Tottenham Hotspur Stadium, London Rehabilitation & Injury Prevention Conference >
Article

The Speed Signature: A new way to understand “how” athletes run

Speed signature
Supported by

Have you ever looked at an athlete and thought, “he / she is not an efficient runner?” Have you ever looked at one of your rehab athletes and thought “that’s not what they looked like before the injury?”

Well, I did that for years, leading physical performance departments in professional field sports since 1998. My interest is not only in how fast I can get athletes to run but how efficiently, as they often must run 11-15 kilometers at a variety of speeds, not just flat out. Not only that, I also try to understand why players get injured and what to do about it when they do.

More and more research has come out in the last decade that clearly describes “how” athletes run [1–3]. Numerous authors have also taken the next logical step and described the deficits that occur post-injury, mainly hamstring injuries [4–6].

My question has always been “How do I practically use this information?” Various authors [7,8] have produced elegant, practical models of describing running gait based on video analysis, a path I have been down more than once. However, when it comes to managing 50+ runners over long periods of time, detailed video analysis, much of which requires manual editing and coding, is impractical.

While a handful of companies attempted to make gait evaluation tools, there was barely any meaningful equipment or methodologies on the market to obtain the type of information I needed.

So, with my innovation hat on, I went and built what I needed.

I didn’t go randomly off on a wild goose chase. There was enough evidence around [9–11] to suggest that it could be done… it just hadn’t been. While I was extremely motivated to innovate my practice by adding this type of data, I had to keep in touch with my analytical side to ensure that the work I was developing was legitimate.

It took me essentially eight and half years. I made lots of mistakes, learned to write computer code and ticked all the clinimetric boxes (see speedsignature.com for more detail) along the way to ensure the technology I was implementing was valid. That said, I’m sure I will get some dissenting opinions from the pure academic community. But I built this for my peers working in the field: S&C coaches, physiotherapists and rehabilitation specialists who have the same questions I had.

Speed signature
Figure 1. Speed Signature

Speed Signature is one-stop shopping for a multitude of factors

For an S&C practitioner, rehab specialist or physio involved in running sports, most of this should make immediate sense. Let’s put a little detail on it to make it immediately usable (but not too much… this can get very nerdy very quickly!).

  • All data is captured in straight line running at constant speed using a commercial IMU secured on the lumbar region with a purpose-built belt.
  • All data is plotted as a Z-score, so it is unitless and relative to a population average for each variable.
  • Data is reported in a 1 m/s speed window, the highest reliability (~0.87) we could demonstrate.
  • Upper and lower range of the population’s data is represented by the grey dotted line.
  • Data represents both limbs.
  • Kinetic variables are surrogate representations of mass relative force (i.e., impulse) in all three planes .
  • Kinematic variables are ground contact time, step frequency and step length (noted as stride).
  • Dynamic stability variables are stiffness and centre of mass displacement, and the ECON variable represents how much the body deviates at the centre of mass in all axes while running.

To validate the speed signature, I synchronised data gathered from commercial GPS units (Catapult S5) and a force plate during running, and found a very strong relationship between ground contact time (GCT) and patterns within each signal. The data was highly reliable within 1 m/s speed bands. This observation led us to develop strict assessments of multiple speed bands for individuals, given the large variability caused by individual strategies to increase speed.

Strive Tech
This article is proudly supported by Strive

One thing I have become acutely aware of using this data over the years is that gym- and physio-based tests don’t really count as much as you might believe. Don’t get me wrong, strength before speed is a core mantra of my rehab practice. But as Morin [1] has shown, it’s how the athlete applies that force in running that makes a difference. Despite great calf raises, Nordic (eccentric hamstring) data off the charts, single leg strength, etc, athletes still quite often have gaping holes post-injury when they hit the track.

I’m not suggesting we dispense with strength measures. Far from it. But extrapolating just gym performance or physio testing to running preparedness is an incomplete picture. Evaluating “how” the athlete is running – their process – at different speeds is far more significant than just evaluating the speed at which they run, their outcome.

“Evaluating “how” the athlete is running – their process – at different speeds is far more significant than just evaluating the speed at which they run, their outcome”

@JasonAWeber
Tweet This

Understanding the Speed Signature by comparing two athletes

The two plots below show two athletes of the same age, same mass and similar athletic ability.

To confirm that I wasn’t seeing things, I used a computational technique called “ensemble machine learning” to see if a computer can find any distinct patterns without me giving it any hints. This is an example of “unsupervised learning.” I identified two patterns that form a continuum. The patterns aligned perfectly with the concepts of aerial and terrestrial running described throughout the sports performance literature [12–14], and is also consistent with contemporary coaching philosophies.

We can classify athletes on the continuum between aerial (left) and terrestrial (right). Aerial athletes demonstrate a low ground contact time, are step frequency dominant and are vertically stiff / have a low centre of mass displacement. Terrestrial athletes have a higher ground contact time, are step length (stride) dominant and have low vertical stiffness / high centre of mass displacement.

There is more to aerial vs. terrestrial than just these three variables, but they are central to the process and are very relevant from a practical coaching perspective.

Speed signature: gait classification
Figure 3. Gait Classification Continuum. Box = mean classification, blue line = CI (variability of each classification), RED dotted line is the population mean, beige area is the population CI (variability) Speed Bands 1 = 4.5-5.5m/s, 2 = 5.5-6.5m/s, 3 = 6.5-7.5m/s, 4 = >7.5m/s

Athlete One (Figure 2, left) is aerial dominant at all speeds (not all speed bands are shown). Athlete Two (right) is classified terrestrial at lower speeds, but as he goes faster he becomes aerial to about midway on the continuum (Figure 3).

Figure 3 shows the need to evaluate these classifications across a range of speed bands to understand the athlete’s strategy for generating constant speed within the velocity band (as opposed to strategies for accelerating). This represents the concepts Sandford [16] presented when observing that many athletes demonstrate higher variability at higher speeds, thus requiring specific work.

You might think all athletes become more inclined toward aerial strategies (lower GCT, greater stiffness, etc) at higher speeds. Some do, but out of 20,000 samples I’ve taken about 30% of athletes become markedly more terrestrial as they get faster. This is not necessarily a good thing nor the most efficient, but it’s how they achieve speed. It speaks to the inefficiencies of many field sport athletes, and is very consistent with recent work by Sandford [15] highlighting the need to understand individual gait patterns to achieve the most accurate training interventions.

Applying the Speed Signature to improving individual athletes

This analysis process has impelled me to approach my athletes differently. Being involved in professional rugby union, AFL and football for over 25 years I have always had a focus on running skill and capacity. Covering 11-16 km in repeat speed modalities is critical to performance in those sports. Now with more information about how athletes achieve different fractions of their speed reserve (also Sandford [15]), I have an another avenue to further develop athletes.

Let’s take another look at the two examples from above.

Athlete One (the aerial athlete) was a very good 2 km time trialist when I met him. At lower speeds, though, he showed a vertical tendency that I believed impacted his acceleration. Force-velocity profiling would likely have confirmed this, but at that time we were not able to gather maximum sprint data, so input from that test was unobtainable.

F-v profiling also assumes all athletes produce identical and consistent vertical force, an observation that I think is inaccurate.

The Speed Signature observation reflects JB Morin’s “technical application of force” [1]. If you look within Athlete One’s kinetic variables, the horizontal force surrogate (CMV_H) is very low relative to vertical (CMV_V). While Speed Signature uses different calculations than Morin [1] for a variety of reasons, the construct is applicable. This athlete achieved speed by ”pushing” off the ground , not by “pulling,” to borrow Stu McMillan’s terms. My interventions with this athlete were aimed at teaching him to actively accelerate his thigh down and back (scissor action) to achieve greater thigh velocity at GCT, such as through bounding and resisted-running modalities.

Athlete Two presented as a solid runner but struggled to produce an elite 2 km time trial, a valuable physical attribute in the AFL. He also struggled with repeat speed after halftime in games, despite a reasonable max speed relative to his peers.

My intervention was based around increased stiffness through his entire body, critically focused on ankle, knee and hip integration. For running specifically, I focused on cadence drills and improving his ability to efficiently control his running skills at higher fractions of maximal speed, taking a page straight out of the anaerobic speed reserve play book [16].

Both players developed into highly successful AFL players. I am not claiming this result is entirely due to my interventions, but simply providing these examples to show the practical application of the data in Speed Signature.

Speed Signature and asymmetries

Quantifying lower limb asymmetries was one of my key objectives in developing this system. The location of the hardware at the lumbar spine makes it easy to assess input from both limbs.

The tricky thing with assessing asymmetry is defining exactly what an asymmetry is. There have been several review pieces in recent years [17,18] that describe the variety of techniques used in this calculation. I went with the most scientifically correct: ”an interlimb asymmetry is defined as a difference between limbs that is greater than the difference within a limb” [19]. To apply that definition, we used repeated measures ANOVA with a Bonferroni post hoc, as shown in the figure on the right below.

Speed Signature with an asymmetry assessment on the CMV_H variable, post proximal hamstring tendon rupture and surgical repair
Figure 4. Speed Signature with an asymmetry assessment on the CMV_H variable, post proximal hamstring tendon rupture and surgical repair (Athlete Three).

Each dot on the box plot (Figure 4, right) represents the average data of one trial, which I found is the best method to communicate accelerometer data. The boxes show the middle 50% of the data with the line in the middle being the mean. The magnitude of the difference between the two boxes is the key observation. In the example above, CMV_H is higher on the right by a margin that is very significant (p<0.001). In this case the asymmetry is real and worth examining further.

Figure 4 is from an athlete who suffered a collision-based proximal hamstring tear on his right side that was surgically repaired. While there are no specific research articles on proximal hamstring reconstructions, based on some good work previously done by Brughelli et al., [5], I would have expected the athlete’s right horizontal surrogate force signal (CMV_H) to go down. In fact, it went up in comparison to the pre-injury report (Figure 5), which also shows a small asymmetry that has markedly deteriorated in an unexpected way.

“Many of these issues aren’t obvious until speed is introduced, which is why I’ve always believed we need to critically evaluate running skill after major injuries, not simply strength tests”

@JasonAWeber
Tweet This
Pre-injury Speed Signature with an asymmetry assessment on the CMV_H variable, Athlete Three.
Figure 5. Pre-injury Speed Signature with an asymmetry assessment on the CMV_H variable, Athlete Three.

The proximal hamstring repair appeared to have created more problems around the hip with large instability on the right limb in CMV_M (indicator of frontal plane stability) and a loss of vertical force production (CMV_V). This meant that the key to improving this athlete’s return to full training and competition was a focus on his glute function in running (both vertical and lateral), not specifically focusing attention on the hamstring, which would have been the obvious conclusion. The “obvious” answer is not always the correct one.

Practical interventions included high volume multi-plane hip control work, vertically oriented running drills and a high volume of coordination work. That last aspect often gets overlooked in lower limb injuries. I think much of the asymmetry we see in the figure above results from coordination issues post-injury. Many of these issues aren’t obvious until speed is introduced, which is why I’ve always believed we need to critically evaluate running skill after major injuries, not simply strength tests (e.g., eccentric hamstring force).

Speed Signature introduces a future of possibilities for athlete performance

Speed Signature was developed to work within a system that quantifies the athlete from multiple perspectives to provide the best opportunity for performance and rehab professionals to design training interventions.

I hope this concept will contribute to improve our collective practice with our athletes and, maybe, encourage people to go out and find the answers to questions they have even when it doesn’t seem possible.

Supported by

STRIVE allows you to compare body input to body output, and provides insights on muscle activity including load, symmetry, and fatigue. With both IMU and EMG capabilities, STRIVE goes beyond measuring only External Load.

STRIVE delivers the most complete, accurate and actionable muscle data in human performance inside a platform that provides powerful insights to elite athletes and coaches.

References

Show