# enorm and dvars

Dear AFNI experts,

I’m sorry for asking again. Is it possible to calculate with AFNI the censor file of dvars values using a threshold of 0.5?

Furthermore, I would like to know what is the output that I can use as a regressor in the linear regression step, both in the case of enorm and in the case of dvars (if it is possible).

Any help is welcome!

Cheers,
Marina.

Hi AFNI experts,

I would like to calculate dvars and use a threshold of 0.5% to generate the censor file which contains 1s (indicating time points to be included)
and 0s (indicating time points to be excluded). In order to calculate dvars, I run the following command:

``````

``````

I saw that the values of the output are in a range of 15-20. Is it normal?

On the other hand, I think that I have to use 1d_tool.py to calculate the censor file that I want to use with 3dTproject. Is it correct? If so, how should I use this > function to calculate the censor file taking into account the values of dvars with a threshold of 0.5?

Thank you very much in advance!
Best wishes,
Marina.

Hi Marina,

I am not positive exactly how DVARS is applied in the field, but 3dTto1D has a few related methods:

1. dvars (or rms) : simply the root mean square of the first differences
2. srms (or cvar) : dvars/rms scaled down by the global mean
3. shift_srms : srms shifted by the global mean
The -help output has more details.

My guess is that you want srms, as it must be in [0,1]. But maybe there is a different scalar that you want, which would also scale into the 0.5 threshold.

There is currently no method to apply this in afni_proc.py, as those curves seemed to noisy to be robust. But if it is something you feel is important, I could consider adding something.

Regarding enorm (this time of the motion parameters), note that it is not generally used in the linear regression. The (de-meaned) motion parameters are used, and/or their derivatives. But enorm itself is generally just used for censoring and QC. Note that enorm would capture little that the motion derivatives would not.

The dvars regressor could be used in the regression (though not currently via afni_proc.py). In that case, scaling would not matter much.

Does that seem reasonable?

– rick

Hi Rick,

My guess is that you want srms, as it must be in [0,1]. But maybe there is a different scalar that you want, which would also
scale into the 0.5 threshold. There is currently no method to apply this in afni_proc.py, as those curves seemed to noisy to
be robust. But if it is something you feel is important, I could consider adding something.

Our initial idea was to do a censoring based on Power et al., 2012 and, particularly, based on intensity instead of the 6 motion parameters, and it would be interesting if AFNI had a way to do it (for example, censoring with dvars or srms), but using enorm is fine! I have also read here the explanations why it’s more appropriate to use enorm than fd (it is a calculation more proportional to the real movement), and for now, we will apply enorm to censoring because it seems very reasonable.

Regarding enorm (this time of the motion parameters), note that it is not generally used in the linear regression. The (de-meaned)
motion parameters are used, and/or their derivatives. But enorm itself is generally just used for censoring and QC. Note that
enorm would capture little that the motion derivatives would not.

It make sense!

The dvars regressor could be used in the regression (though not currently via afni_proc.py). In that case, scaling would not matter much.
Does that seem reasonable?

A doubt emerged related to this: Is it correct to use as a regressor a file that contains a value for each volume except for the first volume?
(Because the first value in the dvars file is always 0.)

Thanks again!

Best wishes,
Marina.

Hi Marina,

Note that we also have outlier censoring (you can do both), which is based on the outlier plots generated at the beginning of the proc script. That is based on the unregistered data, and it might be at least somewhat more logically similar to dvars. But of course, it is still different. See -regress_censor_outliers.

Regarding regressors that are 0 at the first volume, it should not make a big difference. That is true for the motion derivatives (first differences, actually), too. It does beg the question of whether that 0 should be at the beginning or the end. Nothing is perfect.

– rick

Okay Rick! Thank you very much for solving all the doubts!

Best wishes,
Marina.