Dear Vasilis,
Xin's paper and your investigations are spot-on. The difference you
observe is whether the psychological variable is mean-centred or not:
https://github.com/spm/spm12/blob/r7487/spm_peb_ppi.m#L439
Usually, PPI is used in a multifactorial design setting where one
experimental factor is used to drive responses in the seed region (the
physiological variable), while a second experimental factor (the
psychological variable) is used to modulate the influence of the seed
region on target regions. In doing so, the psychological factor is
orthogonal to the factor inducing activations. With a balanced design,
the psychological variable, obtained by contrasting the levels of one
experimental factor, will have a zero mean and the issue of
mean-centring goes away. See, e.g., the example in chapter 36 of the SPM
manual.
Best regards,
Guillaume.
On 28/11/2019 17:40, Ioakeimidis, Vasileios wrote:
> Thanks a lot Xin, that’s a good paper. Not sure if I’m at a level to
> fully comprehend it though.
>
>
>
> From what I understand in the paper, SPM8 does not mean centre the
> psychological variable PPI.P, and this introduces spurious results. On
> the other hand the latest versions in SPM12 have fixed this issue.
>
>
>
> Please see below comparative graphs of PPI.ppi variables resulting from
> SPM8 and SPM12, as well as the PPI.P from SPM8, SPM12 and by hand mean
> centering of the SPM8.
>
>
>
>
>
> Indeed when I mean centre the psychological variable by hand in SPM8, I
> get an almost identical line as the one from SPM12. However I’m still
> confused looking at the PPI.ppi SPM12 line. Are the noisy areas that
> belong to the conditions of no interest in the purple line ‘normal’ or
> should it be flat as the orange line?
>
>
>
> Many thanks,
>
>
>
> Vasilis
>
>
>
> *From:*Xin Di <[log in to unmask]>
> *Sent:* 28 November 2019 13:11
> *To:* Ioakeimidis, Vasileios <[log in to unmask]>
> *Subject:* Re: [SPM] PPI different results SPM8 vs SPM12
>
>
>
> *CAUTION:*This email originated from outside of the organisation. Do not
> click links or open attachments unless you recognise the sender and
> believe the content to be safe.
>
>
>
> Hi, Vasilis,
>
>
>
> Please check this paper
>
> https://onlinelibrary.wiley.com/doi/abs/10.1002/hbm.23413
> <https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fonlinelibrary.wiley.com%2Fdoi%2Fabs%2F10.1002%2Fhbm.23413&data=02%7C01%7C%7C2a362f43dd4a43ec214d08d7742a177f%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C637105596444489898&sdata=dYlln%2B4hcwut4EO0vJX9bBACzS390Hz0Mu7UjMzBNTk%3D&reserved=0>
>
>
>
> Best regards,
>
> Xin
>
>
>
>
>
> On Thu, Nov 28, 2019, 7:45 AM Vasilis Ioak
> <[log in to unmask]
> <mailto:[log in to unmask]>> wrote:
>
> Dear SPM users and PPI enthusiasts,
>
> I've noticed that running the PPI module with SPM8 and SPM12 gives
> completely different results.
>
> Few words on my design:
>
> I have an n-back paradigm with blocks of ['X' 'One-back' 'Two-back'
> 'Three-back'], and these are modeled in the fist-level SPM.mat of
> each subject. I am interested in the negative and positive
> modulation of two-back on the left and right DLPFC, as VOIs. Thus,
> when specifying my PPI for each subject I use the contrast weights
> vector [3 1 1] to tell spm I'm interested in the two-back only.
>
> Then, from the variables [PPI.P PPI.Y PPI.ppi] I'm interested in the
> negative and positive PPI interactions with my VOI, so when viewing
> the PPI results I define the contrasts [0 0 -1] and [0 0 1], for
> negative and positive interactions respectively. Following the
> subject-level PPI estimation, I create a two flexible factorial
> models, one for the left DLPFC and one for the right, with factors
> 1) Group (controls - patients) and 2) PPI interaction (negative -
> positive). When I do these steps with PPIs created with SPM8 and
> SPM12, I noticed completely different results from the 2nd-level
> analysis.
>
> The issue:
>
> I've inspected the subject-level PPIs from SPM8 and SPM12 and
> spotted that the variable PPI.P have 0s assigned for every scan
> except where the two-back is supposed to be, as expected, since spm
> was instructed to ignore the other blocks. However, the same PPI
> specification and estimation with SPM12 does not produce PPI.P with
> 0s in the blocks of no interest, instead it assigns some random
> values at the beginning at the first X block and then a stable value
> (different from 0) up to where the two-back begins. The PPI.Y
> variables are the same from both SPM versions, but since PPI.P are
> different, PPI.ppi are also different. In the end, the within and
> between group results have no relationship with each other.
>
> Has anyone else encountered this issue? Is it normal that with SPM12
> I don't get 0s for the PPI.P corresponding to the blocks of no
> interest? If not, how could I correct this and calculate the correct
> PPI.ppi?
>
> Many thanks!
>
> Vasilis
>
--
Guillaume Flandin, PhD
Wellcome Centre for Human Neuroimaging
UCL Queen Square Institute of Neurology
London WC1N 3BG
|