Hi
> I am still waiting for someone who can answer to the following questions in my last email. In addition, I have carefully looked at my analyzed results from TICA. My impression is that when I inputted all datasets from both patients and controls (i.e. two groups) to Melodic, it decomposed all datasets into independent components instead of separating the analysis for the two groups.
That's right, it melodic only uses the given (temporal or subject/session) design matrices post-hoc to test estimates time courses and subject modes.
> So even thought I created a subject design matrix and contrast matrix for group comparison (e.g. group A > group B and vice versa), there is still no way for me to differentiate the spatial differences of independent components between the two groups.
That's a very ill-posed question - under the model there is no 'spatial differences of independent components between the two groups'. Instead, the tool estimates a set of spatial maps together with estimates that describe how this spatial effect is reflected across time and subject/session mode that reflects how strongly the _entire_ spatial effect is expressed in every single original data set fed into the analysis. The latter is then regressed against your subject/session design to see if these 'strength' values associated with the entire spatial component differs between subjects. If so, the natural interpretation is that this spatial map is expressed differently in your groups.
> The statistical values shown in sessions/subjects mode can only describe if the brain response of specific component in group A is significantly different from the response in group B, but didn't provide a statistical image to show the spatial differences corresponding to that contrast. Am I right?
>
That's right, they are not contrast images in the classical GLM sense
> I did a search on the Web and papers, and found that there is a dual-regression approach for voxel-wise comparisons, especially applying to resting-state data. It suggests to perform group ICA first and select components of interest, and then use dual-regression approach to identify, within each subject's fMRI dataset, subject-specific temporal dynamics and associated spatial maps. Finally, do a statistical test for group comparison. If I want to achieve my goal to find out the spatial differences between two groups for particular component, should I follow this approach for my study? Is it still applicable if my data are task-related with block paradigm, but not resting-state?
Yes, you can use DR on task-related data
hth
Christian
> Or if there is any alternative/easier approach that I can use for my study to do the group comparison?
>
> Your advice on this issue would be much appreciated.
>
> Many thanks,
> Angel
>
> On Mon, Jan 24, 2011 at 4:36 AM, Angel Wong <[log in to unmask]> wrote:
> Hi,
>
> I would like to rephrase a little bit my questions.
>
> First, what is your usual criteria when looking for task-related components? According to my understanding (please correct me if I am wrong), if the p value of the F-test on full model fit shows significant (e.g. p < 0.00000), this may reflect that the timecourse of that component follows the pattern of the given paradigm, but may have some phase shift from the original paradigm pattern. So if the p value of the contrast together shows significant (e.g. p < 0.00000), this means that the timecourse is in-phase with the paradigm, and can be considered as task-related response without phase shift. But if the p value of the contrast shows not significant (e.g. p < 1.00000), this may represent that the timecourse is totally out of phase with the paradigm. So what I interpret is that the p value of the F-test helps to determine whether the timecourse follows the given paradigm pattern, and the p value of the contrast helps to determine the degree of phase shift. Am I right?
>
> Second, what is the meaning of color spectrum shown in thresholded IC maps? Is the red spectrum corresponding to positive response to the decomposed timecourse of that component and the blue spectrum for the negative response?
>
> Third, how can I observe/interpret the group/task differences in spatial pattern of thresholded IC map if I add contrast matrix for the sessions/subjects mode? If the p value of a contrast (e.g. Group A > Group B) shows significant, does the spatial pattern of thresholded IC map represent the contrast differences between the two groups? If I add more contrasts for comparison, will it affect the number of IC decomposition from the input datasets?
>
> Thank you for your attention and help in advance.
>
> Best regards,
> Angel
>
>
> On Sun, Jan 23, 2011 at 4:55 AM, Angel Wong <[log in to unmask]> wrote:
> Hi Prof. Smith,
>
> Thanks for your reply. I have some follow-up questions for clarifying the statistical output. I know if I specify design matrix and contrast matrix for timecourse and subject mode, the Melodic will generate two GLM regression fit tables for the timecourse and subject mode respectively. But if I want to look for task-related components, should I look at the p values in the contrasts column? What does it represent if it is significant for the F-test on full model fit, but not significant for the contrasts and vice versa? As far as I know, if the p value of a contrast shows that it is significant should represent the timecourse of that component is significantly active/valid corresponding to that contrast. So if it shows that the p value of a contrast is significant, then the p value for the F-test on full model fit should also be significant, but it is not always true for the opposite direction. Am I right?
>
> For the sessions/subjects mode, if I define '1' as group1 and '2' as group2, C1 represents group1 > group2 and C2 represents group2 > group1. If p value for C1 is significant, can I say that the 'activated' regions (indicated by red color) shown in that component is significantly larger in group1 compared with group2, or 'deactivated' regions (indicated by blue color) is significantly smaller in group1 than group2?
>
> Maybe I am a bit confused with the basic concept of the TICA, and may not deliver my questions clearly. But I am glad if you can give me a clearer concept for the interpretation of the results (between spatial pattern, temporal pattern and subjects mode).
>
> Best regards,
> Angel
>
>
>
|