JISCMail - FSL Archives

On Sun, Aug 8, 2010 at 5:44 PM, Auer, Tibor M.D. Ph.D. <[log in to unmask]> wrote:

Dear Gelareh,

1.    Your design to compare two subjects should be basically a two-sampled t-test:

Group       intercepts1    Intercept2    Slope1    Slope2

1 1                      0     -4.4    0
1 1            0                   -3.4    0
1 1                      0       -1.4    0

1 1            0                    1.6    0
1 1            0                    7.6    0

2 0           1                       0 -4.4
2 0           1                       0 -3.4

2 0           1                       0 -1.4
2 0            1                       0    1.6

2 0            1                       0    7.6

Contrast                  EV1        EV2        EV3        EV4
common areas          0             0            1            1

2. A second option could be a multi-level design, if you had more subjects:
First level (per subject):

                EV1        EV2
               intecept     Slope
1 1 -4.4
1 1     -3.4
1 1     -1.4
1 1      1.6
1 1      7.6

Second level (one-sample t-test on the COPE-images from the first level): example shown with 4 subjects

                EV1

1                1
1                1
1                1
1                1

Contrast                  EV1
common areas          1

Regards,
Auer, Tibor M.D. Ph.D.
Biomedizinische NMR Forschungs GmbH
am Max-Planck Institut für Biophysikalische Chemie
Am Fassberg 11
37077 Göttingen
Germany
Phone/Work: +49-(0)551-201-1725
Phone/Home: +49-(0)551-387-0076
Mobile: +49-(0)176-8012-7921
Mail: [log in to unmask]
 
2010.08.08. 7:03 keltezéssel, Gelareh Ahmadi írta:

Dear Jesper,

The result in some areas for instant show greater value than previous in time point 1 and 2, but smaller value than previous in time point 3 and then increased in 4 and 5 . In some areas show smaller value than previous in time point 2 but it is ok for the rest time points. So as I understood by your email, this is consistent with the "linear increase" and it is probably the best I can do.

I have another question. If I want to find out common areas between my two subjects that has linear increase during 5 time points , should I design it as below? Is there any other way to design one GLM to show linear increase for each subject and also between all of them?

For common areas with linear increase:

Group       intercepts      slopes   common area

1 1     -4.4    1
1 1       -3.4    1
1 1       -1.4    1

1 1        1.6    1
1 1        7.6    1

1 1      -4.4    1
1 1    -3.4    1

1 1    -1.4    1
1 1        1.6    1
1 1        7.6    1

Contrast                  EV1        EV2        EV3
common areas          0             0            1

Or I should delet intercepts and only have 2 EV (slopes and common area) and for contrast use [0 1]?

Thanks you in advace for your kind assistance,

Regards,

Gelareh

On Wed, Aug 4, 2010 at 7:15 PM, Jesper Andersson <[log in to unmask]> wrote:

Dear Gelareh,

I ran the program exactly in a way that you suggested me, just for one subject, but still it is not correct. My goal is showing that activated areas have increased FA value during 5 time points in this subject.

The activation areas have different colors (red,orange, yellow and creamy yellow), only the pure yellow ones show increased FA value during 5 time points. Would you please help me to find out what still is wrong in my GLM that I can not get the correct result.

The GLM desing was as follow:

                 EV1        EV2
               intecept     Slope
1 1 -4.4
1 1     -3.4
1 1     -1.4
1 1      1.6
1 1      7.6

if I understand correctly your concern is that for some implicated areas you don't see a "linear increase", i.e. not every point (in time) is greater than the one preceding it? I can see why you would see this as an unwanted behavior, but it is nevertheless consistent with a linear model. Let us say you have five points and that the first four points have identical FA values and that the fifth point has a much larger FA. If you now fit a straight line to these points you will find a positive slope. It will no be a "perfect fit", but it will be positive.

There isn't really an easy way to pre-select the areas where each time point has a greater value than the previous. The best is probably to do precisely what you have done, to plot time courses for your different areas.

I hope this clarified it?

Jesper

Contrast EV1 EV2
Subject 1 0    1

Cheers,

Gelareh

On Tue, Jul 20, 2010 at 5:35 PM, Jesper Andersson <[log in to unmask]> wrote:

Hi again,

Your assumption is correct and your GLM looks reasonable.

As I am very far away from mathematics, for my understanding would it be possible to explain for me from where below numbers: -4.4, -3.4, -1.4, 1.6 and 7.6 come from? I can see -3.4 to the end comes from -4.4+ month differences but from where -4.4 come?

these numbers are simply [0 1 3 6 12] - mean([0 1 3 6 12])

And another question: If I desing this, the the contrast would be as below?
Contrast EV1 EV2 EV3 EV4
Subject 1 1 0 0 0
Subject2 0 1 0 0

these contrasts look at the the intercepts for the two subjects. If you want to look at e.g. the slope (i.e. the changes over time) for subject 1 you should use [0 0 1 0]

Good luck Jesper

--
Gelareh Ahmadi, MD
PhD Candidate
Neuroimaging Group
Howard Florey Institute
University of Melbourne
VIC, Australia, 3010

ph   +61 3 8344 1904
fax +61 3 9347 0446

--
Gelareh Ahmadi, MD
PhD Candidate
Neuroimaging Group
Howard Florey Institute
University of Melbourne
VIC, Australia, 3010

ph   +61 3 8344 1904
fax +61 3 9347 0446

--
Gelareh Ahmadi, MD
PhD Candidate
Neuroimaging Group
Howard Florey Institute
University of Melbourne
VIC, Australia, 3010

ph   +61 3 8344 1904
fax +61 3 9347 0446