Hi Vince,
Yes, you are correct. What I meant to tell Jane was that the
non-mean adjusted (non-centered) time courses that correspond to each
spatial independent component found in GIFT or fMRLAB may be positive
or negative. However, we know that when fMRI scans are performed, the
time course signals measured have positive values. Thus, the sign of
any time course corresponding to a spatial component (a column of the
A matrix) can be determined as we know the amplitudes must be
positive, even though the ICA program may not be able to determine the
sign for itself as all ICA results are accurate up to the
muliplicative sign.
Todd
Quoting Vince Calhoun <[log in to unmask]>:
> The spatial maps and timecourses in gift are zero mean (i.e. They have both
> positive and negative values).
>
> regards,
>
> vdc
> ...... Original Message .......
> On Wed, 28 Nov 2007 12:54:50 -0700 [log in to unmask] wrote:
>> Quoting [log in to unmask]:
>>
>>>
>>>
>>>
>>> Hi Jane,
>>>
>>> You're quite right. However, spatial ICA programs like fMRLAB and
>>> GIFT multiply the time courses corresponding to each spatial
>>> independent component IF the time course has all negative amplitudes.
>>> This makes perfect sense because the time course modulates the spatial
>>> weights, whether they are positive or negative AND the fMRI time domain
>>> signals that are measured do not have negative values. Thus, only the
>>> spatial weights can be negative or positive and the corresponding time
>>> course for each must be positive.
>>>
>>> That's my understanding. It seems to be verified by what fMRLAB and
>>> GIFT do with the time courses too.
>>>
>>> Thanks
>>>
>>> Todd Penney
>>>
>>> Quoting Jane Neumann <[log in to unmask]>:
>>>
>>>>
>>>> hi todd,
>>>>
>>>> i am by no means an expert in ICA, but isn't it the case that components
>>>> in ICA can be determined only up to a factor? that is, amplitude and
>>>> sign of the determined component cannot be determined unambiguously.
>>>> maybe thinking along those lines helps.
>>>>
>>>> cheerio,
>>>> jane
>>>>
>>>>
>>>> Todd Penney wrote:
>>>>>
>>>>>
>>>>> Hello all,
>>>>>
>>>>> I have a question regarding some ICA results I received on a
>>>>> series of fMRI images from an epileptic patient who underwent a
>>>>> continuous EEG-fMRI scan. The patient was scanned for interictal
>>>>> epileptiform activity that occurred on her EEG.
>>>>>
>>>>> I performed spatial ICA and found one component that I am
>>>>> convinced is related to the interictal epileptic spikes. The
>>>>> timing of the BOLD response corresponds to that of the epileptic
>>>>> spikes and the activations are occurring in what looks like the
>>>>> temporal lobe (the patient was diagnosed with right temporal lobe
>>>>> epilepsy). However, the activations have a positive z-score
>>>>> but the BOLD response is showing a DECREASE at each of the
>>>>> spike locations in the spatial brain maps (shown by the large
>>>>> signal discharges in the attached images). This is puzzling to
>>>>> me as I am expecting a BOLD signal increase and positive
>>>>> spatial weights. I have attached two .jpg files showing the
>>>>> brain maps of the component. Is it possible that perhaps the
>>>>> BOLD signals are INCREASING because of the sign ambiguity when
>>>>> doing ICA and that the spatial weights should be DEACTIVATIONS
>>>>> with the same z-scores (which means all I have to do is
>>>>> multiply the time course and spatial weights by negative 1)? If
>>>>> not, how is it that I am getting positive spatial weights with
>>>>> a declining BOLD response at each spike? These results are
>>>>> output from the GIFT ICA program.
>>>>>
>>>>> Thank you.
>>>>>
>>>>> Todd Penne
>>>>
>>>> --
>>>>
>>>>
> ============================================================================
>>>>
>>>> Jane Neumann
>>>>
>>>> Max Planck Institute for Human Cognitive and Brain Sciences
>>>> Stephanstrasse 1a
>>>> D-04103 Leipzig, Germany
>>>>
>>>> phone: +49-(0)341-3552-1737
>>>> email: [log in to unmask]
>>>>
> ============================================================================
>>
>>
> ___
>
>
|