Print

Print
Hi Cyril

Thank you so much for your help. I went through the article you had shared regarding implicit vs explicit modeling of baseline.


I decided to change the model specification and removed the 'silence ' condition from the model specification in the first level.
But i could not estimate the model and i kept getting the following error "error using spm est non sphericity".

I use an explicit mask in the first level and masking threshold of 0.8.

I also tried using masking threshold of -inf and removed the explicit mask. Spm did estimate the model but in the results all the voxels are now outside the brain.

Please could you suggest a solution to this.

Thank you so much

Best
Payal



On Sun, Nov 4, 2018 at 5:00 AM PERNET Cyril <[log in to unmask]> wrote:
Hi Payal

Yes this is correct. create that contrast do to your 2nd level one sample giving you the 1st result, and use that result as inclusive mask for your ANOVA that uses the three contrats you already have. As indicated before:

2nd level flexible factorial masking by the result of your one sample at
some liberal threshold (say p=0.001 uncorrected) that way you only look
at voxels with positive responses (ie > silence), boosting a bit your
stats because smaller volume is considered -- and yes 'song' is a single
factor with 3 levels

From: Payal Arya <[log in to unmask]>
Sent: 02 November 2018 21:31:59
To: PERNET Cyril
Cc: [log in to unmask]
Subject: Re: [SPM] second level analysis
 
Hi Cyrill

thanks for your help.

Thank you for the reference article as well.

If i understand correctly, to test if silence is greater than either of the song
1st level contrast song1+song2+song3 (>3*silence if you keep your design as is)

- I should create a new  t -contrast at first level for all the subjects using the following t-contrast
 (-3 1 1 1 ) (silence song1 song 2 song3)

I also want to look at the different regions activated in song 1 or song 2 or song 3
-would it be correct to use the 1st level con images of song1>sil ,song2>sil and song3>sil in the second level analysis.
the contrast would be difference between the song1>sil -song2>sil images?

thank you for your help

Best

payal


On Fri, Nov 2, 2018 at 10:40 AM cyril pernet <[log in to unmask]> wrote:
Hi Payal
> Hi SPM users
> I needed some help with second level analysis.
> My experimental paradigm includes:
> 15 subjects. Each subject is exposed to three kinds of songs in
> between silence periods.
> For the first level analysis at each subject level.
> I generated three  T-contrasts: Song1 > sil, song2>sil, song3>sil that
> resulted in three con images which i  will use in the second level
> analysis.
> But i am not sure how to go about it.
> What i would like to test at the group level is if there is a main
> effect of the song on the subjects brain and is it more than the
> silence. Which means is any of the stimulus (in this case song ) is
> activating brain regions more than the silence?

note that if you have a single run, then you do not need that contrast
by leaving silence implicit (see
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3896880/)

>
> And then specifically i would like to check if there are brain regions
> exclusively being activated when song 1, song 2 or song 3 is being
> played.
> I set up a flexible factorial design
> keeping first factor as subjects
> second factor as stimulus class (which includes three distinct types
> of songs which were played to  the subjects).
> I am not sure should i set up the three kinds of songs a s different
> factors or they count as different level of one factor?

I'd do the following given you want to see only 'activated areas'

1st level contrast song1+song2+song3 (>3*silence if you keep your design
as is)

2nd one sample t-test on this contrast = show me where this is active on
average

2nd level flexible factorial masking by the result of your one sample at
some liberal threshold (say p=0.001 uncorrected) that way you only look
at voxels with positive responses (ie > silence), boosting a bit your
stats because smaller volume is considered -- and yes 'song' is a single
factor with 3 levels

best

cyril



>
> Any help is appreciated.
>
> thank you so much
>
> Payal
>

--
Dr Cyril Pernet,
Senior Academic Fellow
Software Sustainability Institute Fellow
Neuroimaging Sciences

Centre for Clinical Brain Sciences
The University of Edinburgh
Chancellor's Building
Room GU426D
49 Little France Crescent
Edinburgh   EH16 4SB

[log in to unmask]
tel:  +44 (0)131 465 9530
http://www.sbirc.ed.ac.uk/cyril
http://www.ed.ac.uk/edinburgh-imaging


The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.