There is not yet any code in SPM for doing the most "correct" thing
(although I'm sporadically working on it). I tend to think in terms
of model accuracy and internal consistency, rather than correctness.
You could try the approach I mentioned previously, although you may
want to repeat the procedure by estimating the spatial normalisation
parameters from the 2nd time point (instead of the first) to ensure
that you still get the same results.
You could also run Dartel using all scans, but I suspect that this may
not give you the same amount of sensitivity as treating the data
longitudinally. On the positive side, this approach should not
introduce systematic biases that depend on whether the early or late
scan was used to do the spatial normalisation.
Although I have no evidence either way, I suspect that a TBM-like
approach may give greater sensitivity for longitudinal data. This
would involve estimating a measure of difference between the 1st and
2nd time points directly from the data, rather than trying to
independently segment them. The difference between the time points is
likely to be small relative to segmentation errors, so it may be more
accurate to try to directly determine the difference by directly
registering the scans together.
Best regards,
-John
On 11 January 2012 18:42, Kairys, Anson <[log in to unmask]> wrote:
> Thanks John,
>
> I am assuming it would be incorrect to create a template from all timepoints (as we expect differences between trx arms).
>
> Just to clarify, you are saying I should create flow fields from just timepoint 1, and then sequentially apply these flow fields to each timepoint for both GM and WM segments? E.g. u_rc1 from timepoint 1 applied to a GM segment in timepoint 2
>
> Is it incorrect to segment al ldata normally using "new segment" then creating flow fields and normalizing, instead of just segmenting timepoint 1?
>
> Thanks
>
> AK
>
> -----Original Message-----
> From: John Ashburner [mailto:[log in to unmask]]
> Sent: Wednesday, January 11, 2012 1:34 PM
> To: Kairys, Anson
> Cc: [log in to unmask]
> Subject: Re: [SPM] DARTEL template from baseline scan, applied to other timepoints
>
> Yes. That should be possible.
>
> 1) First of all, make sure that all your data are coregistered within subject.
> 2) Then do the segmenting and Dartel estimation on the first time point data.
> 3) Write spatially normalised versions of the first time point data using the estimated deformations.
> 4) Write spatially normalised versions of the second time point data using the estimated deformations.
> etc
>
>
> One thing to be careful of are systematic biases resulting from treating the first time point data differently from the others. See the recent commentaries in NeuroImage Volume 57, Issue 1 to see if these relate to your plans for the data.
>
> Best regards,
> John
>
> On 11 January 2012 18:21, Anson Kairys <[log in to unmask]> wrote:
>> Hi DARTEL users,
>>
>> I am trying to compare timepoints of drug trx (e.g. pre vs post treatment). I want to make the DARTEL template from their baseline scans, then use this template for normalization with the other timepoints. Is this possible without creating flow fields for all timepoints?
>>
>> thanks
> **********************************************************
> Electronic Mail is not secure, may not be read every day, and should not be used for urgent or sensitive issues
>
|