Has-Motion Resources

Challenges and Opportunities with Symmetry Scores for Gait Assessments

It is complicated to use gait assessments in decision-making, whether in a clinical or sporting environments. First is the matter of how complicated gait is by itself. There are so many things we can measure that it is hard to say what we should pay attention to. Second, we need to boil our complex data down into something reliable for what is essentially a yes/no decision. Is a patient’s condition worsening? Can an athlete return to play? And third, it’s not clear what we should be assessing the individual against. Our gait is robust and tolerates a significant amount of variability between strides and between individuals. This robustness is great for walking, we tend not to fall over, but makes it hard to pick out meaningful comparison data.

Photo by Nigel Tadyanehondo on Unsplash

Enter symmetry scores. These scores are a tempting way forward because there is an intuitive argument that healthy gait is symmetric and this gives us a single biomechanically-anchored score that can lead directly into a decision. Oh, and our comparison data? We get that for free from the participant themselves! But there is an inherent tension between simplifying many biomechanical waveforms into a single symmetry measure to support decision-making and throwing out information that is potentially necessary to ensure that this single measure is reliable. At a recent journal club we looked at two approaches to calculating and interpreting symmetry scores for gait.

Paper 1: Comparing Inertial Measurement Units to Markerless Video Analysis for Movement Symmetry in Quarter Horses

Thilo Pfau and colleagues highlighted the hope that symmetry measures will allow biomechanists to reduce variable waveforms into a meaningful score, enabling comparisons between measurements made outside of the lab. Their study actually looked at movement symmetry in quarter horses but we thought it was insightful because of the sensors and modelling approach they used.

The study compared two different sensor systems: IMUs and a 2D markerless motion capture system. The authors’ assessed that neither system could be trusted as a gold standard, which we agree with, so they framed their study as a comparison study and set aside questions of specificity and sensitivity.

Mathematically, the authors modelled symmetry as the similarity between spatiotemporal measures for different strides. Notably they did not require strides to be compared against matched strides, with the goal of making their results robust to automated (and potentially flawed) stride matching algorithms. In this spatiotemporal approach, our data is simplified early and then summary statistics–in this case the spatiotemporal measures–are compared between trials. The authors concluded that the symmetry measures computed for the different sensors were interchangeable but they also noted that the proportional limits of agreement were larger than the clinically-accepted threshold values for the decision of lameness in quarter horses. This is troublesome, since it implies that our decision might change with a second measurement even if nothing about the individual being measured changed! It may be preferable to compute symmetry on the basis of complete waveforms rather than summary statistics.

Figure: The spatiotemporal approach to symmetry measures used in Paper 1. The authors extracted their chosen parameters from biomechanical waveforms, found the differences between these parameters for (usually) matching gait cycles, and then summarized these differences.

Paper 2: Reliability of a Global Gait Symmetry Index Based on Linear Joint Displacements

Silvia Cabral and colleagues asserted that symmetry scores must be both consistent and reliable to be used in meaningful clinical decision-making. Whether the decision is go/no-go for surgery or deciding whether or not a person is getting better, it is important that changes in the symmetry score should reflect changes in the person’s movement. Conversely, if the movement stays the same then the symmetry score should also stay the same. We thought this paper was insightful because of its clarity on what a gold standard symmetry score should look like and the implications of this ideal for current scores in the literature.

Here the authors chose to model symmetry as the cumulative sum of bilateral differences for quantities like joint angles. This cumulative sum approach takes into account differences across the whole movement and delays the moment of simplification. The effect of this process is to give the whole waveform a chance to impact the symmetry score, though previous work has shown that this approach can be overly sensitive to functionally irrelevant differences in movements and inconsistent marker placement. The present study proposed a linear GGA (LGGA) where the symmetry index is the cumulative sum of bilateral differences between linear joint positions in relation to the pelvis. This LGGA was more reliable than other GGA formulations with the trade-off of being potentially less interpretable for the patient or athlete.

Figure: The Cumulative Sum approach to symmetry measures used in Paper 2. The authors compared entire waveforms for matched gait cycles, computed a similarity score from these comparisons, and then summarized these comparisons.

Conclusion

We agree with the authors of both papers that symmetry scores show promise as summary statistics to enable comparisons of movements across research groups and over time. There are still some open questions to resolve before we take the next step, not the least of which is being able to establish a symmetry score that is sufficiently reliable to support decision-making. As a core takeaway, we agree that it is crucial for changes in symmetry scores to reflect meaningful changes in movements and for meaningfully similar movements to have similar symmetry scores.

Along this road towards reliability for symmetry measures, we wondered:

  1. Should symmetry measures account for the fact that some bilateral differences are functionally irrelevant and exclude them from the final score?

and

  1. If inconsistent marker placement or limitations of 2D markerless motion capture are preventing symmetry scores from being reliable, would 3D markerless motion capture systems improve this?

Like this article? Share with friends!

Facebook
Twitter
LinkedIn

Return To Our Resources

HAS-Motion

Our Software Solutions

Latest Resources