No problem! Just to add that whichever strategy you choose, the extract ROI function in SPM should give you some protection from outlier voxels, as it is performing a principle component analysis within the region you have defined.
Best,
Peter.
> -----Original Message-----
> From: Mason, Liam [mailto:[log in to unmask]]
> Sent: 28 March 2014 13:34
> To: Zeidman, Peter; [log in to unmask]
> Subject: RE: VOI thresholding queries
>
> Hi,
>
> Thank you for the helpful response and the code for the % variance,
> much appreciated.
>
> I agree that the threshold is arbitrary but I would still say it is
> necessary - if I used unthresholded (or thresholded at P=1.0) to define
> my subject-level peak (for centre of the VOI), I might expect a high
> chance of that peak being a spike or other artifact.
>
>
> Bw,
> Liam
>
> ________________________________________
> From: Zeidman, Peter [[log in to unmask]]
> Sent: 28 March 2014 11:47
> To: Mason, Liam; [log in to unmask]
> Subject: RE: VOI thresholding queries
>
> Hi Liam,
> You're choosing the voxels to include for individual subjects' ROIs,
> but I think you're saying your inference will be at the group level.
>
> In which case, common practice is to just take all the voxels at the
> individual level corresponding to the group cluster, or alternatively,
> create single-subject rois (e.g. spheres) which vary in their centroid
> up to a fixed maximum distance from the group peak.
>
> As for excluding some voxels, e.g. using a t-statistic equivalent to p
> < 0.5 or p < 0.1, my personal opinion is this is quite arbitrary, as
> you would not necessarily expect a voxel to have significant activation
> at the single subject level, even if it's part of the group peak.
> Others may have different opinions on this...
>
> Best,
> Peter.
>
> > -----Original Message-----
> > From: SPM (Statistical Parametric Mapping)
> [mailto:[log in to unmask]]
> > On Behalf Of Mason, Liam
> > Sent: 28 March 2014 11:34
> > To: [log in to unmask]
> > Subject: Re: [SPM] VOI thresholding queries
> >
> > Hi,
> >
> > My question refers to selecting VOIs at single subject level, that
> lie
> > near to a set of co-ordinates already identified at group level. I
> > also constrain this within an anatomical structure.
> >
> > Any further input from the list greatly appreciated.
> >
> > Thanks,
> > Liam
> >
> > ________________________________________
> > From: Zeidman, Peter [[log in to unmask]]
> > Sent: 28 March 2014 09:54
> > To: Mason, Liam; [log in to unmask]
> > Subject: RE: VOI thresholding queries
> >
> > Hi Liam,
> > Could you clarify whether you want to make inferences at the single
> > subject or group level? And how is your ROI defined - is it based on
> a
> > group activation? Or single subject activation? Or anatomical
> > structure?
> >
> > Note that if you're making inferences at the group level, you would
> > not necessarily expect significance at the single subject level.
> >
> > The variance explained gives you a sense of how much of the signal
> you
> > have captured in your principle component (ROI) - it might be that
> > you are perfectly modelling 5% of the signal, but there are more
> > substantial processes going on that are not included. You'll
> generally
> > find it varies as a function of the size of your ROI. It's shown in
> > the GUI, but I'm not sure if it's stored.
> >
> > Best,
> > Peter.
> >
> > > -----Original Message-----
> > > From: SPM (Statistical Parametric Mapping)
> > [mailto:[log in to unmask]]
> > > On Behalf Of Mason, Liam
> > > Sent: 27 March 2014 15:14
> > > To: [log in to unmask]
> > > Subject: [SPM] FW: VOI thresholding queries
> > >
> > > Hi there,
> > >
> > > Can anyone help with the following VOI queries?
> > >
> > > First, I am having difficulty deciding on a threshold for
> individual
> > > subject VOI - is there a trade-off in terms of 'quantity' (number
> of
> > > voxels included) and their 'quality'? Where poor quality might mean
> > > including voxels with spurious or artifactual activations.
> > > Practicaly, thresholding with P=.1 fails to find any voxels for a
> > > substantial number of subjects, so I will need to relax this. P=.5
> > (or
> > > even P=1) is commonly used but seems too relaxed(?) For subjects
> > where
> > > a robust/non-artifactual peak is identified at e.g. P=.1, won't
> > > dropping this threshold further increase the chance of finding an
> > > artifactual peak?
> > >
> > > Second, is the amount of variance explained by the VOI a good
> > measure
> > > of 'quality'? In which variable would this value be stored for each
> > > subject (can't find it in any of the fields of xY returned by
> > > spm_regions function)
> > >
> > > Any help would be much appreciated!
> > >
> > > Many thanks,
> > > Liam
>
|