Hi,
An alternate approach is to "repair" the data in the bad scans, where repair
means to replace those scans with something that is not an artifact. Two
possible repairs are to replace the offending scans with the mean scan of
the data series, or for single bad volumes, to replace the offending scan with
the average of the before and after volumes. Software to do the repairs is
available at the SPM Extensions website, called ArtifactRepair. The software
has a convenient GUI to pick and choose the bad scans. We have found it
pretty useful for analyzing fMRI scan data from children.
The repairs will affect the SPM estimation process. One approach is to
omit those scans (which is where you are already, and as stated, there
is a bunch of hand-labor), and the other approach is just run SPM without
omitting those scans. The latter approach is easy...it slightly biases the
contrasts to be small, but in practice this does not seem noticeable if the
repairs are done on only 5-10% of the scans. Generally, this method might be
termed replacing large "artifact noise" with smaller "repair noise".
Good luck,
- Paul
On Mon, 9 Oct 2006 12:13:08 +0100, Helmut Laufs <[log in to unmask]> wrote:
>Hi Daniel,
>
>to extend what Cyril has written: you would need [at least] one regressor
>per image volume you want to 'model out', because a single block regressor
>would not
>be sufficient as this would assume motion affecting all volumes in the same
>way. Afraim
>Salek Haddadi once introduced me to an approach ("scan nulling") where he
>modelled - per 'bad volume' - three regressors including the preceeding and
>subsequent scan (I think).
>
>The advantage over 'erasing' the scans is beyond saving you to recalculate
>your onsets, as it also makes sure the HiPass filtering remains accurate
>etc. I think there was a previous discussion related to this on the list.
>
>Best wishes,
>
>Helmut
>
>
>
>
>----- Original Message -----
>From: "cyril pernet" <[log in to unmask]>
>Sent: Sunday, October 08, 2006 11:08 AM
>Subject: Re: Motion during task
>
>
>> Hi Daniel
>>
>>>Hi all,
>>> I have a question about accounting for motion in the scanner. We scan
>>>many children with ADHD, and they tend to move around, and occasionally
>>>there is a period of about 10 seconds of fidgeting in the middle of a run.
>>>In the past, we have removed these volumes and adjusted the timing files,
>>>but I was wondering - would it work to create a block regressor to cover
>>>the
>>>period of motion, and assign it a zero in the model (and remove the events
>>>during that period from the timing file), so that SPM "ignores" the data
>>>from that period? Or would keeping the motion volumes in the model still
>>>affect the results somehow? It would be convenient to use this technique,
>>>because I was also thinking of using the same technique for the infrequent
>>>times when a child falls asleep in the scanner. Does this technique sound
>>>dodgy? Thanks!
>>>Daniel
>>>
>> That sounds more reasonable to me not to remove the bad volumes and model
>> the full time series.
>> You can "nullify" the bad scans by adding an extra regressor to your
>> design matrix with zeros for all your scans but 1 for the bad ones.
>> I did use that to nullify a dummy volume, i.e. there was a missing volume
>> in a series that I substituted by the mean of adjacent volumes and next
>> nullify this way (thanks to Tom Nichols for the trick)
>>
>> I guess this is the same idea for you.
>> Hope this helps
>>
>> Best
>> Cyril
>>
>===========================================================
==============
|