On Mon, Jan 16, 2012 at 5:25 AM, cc yy <[log in to unmask]> wrote:Thank you very much for this detailed information Donald. I apologise for the late response, I have been busy finalising my thesis. I think I will go with 2-b, which was what Michael had suggested too.
Can I kindly ask how to convert test results to binary and add them up?
There are a number of programs that would add them up (FSL, SPM, MATLAB, etc.). I prefer MATLAB (load the statistical image (img1); then convert to 1s and 0s by creating a second matrix (Bimg1) of zeros using the equation Bimg1=zeros(size(img1)); and then Bimg1(img1>0)=1; repeat for the other three; then IMG=Bimg1+Bimg2+Bimg3, then save the image). SPM has several commands for reading and writing nifti files in MATLAB. In FSL (since this is the FSL list):
fslmaths img1 -nan -thr 3 -bin Bimg1
fslmaths img2 -nan -thr 3 -bin Bimg2
fslmaths img3 -nan -thr 3 -bin Bimg3
fslmaths Bimg1 -add Bimg2 summedimage
fslmaths Bimg3 -add summedimaged summedimage
And would the number of subjects make any difference in choosing the alternatives you suggested i.e 2-a vs.2-b. My numbers are low for a t-test (i.e. 31, 21, 42, 28 ).
Nope. The difference between 2a and 2b is in the interpretation.
Thanks a lot again.Cagri2012/1/9 MCLAREN, Donald <[log in to unmask]>
Cagri,
There are two approaches:
(1) Linear Regression where you have a grouping variable that takes values 1-4. Then you evaluate the slope of the grouping variable. This doesn't allow for different variances in each group unless you tell FSL to allow different variances, but usually doesn't because group is now a single continuous variable.
(2) Multiple Linear Regression where you have one EV for each group. There are two potential tests you can use to evaluate this model: (a) 1.5 0.5 -0.5 -1.5; (b1) 1 -1 0 0; (b2) 0 1 -1 0; (b3) 0 0 1 -1. The variance for each group can be modelled separately and should be modelled separately. Test (a) will be very similar to (1) and has the disadvantage that group 2 and 3 don't have to be less than group 1 to get a significantly positive result if group 4 is really lower than group 1 (e.g GM estimate in the four groups are .75 1 1 .25. Tests (b), are what Michael suggested, will test if each successive group is different. Then using a conjunction (I like logical AND), you can assess whether you have the ordered pattern based on successive differences in all 3 b tests. For the logical AND, threshold each (b) test and convert to binary (1 for significant, 0 otherwise); then add them up. Any voxel with a value of 3 has the continuum pattern, otherwise it doesn't. There are other forms of conjunctions that are less conservative that have been described previously.
Hope this helps.
Best Regards, Donald McLaren
=================
D.G. McLaren, Ph.D.
Postdoctoral Research Fellow, GRECC, Bedford VA
Research Fellow, Department of Neurology, Massachusetts General Hospital and
Harvard Medical School
Office: (773) 406-2464
=====================
This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED
HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is
intended only for the use of the individual or entity named above. If the
reader of the e-mail is not the intended recipient or the employee or agent
responsible for delivering it to the intended recipient, you are hereby
notified that you are in possession of confidential and privileged
information. Any unauthorized use, disclosure, copying or the taking of any
action in reliance on the contents of this information is strictly
prohibited and may be unlawful. If you have received this e-mail
unintentionally, please immediately notify the sender via telephone at (773)
406-2464 or email.
On Sat, Jan 7, 2012 at 10:49 AM, Cagri Yuksel <[log in to unmask]> wrote:I see, thank you very much anyways.
I am wondering if it's my misinterpretation of the reviewers comment, so I am copying the exact words of the reviewer to give a better idea about what is asked from us (I just changed the group names):
" the examination of a potential continuum of abnormalities between these groups is likely the more adequate approach. Given the limited statistical power of this study this is probably best examined in the context of a multiple regression model with a group variable (e.g., Group X = 4, Group Y = 3, Group Z = 2, healthy volunteers = 1) and an independent estimation of (potentially unequal) group variances. Given the strong a priori evidence for a continuous increase in gray matter deficits in prefrontal and temporal cortices over groups, one-sided testing of such a model appears legitimate."
Many thanks
Cagri
On Fri, 6 Jan 2012 08:07:23 -0600, Michael Harms <[log in to unmask]> wrote:
>I've never had to test for an "ordered continuum" between groups, so
>maybe others will chime in. Perhaps you could do the conjunction of the
>regions that satisfy 1 > 2, 2 > 3, and 3 > 4 ?
>
>good luck,
>-MH
>
>On Fri, 2012-01-06 at 12:34 +0000, Cagri Yuksel wrote:
>> Thank you Michael, that was enlightening. The answer is no, we can not assume that there is a linear relationship between these diagnostic groups.
>>
>> So how should a model be testing a continuum of GM abnormalities between these 4 diagnostic groups using a multiple regression model ? I really can not think of anything at this point.
>>
>> Cheers
>>
>> Cagri
>>
>> On Thu, 5 Jan 2012 10:56:47 -0600, Michael Harms <[log in to unmask]> wrote:
>>
>> >Whether or not a linear model relating the groups makes sense depends on
>> >on the specific groups, so I don't know whether it makes sense in your
>> >context or not. I'll just note that modeling a linear relationship
>> >between groups is a specific hypothesis that assumes that each step up
>> >in the "group" variable yields an identical change in the dependent
>> >variable (since all the groups were themselves spaced by a delta of 1
>> >unit). This is *not* the same as hypothesizing that there is merely a
>> >continuum in the DV such that 1 > 2 > 3 > 4 (or 1 < 2 < 3 < 4).
>> >
>> >cheers,
>> >-MH
>> >
>> >On Thu, 2012-01-05 at 16:38 +0000, Cagri Yuksel wrote:
>> >> Hello Michael,
>> >>
>> >> Thank you for your answer. Yes, I realized my mistake about the interpretation of the results right after I sent the message.
>> >>
>> >> These diagnostic groups are related and this analysis is to test an a priori hypothesis about a possible continuum of GM abnormalities in these groups, that's why I was thinking a linear model.
>> >>
>> >> Do you think it makes sense ? Do you have other suggestions?
>> >>
>> >> Thank you again,
>> >>
>> >> Cagri