Seyed Mirsattari wrote:
> Thank you so much, Will! I have a related question. I have seen researchers reporting their result
> with corrected p values (i.e. FWE) and others with uncorrected P values with voxel cluster threshold
> with p less than 0.05. The latter is seen more often in the clinical fMRI papers. Any comments?
If you don't have a prior hypothesis about where your activations are going to be then
you should correct for the whole volume ie. use corrected p-values.
If you do have a prior hypothesis ie. that you expect an activation at
x,y,z then you can use the uncorrected p-value based
on the spatial extent on the nearest cluster.
So, I guess in these clinical papers the researchers had an idea
of where the activations would be.
For more on this see Friston, K. (1997) Testing for anatomically specified regional
effects. Human Brain Mapping 5,133-136.
Best wishes,
Will.
> Sorry
> for taking your time.
> Seyed
>
> Will Penny wrote:
>
>
>>Seyed Mirsattari wrote:
>>
>>
>>>Thanks Will. I really appreciated your comments. This young girl developed a condition
>>>called Epilepsia Partialis Continua (EPC). EPC is a continuous focal seizure activity that
>>>can go on for days or months. She still has it after 2 months. Medications do not stop it
>>>unless given by intravenous route that put her into sleep. So I collected activation data
>>>continuously for 3 minutes and then administered the drug in the magnet to stop the movements
>>>and recollected scans for additional 3 minutes. I hope that you appreciate the strains that
>>>was imposed on me for the analysis. I am wondering if there was a way to pull out the signal
>>>out? As a neurologist/epileptologist, we frequently encounter such patients and there is the
>>>expectation that fMRI is going to solve the problem!
>>>
>>What problem is that exactly ?
>>
>>I guess you wish to see areas more (or less) activated in session 1 than
>>in session 2.
>>
>>Many of these areas may be related to movement per se. So to get rid of these
>>you could specify the movement regressors from the realignment stage as user specified
>>regressors in the GLM analysis.
>>
>>Also, if you recorded heart rate variability or galvanic skin response measures you
>>could include these as regressors of no interest. This would help persuade reviewers
>>that any activations you see are of neuronal rather than physiologic origin.
>>
>>Finally, you need to increase the HRF cut-off to 400s as discussed before.
>>
>>If you have more patients that would be great too.
>>
>>Hope these ideas are of some help. But maybe fMRI is'nt the best tool (on its own) for identifying
>>such long-lasting seizures. If you could identify these events
>>electrically (eg via EEG - and there was some variation in the electrical response eg. discrete
>>events, changes in magnitude) and then regress fMRI activity onto EEG you would perhaps be in a much
>>stronger position - but then this would require simultaneous acquisition which is
>>only available in a few labs.
>>
>>Perhaps epilepsy-imaging experts would like to comment.
>>
>>Very best wishes, Will.
>>
>>
>>>Thanks,
>>>Seyed
>>>
>>>Will Penny wrote:
>>>
>>>
>>>
>>>>Dear Seyed,
>>>>
>>>>The experimental design you have used, unfortunately, has a major
>>>>problem.
>>>>
>>>>Your experimental manipulation, 'activation' minus 'rest' has
>>>>a period of 354 seconds ie. the activation regressor is high
>>>>for the first 177s (59 scans) and low for the last 177s.
>>>>
>>>>BOLD-fMRI is very insensitive to manipulations with such long
>>>>time periods.
>>>>
>>>>Also there is such a lot of noise at this
>>>>sort of time scale (scanner drifts etc) that what little
>>>>sensitivity there is will get filtered out by high pass filtering - you have
>>>>chosen the default high pass filter with cut-off 128s which means that any signal with
>>>>a period greater than 128s will be removed ie. your signal will be removed.
>>>>
>>>>I suggest you change your experimental design and collect new data.
>>>>
>>>>Ideally, your experimental manipulation should have a period of the order of 15
>>>>seconds or so. So you need to get your subject(s) to perform a task - one
>>>>that will activate those areas of the brain you wish to study. You could then
>>>>look at the difference in effect of task between the activation and
>>>>rest conditions.
>>>>
>>>>As a quick fix, however, you could increase the high-pass filter cut-off to
>>>>400 seconds and redo the analysis from there. It would be interesting to see the results.
>>>>But this won't address the fundamental problem.
>>>>
>>>>Hope this helps.
>>>>
>>>>Best wishes,
>>>>
>>>>Will.
>>>>
>>>>Seyed Mirsattari wrote:
>>>>
>>>>
>>>>
>>>>>Thanks, Will. I do not mind redoing the analysis with your suggestions. We
>>>>>collected the activation and two rest data separately on the same day without
>>>>>taking the patient out of the magnet but we scanned three separate times. We
>>>>>use spiral sequence for data acquisition. I then put all the three scans in
>>>>>one big folder and subjected them to normalization and soomthing. I also
>>>>>repeated it without normalizing and using her meanfMRI image for display of
>>>>>the results. In the paradigm design I said that the start was 0 with epoch
>>>>>duration of 59 scans (TR was 3 sec). I have attached on analysis for the
>>>>>activation compared to one rest epoch only. I used 1 0 in the matrix design
>>>>>instead of 1 -1. I will greatly appreciate your thoughts. By the way, SPM2
>>>>>course was great! Thank you -+again
>>>>>Seyed
>>>>>
>>>>>Will Penny wrote:
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>>Sorry for such a delayed response !
>>>>>>
>>>>>>Seyed Mirsattari wrote:
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>>Dear Will:
>>>>>>>Thank you for the wonderful SPM2 course in May 03. I am having
>>>>>>>difficulty with SPM contrast manager screen as I am trying to analyze my
>>>>>>>first set of data with SPM2. In SPM99, I entered contrast 1 0 but
>>>>>>>neither that nor any other combinations such as -1 1 0 as suggested by
>>>>>>>the help section seem to work. I am doing a Box-car design with 1 rest
>>>>>>>and 1 activation. I appreciate your help.
>>>>>>>Thanks,
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>Well, a 1 -1 should work for activation minus rest. Tell me more
>>>>>>about the design - unless of course you've finished the analysis,
>>>>>>written it up and moved on to other projects !
>>>>>>
>>>>>>Best wishes,
>>>>>>
>>>>>>Will.
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>>Seyed (Canada)
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>--
>>>>>>William D. Penny
>>>>>>Wellcome Department of Imaging Neuroscience
>>>>>>University College London
>>>>>>12 Queen Square
>>>>>>London WC1N 3BG
>>>>>>
>>>>>>Tel: 020 7833 7478
>>>>>>FAX: 020 7813 1420
>>>>>>Email: [log in to unmask]
>>>>>>URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
>>>>>>
>>>>>>epc_analysis.PPT
>>>>>>
>>>>>>Content-Type:
>>>>>>
>>>>>>application/ppt
>>>>>>Content-Encoding:
>>>>>>
>>>>>>base64
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>--
>>>>William D. Penny
>>>>Wellcome Department of Imaging Neuroscience
>>>>University College London
>>>>12 Queen Square
>>>>London WC1N 3BG
>>>>
>>>>Tel: 020 7833 7478
>>>>FAX: 020 7813 1420
>>>>Email: [log in to unmask]
>>>>URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
>>>>
>>>>
>>>
>>>
>>--
>>William D. Penny
>>Wellcome Department of Imaging Neuroscience
>>University College London
>>12 Queen Square
>>London WC1N 3BG
>>
>>Tel: 020 7833 7478
>>FAX: 020 7813 1420
>>Email: [log in to unmask]
>>URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
>>
>
>
>
--
William D. Penny
Wellcome Department of Imaging Neuroscience
University College London
12 Queen Square
London WC1N 3BG
Tel: 020 7833 7478
FAX: 020 7813 1420
Email: [log in to unmask]
URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
|