Hi,
On 13 Feb 2008, at 00:32, Shawn Yeh wrote:
> Thanks for clarifying Steve,
>
> I guess that answers my last question as well. So, if I am
> understanding
> this correctly, filtering out a component will remove all aspects of
> that
> component.
Well, only to the extent that the component is fully modellable as a
rank-1(i.e., representable as the outer-product ('multiplication') of
the estimated timecourse and the spatial map) and purely-additive
artefact on top of the rest of the data. But yes, this often seems to
be pretty true.
> If part of that component intersected(temporal and spatially
> matched) with EV's of interest, then strength of signal for the
> interested
> EV's will decrease by a proportional amount. This proportional
> amount can
> then be roughly estimated by looking at the % signal change and
> threshold
> bar on the IC report page?
If there is partial correlation between the artefact's timecourse and
the EV of interest, there is still a good change that ICA can separate
it out with no degration of the final activation modelling.
Cheers.
>
>
> On Tue, 12 Feb 2008 07:43:51 +0000, Steve Smith
> <[log in to unmask]> wrote:
>
>> Hi, MELODIC-based cleanup takes a component that you have selected,
>> and 'multiplies' the component's timecourse by its spatial map, to
>> produce a 4D dataset (of the same dimensions as your original data)
>> that represents the artefact you selected. This is then simply
>> subtracted out of the original data.
>>
>> Cheers, Steve.
>>
>>
>> On 11 Feb 2008, at 19:05, Shawn Yeh wrote:
>>
>>> Dear FSL users,
>>>
>>> I am currently looking into using MELODIC to identify and filter out
>>> excessive motion that cannot be adequately compensated by McFLIRT.
>>> The FSL
>>> website explains how to use Melodic, but I am having a hard time
>>> finding out
>>> how the filtering works. My questions:
>>>
>>> Does it average, smooth, or interpolate out the respective voxels
>>> with
>>> respect to the component time courses(as generated by melodic), or
>>> simply
>>> reduce/increase the voxel intensity by the thresholded amount?
>>>
>>> In theory, doesn't including the motion estimates as separate ev's
>>> already
>>> attempt to 'devalue' these voxels associated with motion?
>>>
>>> Since the IC's usually have a periodic aspect to their model, how
>>> much
>>> non-motion voxel changes are being filtered out? I understand
>>> 'rimming'
>>> effects are effectively removed this way, but I am looking at some
>>> interleaving effects from motion that doesn't look as safe to filter
>>> out.
>>>
>>>
>>> Help on this topic greatly appreciated, thanks in advance,
>>> Shawn
>>>
>>
>>
>> ---------------------------------------------------------------------------
>> Stephen M. Smith, Professor of Biomedical Engineering
>> Associate Director, Oxford University FMRIB Centre
>>
>> FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
>> +44 (0) 1865 222726 (fax 222717)
>> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>> ---------------------------------------------------------------------------
>> =
>> =
>> =
>> =
>> =====================================================================
>
---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------
|