Hi there,
it seems as if the conversion process (we do it via a Matlab routine
provided with SPM2) inserted this random noise. But: if I do the
conversion with MRICro and view the data with FSLView in movie mode, the
absolute brightness (global signal) stays the same, but the brain starts
to, well, wiggle, at rate of around 3 TR (~9sec).
This introduces BIG motion artefacts, where I had global noise artefacts
before. Is this a known problem? The header files are a little confusing.
When converting the DICOMs our way (via MATLAB), the header contain the
following info (as given by ImageJ):
Title: f7508-005.img
Width: 200.00 mm (64)
Height: 200.00 mm (64)
Depth: 132.00 mm (30)
Voxel size: 3.13x3.13x4.40
Resolution: 0.320 pixels per mm
Bits per pixel: 16 (unsigned short)
Display range: 0 - 32767
No Threshold
Please note the Display range entry.
Converting the same DICOM with MRICro gives something like this:
Title: 7508_mricro_005.img
Width: 200.00 (64)
Height: 200.00 (64)
Depth: 158.40 (36)
Voxel size: 3.13x3.13x4.40
Resolution: 0.320 pixels per
Bits per pixel: 16 (unsigned short)
Display range: 0 - 1113
No Threshold
So, for one, MRICro inserted 6 additional, but empty, slices, AND: it
reduced the display range. Interestingly, the display range changes over
volumes!! The next one (#006) has a display range of 1028, and so forth.
The *.hdrs of our MATLAB processed data stay at their (16bit)-value.
Now I am a little confused which images tell the "truth" about my data:
the shaky ones or the flickering ones?!
Is there any experience with that sort of problem??
Thanks a lot in advance,
Cornelius
On Sat, 06 Nov 2004 12:39:39 +0000, Stephen Smith <[log in to unmask]>
wrote:
> Hi Cornelius - I'm pretty sure I have the answer to this. I can get quite
> nice looking activation maps for all contrasts:
>
> The problem is that you have high frequency global noise in your data.
> The
> whole brain image is (ie globally) fluctuating at quite high frequency,
> giving huge noise in the analysis. This might be due to a number of
> things, including even the data conversion process (for example, Philips
> scanners can insert a random scaling factor for each image depending on
> how you do the conversion).
>
> The reason that you didn't see this before was that SPM does intensity
> normalisation by default, which largely removes this problem. In FEAT
> this
> is an option which we have turned off by default (because it's generally
> an oversimplistic solution to these general issues) but when I turn it
> back on for your data the results are then nice. (Also, it may well be
> that the temporal smoothing in SPM, if you've used that, also reduces
> this
> "noise". Temporal smoothing also is a non-default option in FEAT, as
> again
> it's generally a bad thing to do.)
>
> Note that this noise in the data is quite easy to see if you view the
> time
> series in FSLView in movie mode; it's also clear as strong global
> components (including one all-white-matter "activation" component) in a
> MELODIC ICA analysis.
>
> So - hope that helps. Cheers, Steve.
>
>
>
> On Fri, 29 Oct 2004, Cornelius Werner wrote:
>
>> Hi there,
>> thanks for your offer. I'll try and get our IT staff to putting the data
>> somewhere accessible. But this will take until the start of the next
>> week,
>> most probably...I'll keep you informed.
>> Thank you very much again, and have a nice weekend :-) !
>> Bye,
>> Cornelius
>>
>>
>> On Fri, 29 Oct 2004 10:16:07 +0100, Stephen Smith <[log in to unmask]>
>> wrote:
>>
>> > Hi - it sounds from your description like you did things right, so
>> yes I
>> > suspect that the problem is somewhere in the details of the 3-column
>> EV
>> > setups. The best thing to do is to make a tar file of the output FEAT
>> > directory plus the input 3-column text files, and put it on a web/ftp
>> > site and we'll take a look.
>> >
>> > Cheers.
>> >
>> >
>> > On Fri, 29 Oct 2004, Cornelius Werner wrote:
>> >
>> >> Dear all,
>> >>
>> >> I conducted an event related fMRI study, where people had to press
>> >> visually cued buttons and were to evaluate the (visual) result by
>> >> another
>> >> button press, which could be either the expected one (c for
>> congruent)
>> >> or
>> >> an unexpected one (ic for incongruent). Sometimes, there were null
>> >> events
>> >> interspersed (fixation cross only). (Actually, there was one more
>> >> factor,
>> >> which is of no interest right now).
>> >> During analysis, I was interested in the hemodynamic response to the
>> >> visual presentation of both the expected result, the unexpected
>> result
>> >> and
>> >> the difference between both. Therefore, I created a "three-columns"
>> >> design
>> >> matrix with the actual presentation times (in seconds) of the
>> respective
>> >> event, the actual duration (in s) and simply "1" for the weighting
>> >> factor.
>> >> My contrasts were specified as follows:
>> >>
>> >> c ic null
>> >> 1 0 -1 Main Effect Congruent
>> >> 0 1 -1 Main Effect Incongruent
>> >> -1 1 0 Differential Activation (ic > c)
>> >>
>> >> All preprocessing options were left on default (except for slice
>> timing,
>> >> which was turned on).
>> >>
>> >> The problem now is that there is NOTHING coming up in the results on
>> the
>> >> single subjects level - no higher level visual cortices, simply nil.
>> >> When
>> >> leaving the data unthresholded, some brainstem lights up, which
>> doesn't
>> >> help too much...
>> >> Now, this really makes me wonder, as the same analysis performed with
>> >> SPM2
>> >> yields the expected results with not too weak t-values even in the
>> >> differential contrast. So I suppose I did something wrong with the
>> >> design
>> >> matrix!?! Or are there any other typical pitfalls a beginner might
>> >> stumble
>> >> into which I am not aware of??
>> >>
>> >> I'd greatly appreciate your help with this!
>> >> Best regards,
>> >> Cornelius Werner
>> >>
>> >> --
>> >> Cornelius Werner
>> >> Institut fuer Medizin (IME)
>> >> AG Kognitive Neurologie
>> >> Forschungszentrum Juelich
>> >> 52425 Juelich
>> >> Germany
>> >>
>> >> Tel. +49-(0)2461-61-8609
>> >>
>> >
>> > --
>> > Stephen M. Smith DPhil
>> > Associate Director, FMRIB and Analysis Research Coordinator
>> >
>> > Oxford University Centre for Functional MRI of the Brain
>> > John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
>> > +44 (0) 1865 222726 (fax 222717)
>> >
>> > [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>> >
>>
>>
>>
>> --
>> Cornelius Werner
>> Institut fuer Medizin (IME)
>> AG Kognitive Neurologie
>> Forschungszentrum Juelich
>> 52425 Juelich
>> Germany
>>
>> Tel. +49-(0)2461-61-8609
>>
>
> Stephen M. Smith DPhil
> Associate Director, FMRIB and Analysis Research Coordinator
>
> Oxford University Centre for Functional MRI of the Brain
> John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
> +44 (0) 1865 222726 (fax 222717)
>
> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>
--
Cornelius Werner
Institut fuer Medizin (IME)
AG Kognitive Neurologie
Forschungszentrum Juelich
52425 Juelich
Germany
Tel. +49-(0)2461-61-8609
|