Dear Susanne,
the flexible factorial model in SPM12 will not handle your scan-related covariates properly. Maybe the SWE from Tom Nichols is an alternative, but I have'n tried yet to define covariates in that toolbox:
https://www.nisox.org/Software/SwE/
Maybe somebody else has an idea about alternatives.
Best,
Christian
On Wed, 17 Aug 2022 15:42:48 +0100, Susanne S <[log in to unmask]> wrote:
>Dear all,
>
>I need a little help with a regression analysis with longitudinal VBM data using the Cat12 toolbox. I have a VBM data set with 5 time points and at each time point I have collected a questionnaire which scores change over time. I was hoping to find a relationship between the changing scores and the longitudinal GMV change. but...
>a. When I enter the covariate into the Flexible Factorial (which is recommended in the cat12 toolbox for longitudinal data), I already fail in how to set the contrasts, since the 5 time points are in one covariate and accordingly in a single column of the design matrix. To make 5 covariates out of the questionnaire time points would mean that I would have to enter a bunch of zeros in each covariate, but zeros would not mean missing values, but actually the value zero. Calculating contrasts on this would produce biased results I think?
>b. To calculate a multiple regression I don't quite understand either. If I understand the statistics in the longitudinal analysis halfway, then I should not use the smoothed GM data of a single time point and correlate them with the questionnaire values of this time point, because in the preprocessing of the longitudinal data each individual GM file refers to the previous and following data, so they are not independent of each other? I.e. the individual GM data only make sense when looked at in conjunction of the other data, right?
>c. If I preprocess the whole time points independently in a crosssectional design and then do a multiple or multivariate regression, then I can't speak of neuroplasticity in the results over a period of time in relation to the questionnaire scores, because then it would always be just snapshots of the brain at time X and score at time X. There are a few papers that have done something like that, though, and then speak about longitudinal correlations. But I think that's wrong, isn't it??
>
>Is there any other way or toolbox to process the longitudinal data in such a way that I get some kind of map of GM change between two time points, which I can then correlate with the scores? So, similar to deformation based mapping (which to my knowledge is not for longitudinal data though). I have read about the Jacobian determinant, but this is also only provided for cross-sectional segmentation, not for longitudinal segmentation.
>
>Help..
>
>Best, Susanne
|