I am currently looking into verifying 20 subjects models from gait, sit-stand, and stairs. I have collected 16 channels of EMG on selected lower limb muscles synchronised with my motion capture data.
I would like to do a validation of the muscle recruitment solver I’m using (Polynomial with a Qp penalty), by comparing Pmet with EMG activity. One way of doing this is to define the EMG and metabolic output as active or inactive giving a binary output (1’s and 0’s). But I’d like to compare the two in a little more detail.
Has anyone looked at comparing EMG with modelled muscle outputs and what stats did you use?
That is a very good question, and unfortunately I do not know the answer. But I hope to find out in a few months when our new PhD student on validation begins his work. He will be looking into how we can develop a simulation of EMG in AnyBody, so that we have a direct measure to compare.
Any input to this process from the community will be most welcome.
thanks for you reply, I will shortly be working on a mini validation project for this EMG vs Pmet comparison.
My thoughts would be going towards getting MVC values for your given muscles, then using RMS of your signals to calculate a % effort during your functional trials (gait). Are you still of the opinion that Pmet is the better output to compare the EMG against?
I believe that metabolism is better than muscle activation as a target measure, but perhaps not in the form currently implemented in AnyBody.
There is good experimental evidence that, for static cases, EMG is propotional to muscle activation as well as metabolism. For dynamic cases we know that muscle activation in not a good measure, but we do not know whether metabolism is a good measure.
The metabolism computed by AnyBody is very primitive and fails to capture some of the sources of combustion in the muscle, and for static cases, AnyBody computes the metabolism to zero, which is clearly wrong.
So I would propose to implement a better calculation of metabolism and compare that to the EMG you measure. The better calculation is not difficult to implement directly into AnyScript or in Matlab based on an output file for the muscle in question.
I would like to make some more general points on validation with the help of EMG. First thing to realize that validation like this is really tricky business.
My experience is that if you want to validate muscle recruitment there are two thing you can look at. The first thing is timing. And the second is amplitude, also with respect to other muscles.
The timing is best validated by analyzing a cyclic movement i.e. gait or cycling. Sit-to-stand is in that respect not the best choice, because for example the quadriceps muscles will be active almost all the time in that movement.
Validating for amplitude is difficult, because basically you compare cats and dogs. Anyhow first of all some normalization of the EMG is necessary (Taking MVC is a good possibility, but taking good MVC is not always easy). In my opinion the best way to validate the amplitude in a way is to make use of so called trend measurements. For example if you take the sit-to-stand, where you could measure the movement systematically for different starting heights. This might change the EMG ratio between the quadriceps and hamstrings. If the model reacts in the same way as in reality you have a strong point for the validity of the model. This kind of the validity is in my opinion really important if you want to do simulations with the model where you would like to investigate the influence of a certain parameter.