Separate the data into High vs Low subjects (in regards to autonomic health variable score) similar to the control vs patient in the tutorial.
Or if I should use the GLM matrix to include my variable of interest. In which case, can someone please help me understand how to correctly do this or point me to where I can read more about this?
The approach recommended for use depends on the statistical properties of the numeric variable of interest. If going for the second approach, there is an implicit assumption (assuming you don’t do something more clever) of a linear relationship between a diffusion-derived quantitative measure and that variable. For instance, if it is reasonable to expect that the difference in FDC between two subjects with values 1 and 2 will be equivalent to the difference in FDC between two subjects with values 9 and 10, then I would personally advise going for the direct regression rather than binning your cohort.
If not performing statistical inference at the resolution of individual fixels, but instead aggregating values within ROIs, then such data can be fed into any statistical software of your choice, and you would therefore no longer be constrained by the capabilities of the GLM.
As far as using the variable of interest, it’s no different to adding a nuisance regressor to your design matrix, it’s just that you set the contrast vector to extract the corresponding beta coefficient since that’s what’s of interest. Or to think about it another way, a group comparison is a subset of the case of a numeric variable of interest, where each subject possesses a value of either -1 or 1.