Baseline or spectrum normalization?

1

Hi Francois and team,

I am doing a research to compare the time frequency power for 8 auditory stimuli. In our experimental design, I have 30 subjects and all of them will listen to all auditory stimuli ( 3 minutes each) while spontaneous EEG and MEG are being recorded. Prior to starting of the experiment a 3 minutes of resting state MEG/EEG will be recorded. My questions are:

  1. Which time frequency normalization method is the most suitable for my experimental design? Baseline or spectrum normalization?

  2. If baseline, can I use the resting state ,where the subjects received no auditory stimuli, that is recorded prior to stimuli as the baseline?

  3. If spectrum normalization, should I normalize before or after i group them in frequency bands (delta, theta, beta, etc...)?

thank you very much for your feedbacks.
syairah

2

Hi!
If I where you, I would use event-related spectral perturbation normalization type, comparing the eeg with stimuli with baseline (resting eeg prior to stimuli),

I think you should use that normalization as final step in your analysis, actually when you compute time-frequency analysis there is an option to get the output in eeg bands,

event related perturbation is a baseline type normalization, you just have to select your baseline and software will compare it with your event related eeg data, output comes in percent %.

check the tutorial:

https://neuroimage.usc.edu/brainstorm/Tutorials/TimeFrequency

3

TQ Tourette for your opinion. Very appreciate it. We only has 1 resting state recording that was recorded in the first place before continued with other 8 stimuli. so, the same resting state will be used for other stimuli, if for baseline normalization method.

peace,
syairah

4

Which time frequency normalization method is the most suitable for my experimental design? Baseline or spectrum normalization?

The recommended approach is to normalize the post-stimulus time-frequency maps with respect to a baseline, as illustrated in the tutorials. The option "Spectrum normalization" is mostly to be used for display purposes, not for group analysis.
https://neuroimage.usc.edu/brainstorm/Tutorials/TimeFrequency#Normalized_time-frequency_maps

I am doing a research to compare the time frequency power for 8 auditory stimuli. In our experimental design, I have 30 subjects and all of them will listen to all auditory stimuli ( 3 minutes each)

Are you planning to study particular events during these 3 minutes?
If you are simply expecting to compare the power in various frequency band between different conditions over the entire 3min segment, maybe you should use simpler measures which do not give you time-resolved results, such as the spectrograms you can obtain with the PSD process. The resting state tutorials illustrate how to use such measures on relatively long recordings:
https://neuroimage.usc.edu/brainstorm/Tutorials/RestingOmega#Power_maps

If baseline, can I use the resting state ,where the subjects received no auditory stimuli, that is recorded prior to stimuli as the baseline?

Yes, you could probably normalize your experimental data relatively to the power during the resting period. Use the Process2 tab for that.

If spectrum normalization, should I normalize before or after i group them in frequency bands (delta, theta, beta, etc...)?

Probably after, I guess it would be less noisy, but I'm not sure.

5

Thank you Francois, for your comment and reply.Very appreciate it

Syairah

6

Hi Francois,
May I know why spectrum normalization is only for display and not for group analysis?

Thank you for your reply

7

This option "1/f compensation" multiplies the power at each frequency with the frequency itself, it makes the figures more readable but doesn't normalize the maps with respect with a baseline, and doesn't bring all the participants to the same range of power values.