Z-score and spatial smoothing

Discussions
1

Dear BrainStormers,

A question concerning the best order to do normalization and smoothing for group analysis of ERFs in MEG data.

Concerning smoothing, it makes sense to me to perform it at the single-subject level before projecting the data on the template anatomy for group analysis (as recommended here: http://neuroimage.usc.edu/forums/showthread.php?1409-spatial-smoothing).

I would therefore do the following:

  •   average trials relative to each condition within each subject

  •   compute source reconstruction

  •   compute absolute value of average source time series

  •   z-score at each voxel wrt baseline

  •   apply smoothing (on signed values, not absolute)

  •   project smoothed z-score av sources to default anatomy

  •   ready for averaging/stats across subjects

However, McGill workflow suggests to do smoothing after projection: http://neuroimage.usc.edu/brainstorm/Tutorials/WorkflowGuide. Is there a specific reason for that?

Thanks for your feedback,

Marco

–
Marco Buiatti

2

Hi Marco,

We haven’t evaluated the difference between the two approaches.
Smoothing after projecting ensures that all the maps are smoothed in the same way.
Smoothing before projecting is maybe more anatomically correct.

You can compare the results you get the two methods and use the one you feel the more comfortable with.

Francois

3

Dear BrainStormers,

I have noticed that now you advice NOT to compute the absolute value of the source times series before performing group analysis at the source level:
http://neuroimage.usc.edu/brainstorm/Tutorials/WorkflowGuide

First question: why did you change advice? Is it because of the odd effects of taking the absolute value as you described here: http://neuroimage.usc.edu/brainstorm/Tutorials/Difference, or did you change something in the source reconstruction algorithm?

Second question: This raises some concerns about spatial smoothing at the subject level. As Dimitrios Pantazis was mentioning in another thread (http://neuroimage.usc.edu/forums/showthread.php?1409-spatial-smoothing):
" Due to the folding of the cortical surface, neighboring voxels in MEG maps will have very different activation values and opposite sides of sulci will have opposite signs (in the case of orientation constrained reconstruction). Thus, when working on MEG orientation-constrained cortical maps, you should never smooth the signed data because neighboring strong positive and negative values will jointly become zero. You should only smooth absolute value maps or power maps."

What is your advice on smoothing then? To be safe, should it be done at the very end, i.e. on the source map of the absolute value of the difference between two conditions?

Thanks for the clarification,

Marco

4

Hi Marco,

I have noticed that now you advice NOT to compute the absolute value of the source times series before performing group analysis at the source level:
http://neuroimage.usc.edu/brainstorm.../WorkflowGuide

This is some internal documentation for McGill students (as the title says) and it is not up to date.
You should refer to this tutorial instead (still being written, but should be mostly finished at the end of May):
http://neuroimage.usc.edu/brainstorm/Tutorials/Workflows

Short answer: do not rectify before Z-score, do not rectify for subject-level analysis. The sign is meaningful for a given anatomy.
But rectify before group analysis, or rectify the difference (|A-B|) because the sign is not meaningful between subjects.

First question: why did you change advice? did you change something in the source reconstruction algorithm?

Nothing changed in the source reconstruction. There were simply no recommendations for group analysis before.
These example pipelines in the "workflows" page were not part of the documentation, it was up to the experimenter to decide what to do.
We're now pushing the documentation a step further to help non-experts make the best default choices.

Second question: This raises some concerns about spatial smoothing at the subject level.

http://neuroimage.usc.edu/brainstorm/Tutorials/Workflows#Average:_Group_analysis

Note that this may change again soon.
Read this page again when the mention "[TUTORIAL UNDER DEVELOPMENT: NOT READY FOR PUBLIC USE]" is not on the page anymore.

Cheers,
Francois

1 Like
5

Thanks Francois,

From now on I’ll keep as reference the tutorial pages only.

For the absolute value, I understand your explanation:

[QUOTE=Francois;10395]
Short answer: do not rectify before Z-score, do not rectify for subject-level analysis. The sign is meaningful for a given anatomy.
But rectify before group analysis, or rectify the difference (|A-B|) because the sign is not meaningful between subjects.
[/QUOTE]

and I retain that I should first z-score at the subject level and then rectify the single-subject source signals before group analysis.

For spatial smoothing, I retain that the two options (smoothing before or after projecting) are still both valid, until further notice.

Best,

Marco