Re: sparse temporal sampling

[Jonathan Peelle <peelle@xxxxxxxxx>]

Hi Theresa,

Just to add to Matt's very sound advice: there are also a number of
posts on the SPM email list [1] that address some of the issues raised
in detail.  If you want to use a canonical HRF (Matt's point 3), then
it's important to get the microtime onset (also known as SPM.T0) and
microtime resolution (also known as SPM.T) set correctly.  There are
many posts on this, including, e.g.:

https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;27ab2eb2.1010

which may be helpful to you.  Good luck!

Best regards,
Jonathan

[1] https://www.jiscmail.ac.uk/cgi-bin/webadmin?REPORT&z=4&1=SPM&L=SPM



-- 
Dr. Jonathan Peelle
Department of Neurology
University of Pennsylvania
3 West Gates
3400 Spruce Street
Philadelphia, PA 19104
USA
http://jonathanpeelle.net/



On Fri, Jul 15, 2011 at 4:07 AM, Matt Davis
<Matt.Davis@xxxxxxxxxxxxxxxxx> wrote:
> Hi Theresa
>
> Here's a few suggestions for analysis of sparse fMRI data. I don't know of
> a single methods paper that explains everything. You might to read a paper
> like Hall et al (1999, Human Brain Mapping).
> Most of the preprocessing and analysis steps are as usual for fMRI data. A
> few points of difference between sparse and standard fMRI analysis are as
> follows:
>
> 1. Don't perform slice acquisition time correction. Given the 15s delay
> between scans, interpolation between successive scans is ineffective and
> will damage your data.
>
> 2. At the analysis stage, I often use an FIR basis set of duration 15
> seconds (equivalent to your scan repetition rate), with a single time bin.
> This models the data as a box-car covering the single scan  following each
> condition. This is a simpler model to set up than the typical haemodynamic
> response model, but usually effective for sparse data.
>
> 3. If you do want to use an HRF model - for instance, if you have
> differences in the timing of trials within your silent period - then you
> also need to take care over the specification of SPM.xBF.T and SPM.xBF.T0
> in your model.
>
> 4. You need to take care in specifying the low-pass filter and AR(1)
> parameters in your model. I often turn these off entirely since my goal is
> to do second level, group analyses rather than computing single subject
> statistics. The reason for concern is that scan-to-scan auto-correlation
> is greatly reduced with a long TR design, and the slow changes in
> activation between conditions are at a much lower frequency in sparse
> designs.
>
> Good luck!
>
> Matt
>
>
>
> On 15/07/2011 05:06, "AUDITORY automatic digest system"
> <LISTSERV@xxxxxxxxxxxxxxx> wrote:
>>
>>Date:    Thu, 14 Jul 2011 13:44:23 -0700
>>From:    theresa veltri <theresaveltri@xxxxxxxxxxx>
>>Subject: sparse temporal sampling
>>
>>Hi, I am currently starting some fMRI analysis in SPM8. Unfortunately I
>>am
>>unsure how to account for the sparse temporal sampling design that was
>>used.
>>For example, each trial was 15s, but TR is only 4s. Does anyone have expe
>>rience with specifying such parameters in SPM? Does anyone have any
>>referen=
>>ces the could suggest? any help would be appreciated.
>>Best,
>>
>>Theresa Veltri
>>MSc in Cognitive and Computational Neuroscience
>>University of Sheffield, UK
>