Hello,

I have a an experimental setup where I have a baseline recording, a period of seizures, and then a period after we inject an anti-epileptic drug. I have produced the PSDs and made the z-scores for both the seizure and post-drug period.

My goal is to compare the PSDs of the seizure period and the post injection period to the baseline, hoping the drug will bring the z-score closer to 0 after we inject the medicine. I did have a few questions.

The z-scores I am getting seem extremely high (ie. in the hundreds). z-scores seem to be much lower than that in other papers I have read. Can this be normal?

How can I do a t-test of the z-scores I have for both seizure and after-drug conditions with the baseline? Again I would hope to see that the seizure condition is extremely different than the baseline, while the condition after drug gets closer to baseline.

Please let me know if you have any ideas or tips, thank you for your help!

It depends how you compute these Z-scores. Can you describe what you did?

Please include some screen captures of your database explorer showing the files you have been using (on Windows and Linux, you can press ALT+PrintScreen and then CTRL+V directly in this forum)

How can I do a t-test of the z-scores I have for both seizure and after-drug conditions with the baseline? Again I would hope to see that the seizure condition is extremely different than the baseline, while the condition after drug gets closer to baseline.

I'm not sure you need any normalization before the statistical analysis.

If you have enough repetitions of all the conditions (baseline and seizures), compute the PSD for each trial, then use the Process2 tab to test for seizure vs. baseline, separately for the two conditions (with and without the drug). Other option, test directly seizure/drug vs. seizure no drug.

https://neuroimage.usc.edu/brainstorm/Tutorials/Statistics#Example_2:_Permutation_t-test

@Sylvain Does it make sense?

Thank you for your answer! I will describe exactly what I did, so I apologize for the long message.

First off, here are my 2 EEG recordings, first is 30 minutes of baseline, and then 30 minutes of seizure activity:

EEG_All_2019_2_11_1_30min_basline.tif (214.5 KB)

EEG_All_2019_2_11_2_30min_seizure.tif (412.0 KB)

I then compute the PSD for both recordings using the default setup and with a 2s window, which gives me these PSDs (first PSD is for the baseline and the second is for the seizures):

EEG_PSD_2019_2_11_1_30min_basline.tif (108.2 KB)

EEG_PSD_2019_2_11_2_30min_seizure_PSD_1799_2000ms_Power_EEG.tif (107.7 KB)

Once I have the two PSDs, I perform a z-score transformation, and that is where the z-value turns out in the hundreds, or in this case even thousands..

EEG_PSD_2019_2_11_2_30min_seizure_PSD_1799_2000ms_Power_EEG___zscore.tif (111.5 KB)

From my (limited) understanding of the z-score, it should show me how much the seizure condition varies from the baseline, so a high z-score would be good in this case, but should it be this high?? I have never seen a z-score higher than 10 in the papers I have read.

I am not sure using the z-score is a valid way to compare the conditions, so please let me know what you think!

Again, thank you for your help and I hope to hear from you soon

One more question: when I try to do a t-test using my PSD graphs, I get the following error message: "You are testing the power values, while a more standard analysis is to test the magnitude (ie. sqrt(power)).

Option 1: recompute time-frequency maps using the option "measure: magnitude".

Option 2: run the process "extract > measure from complex values", with option magnitude."

My issue is that I cannot figure out how to do either of these options. I cannot find these computing options when I do my PSD graphs.

Thank you, and sorry for all the questions

First, there is a process under pre processing called apply Matlab command. You can use this to apply sqrt() to the power spectrum.

Regarding your z-scores, it sounds like you're doing a z test (which is not the same). Is this across a group, or a single subject across trials? I'm just wondering how you get the variances. With your seizure psd being on the order of 100 times larger than the baseline (10 times the amplitude), it wouldn't surprise me too much to have a z test value of 100. However, your z-score graph looks oddly similar to your seizure graph (just rescaled). Did you use a single variance value?

I perform a z-score transformation, and that is where the z-value turns out in the hundreds, or in this case even thousands..

How do you compute this Z-score?

The idea of a Z-score transformation is to normalize your values with the standard deviation across multiple repetitions of the same measure. In what you describe here, you don't have any repeated measure for your baseline but only one PSD value per sensor and per frequency, therefore you cannot estimate a standard deviation...

The Z-score transformation described in the source estimation tutorial (https://neuroimage.usc.edu/brainstorm/Tutorials/SourceEstimation#Z-score) is done in time domain, not frequency domain. It considers that the time points over a baseline are repeated measures, and we establish a mean and standard deviation for the "baseline value" across time using all the time points in the baseline.

In your case, you don't have the time dimension anymore in these PSD files, you have only one value per frequency. And you cannot consider that the values obtained from the different frequencies are repeated measures of the same thing, they are not exchangeable. If you want multiple repetitions, you need multiple PSD files.

Otherwise, what you can do is for instance to subtract your two files (PSD seizure - PSD baseline), or test for differences with a non-parametric paired t-test across all your patients (or repetitions of the same measures).

Great! Thank you very much for the clarifications, I think I figured it out thanks to your help. Thanks for the support, I am very satisfied with Brainstorm!

Remi J