Dear Thierry,
A blast from the past ;-)
The dimensionality here refers to the dimensions of the search space,
such that 1D for fMRI data would mean searching for an effect in space
through a line across the brain. This is what we use for example when
analysing ERP/ERF data from EEG/MEG where we search for an effect across
1-dimensional time from a single sensor (or averaged sensors). To use
this in SPM, you need to create 2D NIfTI images where the first
dimension is time (or space or whatever) and the second dimension is 1
(twisting the NIfTI specs a bit), see e.g.:
https://github.com/spm/spm12/blob/master/spm_eeg_convert2images.m#L190-L193
I presume this is what SPM1D does. There used to be a standalone
spm_1d.m function floating around but can't find it now (RFT in 1D is
fairly simple to implement).
That said, it doesn't seem to be what you are looking for as you are not
searching for an effect through time. If you have a single
neurophysiological time series and multiple behavioural predictors, you
can just use a GLM where the latter will form the design matrix:
https://github.com/spm/spm12/blob/master/spm_ancova.m
With multiple neurophysiological time series, you could use a
multivariate analysis such as CVA:
https://github.com/spm/spm12/blob/master/spm_cva.m
Or (and perhaps understanding better only now your setting) just fit a
GLM for each neurophysiological time series in a mass univariate sense;
you could use SPM for this by saving your data in multiple voxels with
arbitrary neighbourhood - just don't smooth in 'space' and use
Bonferroni instead of RFT.
As you mention, temporal non-sphericity has to be taken into account.
This would be an input in spm_ancova or SPM could estimate it with the
last approach, provided you have enough 'voxels'. If your inference is
at the group level, it might still not be an issue as the beta estimates
are unbiased.
Sorry for the convoluted answer and if I entirely misunderstood your
problem.
Best regards,
Guillaume.
On 03/06/2021 17:25, Thierry Chaminade wrote:
> Dear SPM users and experts,
>
> As a long-time user of SPM (I started with SPM96 and PET data), I've
> been told several times that the SPM statistical machinery could be used
> for all types of data... including 1D time-series. I'm currently trying
> to "hijack" SPM for such use, but fail miserably. I am turning to you
> for some insights and help.
>
> Simply, I have multiple1D time series (connectivity between areas) and a
> series of predictors (behavioral descriptors), with subjects and
> repetitions. I would like to know how significantly these
> behavioral predictors can "explain" the neurophysiological time-series.
> This is what SPM does for whole-brain analysis of fMRI (4D) time-series.
> But how can I use the same machinery (integrating time autocorrelations,
> not spatial dependencies) when the data is 1D (single value per time-point)?
>
> I thought a package named SPM1D
> <https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fspm1d.org%2F&data=04%7C01%7C%7C3cf705cc533a4f96edcf08d926ac3d84%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C637583343894585931%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=AbQj2MmgurTvtel%2Fmho5LjDduLHyc%2F9iZ%2Bx%2FEDl64vQ%3D&reserved=0> could
> do the trick, but it seems it is giving correlations through time, which
> is not what I am looking for. For example, given the connectivity
> between the IFG and the MPFC through time, I want to know how many
> behavioral time-series can explain the neurophysiological one ("betas").
>
> I started this project thinking it would be trivial, well, it is not.
> And as I don't have the (statistical, mathematical, programming,
> knowledge...) skills of the SPM community, I am asking you whether 1)
> this makes sense, 2) it is already documented somewhere, and 3) it can
> be implemented ?
>
> Thanks for any feedback that could be helping us!
>
> Best wishes, Thierry Chaminade.
>
>
--
Guillaume Flandin, PhD
Wellcome Centre for Human Neuroimaging
UCL Queen Square Institute of Neurology
London WC1N 3BG
|