Too high ankle joint moment during running

[Lars van Rengs]

Hi OpenSim users,

We are having some issues with the ankle moment calculated using the inverse dynamics tool in OpenSim. Compared to literature, the ankle moment is clearly higher than expected. Our peak ankle joint moment is about 3.5 Nm/kg, where it should be about 2.6 Nm/kg (at a similar speed). The moments of the hip and knee are largely in agreement with the literature, although the knee moment is a bit lower compared to some studies.

We also compared the joint moments from the OpenSim analysis with a different model (Human Body Model), and the Human Body Model produced similar values as the literature (attachment 1). Both models use the same C3D input, so it seems the input (GRF, markers, COP) is good, and the issue occurs within OpenSim.

We have tried many different ways to solve this problem with the joint moments. Unfortunately all without result. A small overview of the things we have tried:
1. Changing the marker positions of toe markers on the generic and scaled model
- Very small effect on kinematics, but the ankle joint moment did not change
2. (un)locking MTP and subtalar joints
- Locking subtalar joint only makes the joint moment worse
- Unlocking MTP is also not the solution
3. Other scaling-settings of the foot (attachment 2)
- First, we use foot xy to scale talus, calcaneus and toes. Where foot xy is the average distance between left heel and LMT2 and right heel and RMT2.
- We also tried to use shank xy to scale the talus, but this only reduced the ankle joint moment with 0.1 Nm/kg. We set the shank xy as the average distance between right lateral knee marker (epicondyle) and right lateral ankle marker (malleolus); and left lateral knee marker (epicondyle) and left lateral ankle marker (malleolus).
4. We also tried another generic model.
- We used the Catelli Human Body Model with or without arms but there was no difference in joint moments between the two models, nor a difference when using the Hamner model

[Ton van den Bogert]

Hi Lars,

This is an interesting observation.

Normally the ankle moment is very reliable and not sensitive to modeling assumptions. When Antoine Falisse compared HBM results to OpenSim for walking, the ankle moments were almost exactly the same. Link to the article: https://pubmed.ncbi.nlm.nih.gov/29809082/

My first suggestion would be to make sure that the low-pass filter settings are the same between the two software systems. In running, frequencies are higher than in walking. The default HBM filter cutoff may be too low. This could cause a reduction in the peak. When comparing to results from the literature, keep an eye on what filters they used in those studies.

For running, I have used 8 Hz before (https://pubmed.ncbi.nlm.nih.gov/17943488/). You can probably go up to 15 Hz with modern motion capture equipment. Especially for the ankle moment, which is not affected much by noisy accelerations, since the foot mass is low.

My process is that I try various filter settings, higher and higher cutoff until the noise becomes too high. Then you're sure you are not removing important parts of the signal. Your HBM results look too smooth. Try to modify the filter settings until you start seeing noise.

Ton van den Bogert

[Lars van Rengs]

Dear Prof. van den Bogert,

First of all, thank you very much for your quick response.

We already applied a high filter frequency of 20 Hz for the HBM model, which corresponds to 24.9 Hz for the ground reaction forces in the OpenSim model and 23.2 Hz for the kinematics according the the Fallise paper. We used the same filter frequency for the kinematics and the ground reaction forces in the HBM model to avoid artificial variations in joint moments, and also experimented with this in OpenSim (e.g. both GRF and marker data with 24.9 Hz). Moreover, we performed the ID also with lower filter settings (e.g. 6 Hz). Regardless of the filter settings, the ankle joint moment in OpenSim however remained too high. Therefore, it seems to us that the filter frequency is not the issue for the high ankle joint moment.

With regard to the smoothness of the HBM signal, we would like to clarify that this is an average signal: we took the average time-normalized data over 60 seconds of running, and this explains why the signal looks so smooth.

We would be very happy to hear if you (or any other OpenSim users) have any other suggestions for this issue as we don't know what other options we could explore to solve this.

[Ton van den Bogert]

After your experiments with the filter settings, I totally agree it is not a filtering issue.

Something is probably wrong in the OpenSim model.

You may have done this already, but I would visualize the skeleton motion and the GRF vector in OpenSim, and make sure that the center of pressure is in the right place, relative to the ankle joint center. Especially at the time of the peak moment. It could have gone wrong due to scaling, or inaccurate marker placement in the model. The residuals of the inverse kinematics could give an indication also.

You can do the same visualization in HBM and compare.

The error small enough to make it hard to troubleshoot, but large enough that you should not ignore it.

Ton van den Bogert

[Lars van Rengs]

Dear Prof. van den Bogert,

Thank you very much for your response.

We visualized the GRF vector in OpenSim and this looked accurate. In the attached figure we have determined the distance between the ankle joint center (AJC) and the center of pressure (COP) for the X, Y and Z directions. The dashed lines indicate one stance phase. From this figure it can be seen that the difference between the OpenSim and HBM AJC-COP distance is <0.5 cm. Moreover, the distance between the AJC and COP is even even slightly smaller for OpenSim than the HBM model, so this is also unlikely to explain the higher ankle moment.

With regard to scaling, we have tried many different options. For example, assigning different weights to various markers, and moving the markers of the toes and heel more anterior/posterior as well as up and down on the foot. We have also attached the markers on the toes to the talus instead of the toes as an experiment, but all options did not resolve the high ankle moment. When we visualized both the experimental markers and virtual markers in OpenSim, they match very well and the deviation is only minimal. Therefore, this also seems unlikely to be the cause.

We did notice however that the peak ankle joint computed in real-time with the HBM model using D-flow was also ~3.5 Nm/kg, and in line with the OpenSim ankle moment (see figure below). Yet, this moment is reduced to ~2.5 Nm/kg after offline re-processing in the Gait Offline Analysis Tool (GOAT; we do not know why this occurs as the GRF, COP, and kinematics are all very similar). This could suggest that the ankle moment we currently get with OpenSim is correct, and that GOAT is producing an incorrect ankle moment. Yet, a peak ankle moment of ~2.5 Nm/kg is in line with other literature, and ~3.5 Nm/kg is substantially higher. Moreover, a peak ankle moment of ~3.5 Nm/kg also required our model strength to be increased by at least 10 times for the muscles to reproduce this joint moment, which we also consider unrealistic since the model is already quite strong compared to other models.

Therefore, we don't know what other things could cause this issue and would very much appreciate if you or anyone else has other suggestions regarding what we could look into.

[Ton van den Bogert]

It definitely looks like something is wrong with the GOAT results. It is supposed to be the same as HBM, so you should contact Motek support and have them look at it.

The lower peak, and the overshoot to negative moment after push-off, are important clues. It almost seems as if GOAT used a filter with incorrect coefficients, but it could be something else.

I would not assume that you should expect a peak of 2.5 Nm/kg. Ankle moment could be sensitive to a small difference in speed, footwear, and possibly vary a lot between subjects.

I can't really judge your COP values, the vertical scale is in meters. You would need to zoom the vertical axis to about the size of the foot to make differences more clear. Don't worry about COP during the swing phase.

It looks like OpenSim was correct all along, but I would want to know what Motek says. That's not for this forum though.

Ton van den Bogert

[Lars van Rengs]

Dear Prof. van den Bogert,

First of all, thank you very much for your quick response. We also want to thank you for your time and input.

We have already contacted Motek about a week ago with our question about the high ankle moment and the difference between before and after GOAT analysis. When we have a response, we will let you know through this forum.