CMC activations differ with same data

[Andrew Lewis]

Hi there,
Can anyone explain why I seem to be getting different muscle activations from CMC when I run exactly the same data twice?

I have been running normal walking data through the CMC tool using the gait2392 model (with locked subtalar & MTP joints). My intention is to compare the differences in simulated CMC muscle activation with changing muscle strength (ie altering the <max_isometric_force> values in the model).

I got some unexpected changes in the muscle activations with a slightly stronger model (see CMC-Models03-&-04.jpg - black line is recorded EMG). Most of the activations were slightly reduced as you would expect with a stronger model but there were certain muscles with higher activations at the end of the gait cycle (notably glut_med3, bifemsh, iliacus, psoas, peri, vas_med, vas_lat, tib_ant, per_tert, ext_dig, ext_hal). The kinematics and residuals for both CMC datasets were effectively identical.

I ran some further simulations to try and isolate the cause of these muscle activation "end effects". If you run the same data through CMC twice, you understandably get exactly the same results (see CMC-Models03-repeat1&2.jpg) but if you simply change the name of the model file (ie the <model_file> xml tag) and change nothing else and re-run the CMC then similar "end effect" altered muscle activations are obtained (see CMC-Models03-&-03Renamed.jpg). Once again the resulting kinematics and residuals for both CMC datasets were effectively identical.

This change in muscle activation for essentially the same data seems very baffling to me! Any help in explaining when is going on here would be greatly appreciated.
Thanks,
Andy

[Ursula Trinler]

Dear OpenSim community,

I have experienced the same issue. I've run the CMC pipeline 7 times for the same trial with the same input variables and got different results for every try. Threfore, I question myself now if this is normal or if I should expect exactly the same results when running the exact same data a second time with CMC? Anyone else experienced the same?

I've attached three graphs showing the simetendinosus, rectus femoris and gastroc medials during normal gait. X-axis in normalised to 100% GC.

Best,
Ursula

[Ursula Trinler]

Hi there,

I still receive this variability in my muscle force output data and I'm not sure if this is normal or not. Does anyone has received same results? Is it normal that CMC results differ slightly from run to run with same input data? Or should I receive exactly the same results every time I run the CMC pipeline with the same experimental data?

Best,
Ursula

[jimmy d]

Hi Ursula,

No, this is not what you should expect. Running cmc repeatedly with the same inputs will give the same result every time.

Are you running the analysis in the gui or through scripting?

Cmc can change the kinematics slightly, I would check that you are not using these new kinematics as inputs.

Cheers,
-james

[Ursula Trinler]

Hi James,

Thank you for your quick reply! I have directly checked my desired kinematics but they are all the same (I use the RRA kinematics as input). Also, I have double checked the GRF data and they are all the same as well.
Other inputs are as follows: the gait2392_CMC_Task.xml file as the tracking tasks file, and the actuator constraints file gait2392_CMC_ControlConstraints.xml.

I've run the trials separately and have done new folders/paths for them as well. What I have not done is rename the files.

Best,
Ursula

[Ursula Trinler]

...sorry, yes I run the analysis through the GUI.

Best,
Ursula

[jimmy d]

Hi Ursula,

Could you confirm if the issue was occurring due to different RRA models being used in CMC. The different masses and kinematics would result in slightly different muscle activations/forces

[Ursula Trinler]

Hi James,

Sorry for the late response!
Yes indeed, it does happen if I use differnt RRA models but as well as if I use the same model.

Best
Ursula

[Andrew Lewis]

Hello,
I still seem to be getting this problem. More recently I have been trying to eliminate mid-stance tib_ant activation in my CMC data "caused" by ankle_angle acceleration "ringing" (see https://simtk.org/forums/viewtopic.php? ... 476#p12476) and have been getting different activations when running CMC with the same settings and on the same data.

This came about as I was re-running CMC on a colleague's data-set of normal walking (using all of their original settings) so see if I would get the same results. I didn't but I thought this may have come about from different versions of OpenSim or differences in running CMC through the GUI or CMD or Matlab. I had run this first CMC through the OpenSim GUI.

So I re-ran the CMC 3 times, once using CMD, once using Matlab, and for a second time using the GUI, but importantly all three using exactly the same settings and data (the only thing I changed was the output directory for where the results were saved). All four CMC runs gave me slightly different activations - see Fig.1 which shows the input (RRA) and output (CMC) ankle_angle acceleration (dudt) and corresponding tib_ant activations.

It appears therefore that the CMC algorithm does not produce repeatable results (!???). Surely this can't be correct! Is anyone able to explain why this is happening?
Many thanks,
Andy