Dear Tang Le,
Following Torben's advice, if you want concatenate fMRI sessions, you
might want to model session effects and high-pass filter within session:
SPM12's function spm_fmri_concatenate.m was written to help you doing
this easily.
https://en.wikibooks.org/wiki/SPM/Concatenation
Best regards,
Guillaume.
On 12/12/16 08:22, Torben Lund wrote:
> it could be due to boarder effects between the different sessions, when
> you concatenate the different runs. These can be reduced using a run
> specific highpass filter and a baseline, which is what you will get if
> you use a multi session 1st-level design, but you could also create the
> HP-regressors manually using spm_filter.
>
> Best
> Torben
>
>
>
>
>
>> Den 9. dec. 2016 kl. 22.07 skrev Thang Le <[log in to unmask]
>> <mailto:[log in to unmask]>>:
>>
>> Hi Torben,
>>
>> Thank you for the suggestion. Unfortunately, we did not record any
>> physio data for this dataset. This is an event-related design.
>> Previously when analyzing certain events separately, I did not see
>> this pattern of artifact. Only when the entire trial length is treated
>> as a continuous event does this activation become visible. Regarding
>> your last question, I did indeed apply the 128s high-pass filter.
>>
>> Thang Le
>>
>> On Thu, Dec 8, 2016 at 2:56 PM, Torben Lund <[log in to unmask]
>> <mailto:[log in to unmask]>> wrote:
>>
>> Dear Tang Le
>>
>> You should try to explicitly model physiological noise. If you
>> dont have pulse and respiration recordings, you can try to extract
>> time-courses from the ventricles and use them as a user specified
>> regressor. What was the paradigm like, was it reasonably fast and
>> did you apply the standard 128s High-pass filter?
>>
>>
>> Best
>> Torben
>>
>>> On 8 Dec 2016, at 17.29, Thang Le <[log in to unmask]
>>> <mailto:[log in to unmask]>> wrote:
>>>
>>> Sorry everyone, the attached image is somehow below my signature.
>>>
>>> On Wed, Dec 7, 2016 at 6:15 PM, Thang Le <[log in to unmask]
>>> <mailto:[log in to unmask]>> wrote:
>>>
>>> Hi everyone,
>>>
>>> I found this artifact when examining the first level analysis
>>> of task-based data in several subjects. As you can see the
>>> example of a subject below, there's activation around the
>>> ventricles. This doesn't look right. Some background: 1. each
>>> subject has 5 runs and I have concatenated these runs to do
>>> connectivity analysis, 2. the activation shown here is from
>>> the entire length of trials belonging to a task condition,
>>> and 3. I have used Artifact Detection Tools (ART) to mark
>>> scans with excessive motions and regress them out during
>>> 1st-level analysis (as per our lab's usual protocol).
>>>
>>> Does anyone have any suggestion how to handle this artifact
>>> beyond ART?
>>>
>>> Thank you.
>>>
>>> Thang Le
>>> <S1.jpg>
>>>
>>>
>>
>>
>
--
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
|