Schutte muscle deprecated

[Ilan Eskinazi]

Hello,

I was wondering why was the Schutte1993Muscle class deprecated with the release of OpenSim 3.0. Does it have an implementation error as did the Thelen2003Muscle? Moreover, will this deprecated muscle class be supported in future OpenSim releases?

Thanks.

[Ayman Habib]

Hi Ilan,

Yes, we did find some implementation issues with old muscle models, that's why they were deprecated and new more robust muscle models were introduced. As we move forward we'll be doing more development/testing and support on the new muscle models and various tools will be developed and tested against new muscle models only so we recommend moving away from the deprecated muscle models Please consult the webinar http://opensim.stanford.edu/support/eve ... html?id=47

for more details about the specific issues for deprecated muscle models and how they were addressed in the new muscle models.

Best regards,
-Ayman

[Ilan Eskinazi]

Thanks!

[Wendy Murray]

Hi Ayman,

Could you provide us with more specific info about the deprecation of the Schutte muscle model? Our lab watched the great webinar by Matt Millard, but it mostly discussed the new model as it compared to the deprecated Thelen model.

We discussed the differences in implementation of the Schutte model in SIMM and OpenSim with Pete Loan, and have compared model performance in both of those environments, but these conversations took place before the 3.0 update and the deprecation of the Schutte model.

Is there any more specific information you can provide?

Thanks, we appreciate it.

Best,

Wendy

[Ayman Habib]

Hi Wendy,

Deprecating the muscle model didn't change anything about its implementation, it was just a renaming to indicate to our users that we will not continue to develop and test this muscle model going forward. This was done either because it couldn't be retrofit to meet our new specification for how muscles should perform (e.g. Muscle interface, continuity requirements of curves , etc.) or because the deprecated implmentation was not a true match to the publication it referenced. The answers provided by the deprecated muscle models, however, should be identical to the pre 3.0 implementation (I believe we have test cases to verify that).

Please let me know if that answers your question about the deprecated muscle model(s).

All the best,
-Ayman

[Wendy Murray]

Hi Ayman,

Yes, thanks, this helps a lot.

Is there a way to use different normalized curves (e.g., force-length, force-velocity, tendon force-length) in the Millard model?

In the long run, we hope to switch to using the new muscle model, with all of the details provided in the paper and available in OpenSim. In the short run, we'd like to first establish whether or not the only difference that results from using the different muscle model is the new normalized curves. So, we'd like to compare results from our model using the deprecated Schutte model in OpenSim and results from our model using the Millard model, with the normalized curves that we currently use in Schutte.

We have reviewed this a bit, and the method to substitute one normalized curve for another is not immediately obvious to us.

Thanks again,

Wendy

[Ajay Seth]

One of the main reasons for deprecating the Schutte model, in addition to not referencing a published model, is that it allowed generic user-supplied splines (later Functions) for specifying muscle curves. In almost every case we reviewed the curves, including the built-in values, generated splined curves with unexpected behaviors. For example, the tendon force-length curve has negative derivatives in the toe region, which means negative stiffness at low force. These anomalies make using robust iterative solvers to find fiber-tendon equilibrium very difficult and they cannot be resolved to high precision. Matt Millard did extensive work to make sure the parametrization of the muscle curves do not produce numerical anomalies like unexpected wiggles or discontinuities. There is no such guarantee with generic user-specified Functions.

To modify the curves you can edit the properties/parameters of the individual curves. These
are explained in detail in muscle model code's documentation. Specifically, the details for the curves are described within the individual classes. See the Detailed Description sections for the following classes in Doxygen: ActiveForceLengthCurve, TendonForceLengthCurve, etc...
We hope these are helpful. Please let us know if you are able to achieve the range of behaviors you need with these parameters.

[Wendy Murray]

Hi Ajay,

Thanks for the details and direction for the documentation. We'll do some testing and keep you posted.

In the long run, we'd like to switch to the data-based curves described in the paper. It will be helpful to understand whether the factors you describe above are influencing our current results, as well as to appreciate the differences that result from the most up to date experimental data.

Best,

Wendy