Jason,
Thanks for the response. Glad to hear others have been concerned by this.
One thing we have noticed is that individual subject masks often look bad
because of a single or a few bad scans in the time series. Although we do
image realignment and use a bite bar, minimizing subject motion, we still
find that the "dodgy" scan is associated with a jump in one of the movement
parameters. We have started to "clean" the data by looking for these bad
scans, particularly associated with movement, and replacing them with the
mean of the immediately preceding and subsequent scan in the time series.
This ended up improving the masks enormously, especially for second level
analysis. Nonetheless, we still are noticing a loss of a surface layer of
voxels around the cortex in the second level analysis due to individual
differences in the surface extent of the brain. I've been wondering if
normalizing with more basis functions would reduce or eliminate this
intersubject variability.
Also wonder if what we are observing is a characteristic of most group
level fMRI data, or if we need to fine tune our processing steps.
Dan
>Hello,
> I agree that I find this quite alarming. There is quite a lot of
>brain matter lost at the second level. Also just one subject having more
>frontal drop out than others greatly limits the entire group analysis. One
>thing that I have done is to look at each mask image from each subject and
>see how each subject affects the group mask. I then go through each
>subject that is limiting the group mask and recheck the preprocessing to
>make sure that everything is done properly. Just one non ideal
>preprocessing step in one subject is cause for a large droput of data at
>the group level. So the point is I would also be interested in what others
>do.
>
>This shows me how careful one has to be at every step.
>
> Jason.
>
> On
>Tue, 4 Jun 2002, Daniel H. Mathalon wrote:
>
>> Dear SPMers,
>>
>> We have been examining the mask.img produced from a second level random
>> effects analysis (one-sample t-test) across 14 subjects. We find it a bit
>> alarming to see that the mask.img falls well within the perimeter of the
>> MNI template brains in all 3 dimensions. We suspected that this results
>> from variation across the individual 14 subject mask.imgs created at the
>> first level analysis, which in turn results from variation in the
>> normalization to the template EPI brain across subjects. Confirming this,
>> when we overlay the individual subject mask.img images, there are small
>> variations between subjects in the perimeter or edge of the brain, visible
>> in all planes.
>>
>> Since the mask.img created at the second level analysis comprises the
>> intersection of all of the individual subject masks, the result is group
>> mask that is smaller than the most of the individual masks and is also
>> smaller than the template. The concern is that the second level analysis
>> masks out a significant layer of cortical gray matter because of this
>> cross-subject variation.
>>
>> Have others observed this?
>>
>> Does it suggest between subject error in normalizing to the template?
>>
>> Should we ignore it or try to fix it?
>>
>> Any advice would be most welcome.
>>
>> Daniel H. Mathalon, Ph.D., M.D.
>> Assistant Professor
>> Department of Psychiatry
>> Yale University School of Medicine
>>
>> Mail address: Psychiatry Service 116A
>> VA Healthcare System
>> 950 Campbell Avenue
>> West Haven, CT 06516
>>
>> Phone (203) 932-5711, ext. 5539
>> FAX : (203) 937-3886
>> Pager 203-867-7756
>> e-mail: [log in to unmask]
>>
>
>Jason Steffener, RTS V
>Department of Psychiatry
>New Jersey Medical School
>Newark, NJ USA
>(973) 972-1604
>http://www.umdnj.edu/~steffejr
Daniel H. Mathalon, Ph.D., M.D.
Assistant Professor
Department of Psychiatry
Yale University School of Medicine
Mail address: Psychiatry Service 116A
VA Healthcare System
950 Campbell Avenue
West Haven, CT 06516
Phone (203) 932-5711, ext. 5539
FAX : (203) 937-3886
Pager 203-867-7756
e-mail: [log in to unmask]
|
