Using Inertial Sensors in Research: What to Take into Consideration
A Conversation with University of Missouri Professor of Equine Science Kevin G. Keegan, DVM, MS, DACVS
WHY USE INERTIAL SENSORS OVER OTHER METHODS OF OBJECTIVE MEASUREMENT?
DR. KEVIN KEEGAN In my honest opinion, if you are only interested in measuring lameness in horses, for whatever reason, then the only way to do it practically today is with body-mounted inertial sensors. If you are interested in measuring something else, for example rider position on the horse, limb movement effects with shoeing, or if you are interested in developing a method of lameness evaluation that is not based upon vertical movement of the torso, then other methods may be for you. I will add, as words of caution, I doubt you will find a better method of lameness measurement based on anything other than vertical movement of the torso (I have tried). And be prepared to spend much effort – with some methods just trying to get enough measurements and with other methods searching for meaningful patterns in mountains of data.
WHY ARE INERTIAL SENSORS SUPERIOR TO THE STATIONARY FORCE PLATE FOR STUDYING AND MEASURING LAMENESS IN HORSES?
KEEGAN Lameness is an elusive entity, with much variability, and lameness in one limb depends on what is going on in the other limbs, and this elusiveness is compounded in quadrupeds. The stationary force plate gives you data on one stride, in one leg. You need much more than this. Also, you will have to look at the horse always in the laboratory situation, under controlled environmental conditions of speed and surface.
WHY ARE INERTIAL SENSORS SUPERIOR TO LINE-OF-SITE TECHNIQUES USING CAMERAS WITH OR WITHOUT BODY MARKERS, OR THE VIDEO-BASED, KINEMATIC MOTION ANALYSIS SYSTEMS?
KEEGAN I have used two of these systems for many years and am very familiar with how they work, the data they generate, their accuracy and sensitivity (they are high), and their limitations and pitfalls; however, they are still not very practical. There still is one major limitation, unlikely in the near future to change with line-of-site kinematic video systems. Your sensitivity is limited to the ratio of the size of your subject (the horse) and the area of the field of view. When the subject of measurement is small relative to the area of the field of view spatial resolution suffers. Also, things get in the way. Only highly thought-out (and expensive, like several cameras arranged side by side to cover large fields of view) or unnatural (treadmill, expensive too) measurement areas make it possible to get enough sensitive data to handle the many-times mild and frequently variable lameness seen clinically in horses.
When using inertial sensors, or specifically, the Equinosis® Q, to study lameness, there are several things one should take into consideration. The remaining discussion focuses primarily on using inertial sensors to study change (e.g. from treatment or diagnostic analgesia).
WHAT VARIABLES SHOULD YOU LOOK AT WHEN USING THE Q TO STUDY LAMENESS?
KEEGAN For evaluating forelimb lameness change over time use the mean Vector Sum (Total Diff Head) as a measure of amplitude of forelimb lameness. It is important to note that the trial report itself does not sign the vector sum for side of lameness – i.e. if the lameness switched sides at some point, the VS does not change from positive to negative or vice versa on the report. However, if you export your data into a CSV (you can do this from the program), it will include a signed VS; so, LF lameness will have a negative VS and RF lameness will have a positive VS. This allows you to calculate change if the asymmetry changes sides.
For hind limb lameness, it is best to calculate change in Diff Max and Diff Min pelvis separately, as they are independent variables, with Diff Max being a measurement of lack of push off and Diff Min being a measurement of lack of impact. For this reason, a horse can measure with lack of impact in one limb and lack of push off in the other limb within the same stride (unlike the forelimb).
Ex A. A horse with a Diff Max Pelvis of -16 and a Diff Min Pelvis of – 5 has a left hind push off lameness of lameness of -16 and a left hind impact lameness of -5.
Ex B. A horse with a Diff Max Pelvis of -16 and a Diff Min Pelvis of + 5 has a left hind push off lameness of -16 and a right hind impact lameness of + 5.
IF YOU NEED TO CATEGORIZE BASELINE SEVERITY OF HIND LIMB LAMENESS AS A WHOLE FOR CASE SELECTION PURPOSES…
- YOU CAN ALSO ADD DIFF MAX AND DIFF MIN TOGETHER to obtain a total measurement of hind limb lameness BUT ONLY if they are of the SAME SIGN and BOTH variables are ABOVE threshold.
METHOD 1 EX. A horse with Diff Max Pelvis of -16 and a Diff Min Pelvis of -5 has a left hind lameness of -21 (no timing indicated).
- YOU COULD ALSO USE THE ABSOLUTE VALUE of Diff Max Pelvis or Diff Min Pelvis BUT ONLY if one measure is above threshold. For instance, you could use |Diff Max| if Diff Min is below threshold or |Diff Min| if Diff Max is below threshold.
METHOD 2 EX. A horse with a Diff Max Pelvis of -10 and a Diff Min Pelvis of + 2.2 has a left hind limb lameness of 10.
- A COMPLICATION OF COMBINING DIFF MAX AND DIFF MIN PELVIS into a total score to evaluate change is that if the impact component gets worse but the push off component gets better, conclusion of improvement is confused.
EX: A horse has a Diff Max Pelvis of -4 and Diff Min Pelvis of -10 before treatment. After treatment Diff Max Pelvis is -10 and Diff Min Pelvis is -4. This would be unusual; however, you can see the confusion! Diff Max and Diff Min show statistically significant change, but overall amplitude of lameness is unchanged.
WHAT ARE SOME OTHER THINGS YOU SHOULD TAKE INTO CONSIDERATION REGARDING SUBJECT SELECTION?
KEEGAN Bilateral Lameness: Any system or method that measures lameness as asymmetry will not be able to detect with high sensitivity, or measure with high precision, forelimb or hind limb lameness that is truly bilateral – especially if the lameness severity is distributed evenly between right and left limbs in every stride.
This is true for all methods – including body-mounted inertial sensors, line-of-site kinematic analysis (video), and the stationary force plate. It is the method (measuring asymmetry), not the equipment, that produces these results.
For instance, a horse with a hypothetical grade 3 lameness in the right limb and a hypothetical grade 2 lameness in the left limb, will be measured with a hypothetical grade 1 (because 3 – 2 = 1) lameness in the right limb. In the practical sense, however, it is very unusual for a horse with a bilateral lameness to have the same amplitude of lameness in both sides on all strides; so detecting the existence of bilateral lameness in the clinical evaluation with methods of asymmetry measurement is generally not an issue – it just requires some diligence.
Here is a more specific example that I remember from reviewing a paper several years ago. The authors were evaluating horses with hind limb distal tarsal arthritis with a stationary force plate and using difference in vertical ground reaction force (in units of % body weight) between right and left limbs as the measure of the amplitude of lameness. A +10% difference was a right hind limb lameness. A -10% difference was a left hind limb lameness. A +10% difference was a worse right hind limb lameness than a +5% difference and vice versa for a left hind limb lameness (i.e. -10% difference worse than -5% difference). In this study a horse with a +10% difference before treatment and a +5% difference after treatment was considered improved! This may not be correct; the horse may not have really improved.
As an example, assume that before treatment a horse had a grade 3 lameness in the right hind limb and a grade 2 lameness in the left hind limb. Let us assume that this was equivalent to a +10% difference in vertical ground reaction forces between the right and left limbs. And then, after treatment, the horse had grade 3 lameness in both hind limbs – a real worsening of the overall lameness. But the grade difference after treatment is 0 and the difference in vertical ground reaction force between right and left hind limbs is less than before treatment. This horse got worse; but, based on asymmetry, it would have been assessed as improved. It is best to avoid subjects with bilateral lameness when evaluating change over time.
Compensatory Lameness: Watch out for potential compensatory forelimb lameness due to primary hind limb lameness or vice versa. If a subject has multiple limb lameness, be sure you are evaluating the primary lameness and not the compensatory lameness, which can vary trial to trial and evaluation to evaluation; and would not be suspected to improve if that limb is the treated limb.
WHAT ARE THE DATA COLLECTION PROTOCOLS YOU SHOULD FOLLOW?
KEEGAN Collect trial data with low stride-by-stride variability (standard deviation not greater than mean). It is most important to evaluate the standard deviation of Diff Min Head, as Diff Min Head determines the side of forelimb lameness, where Diff Max Head determines timing (impact, midstance or push off). For hind limb lameness, consider Diff Max and Diff Min and their respective standard deviations independently.
STRAIGHT LINE: Collect at least two trials back to back, obtaining at least 25 strides in the analysis. If the two trial results are not within the 95% confidence intervals (i.e. 8.5 mm for head VS, +/- 3 mm for pelvis DMax and/or DMin), collect a third trial. This is stabilizing the lameness.
FLEXION TESTS: Collect a separate baseline before flexion of, ideally, 8-10 strides; and the same number of strides after flexion. Use consistent collection conditions, such as only when the horse is trotting away, in both before and after trials.
LUNGING: Collect more than 25 strides in each direction. While there is no experimentally determined number of strides recommended to evaluate lunging horses, a higher number of strides helps to reduce stride by stride variability inherent on the lunge. A good rule of thumb is 45-50 strides.
UNDER SADDLE: If your study includes posting, use the full arena to reduce the effects of induced torso tilt seen on a small circle. Lunging or small circle artifacts on soft ground counteract posting artifacts, thus potentially cancelling asymmetries induced by the other activity.
RACING TROTTERS AND PACERS: I have evaluated a few Standardbred horses at the track, jogging and at training speeds (going the right way of the track). In these situations, I found it less important to separate out strides collected in the turns from strides collected in the straight away, as the horses do not jog or train in both directions equally, and you will not get the equivalent analysis “in the other direction”, as you would when comparing lunge left to lunge right, or ride left to ride right. Also, at the track one is capable of collecting increased numbers of strides (I have collected easily in excess of 400 strides in one trial) with very low stride-by-stride variability, so that the effect of the turn on the overall assessment for lameness is minimized over the whole trial. I will also just add this. Little is known about how lameness changes, both in terms of amplitude and variability, as trotters and pacers approach closer to racing speeds (pacers have less stride-by-stride variation when they go from slow to fast jogging/training). Do they become more or less asymmetric as a consequence of increased speed, whether they have pain or pathology or not? More work needs to be done with inertial sensors on this question.
WHAT ARE THE ADVANTAGES AND DISADVANTAGES OF USING A TREADMILL FOR DATA COLLECTION?
KEEGAN While no published study to date has directly compared vertical head and pelvic movement displacement measurements between overground and treadmill trotting, there is evidence to suggest that the biomechanics of movement are different on a treadmill versus over ground and that differences should be expected. Thus, studying certain aspects of lameness may not be appropriate on the treadmill. The decision to use a treadmill for data collection is often based on the desire to maintain a consistent velocity within and across trials. Variability may be lower on a treadmill, but only when using horses that have been adequately trained to load onto and move on it. Due to the expected alteration of biomechanical forces on the limbs, studying clinically relevant manifestations of naturally occurring lameness on the treadmill is not advised. Studying effects of change on measured lameness on the treadmill may be useful.
In closing… Prior investigative efforts into lameness in horses required time-consuming and expensive and/or complicated laboratory setups with multiple cameras, force-measuring devices, etc. With the advent of body-mounted inertial sensors and software specifically designed to measure lameness in horses, multiple avenues of investigation have been opened up to basic scientists and clinical investigators interested in studying lameness in horses.