies that used low frequency (less than 40 Hz) accelerometers have estimated stride time, stride length and walking speed in addition to behavioral measurements, while accelerometers with higher sampling rates have measured more detailed parameters of a gait cycle including stance and swing duration as temporal kinematic outcomes and foot-load, heel-off, and toe-off as kinetic outcomes.

Force platforms.

Force platforms (FP) are considered the gold standard because of their recognized accuracy and high frequency measurements. Beyond vertical force, 3D force platforms can output shear forces (i.e., transverse, and longitudinal forces). This allows researchers to investigate more in depth the gait attributes affecting transverse and longitudinal components of force [56]. Nonetheless, one of the FP limitations is that it is challenging to measure individual footfalls when more than one foot touches the plate at the same time (i.e., the best results are achieved when only one foot fits on the plate). Based on this, force plates generally come in multiple-platform layouts to register forces exerted by each hoof strike. However, walking animals on a multiple-platform setting to fit footfalls on the designated plates is challenging per se.

Research aim analysis shows that efforts have been made to employ FP as an automatic lameness detection system, generally using a commercial version of force platform for on-farm use (i.e., StepMetrix^™, Boumatic LLC, WI, USA). Despite this, the use of FP in studies dropped since 2015, suggesting that the weaknesses of FP may outweigh the strengths, especially as other technologies, by comparison to FP, are capable of measuring a more diverse assortment of variables that can more robustly analyze gait.

Pressure mapping systems.

Unlike the FP, pressure mapping systems (PMS) can distinguish and segment individual footprints regardless of how many feet are on the ground. As such, they can be used as a single mat or plate in different dimensions which makes them more applicable for kinetic analysis in quadrupeds. PMS can also quantify a wide range of pressure/force (only in vertical dimension), temporal, and spatial parameters, depending on the device’s specifications. Research aim analysis shows that PMS have been utilized to address different aims including evaluation the effects of hoof trimming, local analgesics, and flooring type on cows/claws’ biomechanics. In addition, lameness detection is another research aim that was studied, particularly using a longer version of pressure platforms (i.e., Gaitwise system). Utilizing a longer pressure mapping system allows registering more gait cycles and steps during a passage, which is a key for lameness detection purposes. In comparison with FP, PMSs are more practical and efficient in terms of easy implementation and generating more diverse outcomes either in research studies or at the farm level.

Weight distribution platforms.

While being a limitation for FP and PMS systems as well, numerous failed measurements because of placing hooves outside of the measuring zones have been reported as the primary weakness for WDP. WDP measures variables only while an animal is standing, so the outcomes are limited to static measures and no information on gait cycle kinetics can be taken. Nevertheless, the strength of WDP is the fact that it is not just a quick snapshot of the cow walking, but a longer review of how she distributes her weight. A lack of even distribution of weight makes it easier to pinpoint the limb of issue in these cows with greater certainty, as the reluctance to bear weight in this case is usually a sustained measure. Moreover, they can be easily implemented in farms with the least space occupation, such as embedded in the flooring of milking robots or cattle chute. The results of our scoping review confirm this, since WDPs are primarily used for lameness detection purposes at the farm level.

Video analysis.

Video analysis is the gold standard for kinematic measurements. They are able to provide a variety of gait variables from distances to ROM, velocities, and trajectories. They allow researchers to investigate multiple anatomical locations on the body with a single passage, yielding a better understanding of different aspects of an animal locomotion. Since impaired locomotion (lameness) can manifest in different ways (e.g., from short stride length to back arch, joint stiffness, asymmetries), a system that can consider all of these variables can provide a more accurate analysis. However, these systems, especially marker-based VA systems, pose several limitations, including marker placement error, observer (digitizer) error, and the time-consuming process of manual/semi-manual tracking. Moreover, it requires animal preparation (attaching markers on animal’s body) and a specific environment (room) wherein lighting conditions can be controlled and ample space can be provided so as to not block the cameras’ views of the markers. All these procedures can be challenging and make these techniques unfeasible to implement on commercial farms. The results of this review illustrated that the use of VA techniques have decreased over time, from 8 studies between 2005 and 2009 to only 1 study over the last five years.

Moreover, all the previous studies that used VA technology were limited only to 2D kinematic analysis which includes several limitations by comparison to three-dimensional (3D) analysis, where multiple synchronized cameras are being used. Some of these limitations include parallax error and perspective error that occur when subjects move away from the optical axis of the camera and subjects are restricted to movement within the plane of calibration and accuracy is compromised. Also, 2D angular data are limited to one degree of freedom (DOF) for joint rotation within the sagittal plane, while 3D kinematic data allow the researcher to explore the full range of motion (ROM) and orientation of segments and joints within 3D space [103]. Therefore, although the use of VA techniques is challenging, the high accuracy of these systems, especially the 3D video analysis systems, makes them a reliable reference for the future works of using marker-less VA technologies.

Image processing.

By comparison to VA systems, the downside of the IP methods is that, since they usually focus on one anatomical region as a lameness indicator, their outcomes are limited to the targeted region and the other facets of lameness will be missed. However, use of machine learning algorithms and eliminating the requirement of animal preparation (e.g., with reflective markers) allows them to be utilized as automated systems on commercial farms. Also, the use of overhead 3D depth image cameras instead of the side-view digital cameras benefited this technology to overcome the challenges of side-view image capturing at farm conditions, including occlusions and sensitivity to lighting variance in order to facilitate the image segmentation of the foreground from the background. Our results showed that efforts for implementing IP technology as automated lameness detection systems on farms has become more prevalent in gait analysis studies over the last decade (13 studies between 2010 and 2020).

Accelerometers.

Accelerometers have been used as a technology for gait analysis in cattle studies since 2009 and have been the most widely used technology in the last 5 years. Previous studies showed that pedometers (accelerometers) can be a promising tool for detecting lameness and foot pathologies in dairy cows. Uncontrolled walking speed and cow traffic as well as behavior are the usual limitations reported for gait analysis using FP, PMS and WDP and vision-based technologies, but not for accelerometers. Accelerometer outcomes are not dependent on a single passage or a specific walking corridor. At present, however, accelerometers still carry limitations. For example, not all variables can be measured using a single pedometer, with useful gait metrics for detecting impaired gait, such as asymmetry of acceleration between legs, requiring 2 or more devices. Technologies that require more hardware can be a barrier to adoption on commercial farms [104]. Moreover, current pedometers used on farm generally have lower data sampling rates as they are less costly than high-frequency pedometers; however, this trade-off can compromise the efficacy of the pedometer use for detecting impaired gait. Low frequency accelerometers can make gait analysis more challenging and requires more research.