Print

Print


Dear FSL experts,

We are trying different denoising pipelines and different nuisance regression tools. Now we are focused on regressing the six motion traces and the mean CSF signal. However, we noticed that the output of fsl_regfilt and fsl_glm differs in terms of functional connectivity. Across our 55 subjects, the correlation between S1right-S1left computed after using fsl_glm seems to shifted up of ca. 0.2/0.25 in comparison to fsl_regfilt (nuisance regression is followed by BP filtering and smoothing in both cases). It must be said that FC measured with the two tools correlates quite well, with r=0.85 for S1right-S1left.

In the bash scripts we use

fsl_glm \
    -i ${subject}_registered.nii.gz \
    -d ${subject}_to_regress.mat \
    -m functional_template_mask.nii.gz \
     --out_res=${subject}_regressed.nii.gz

or

fsl_regfilt \
    -i ${subject}_registered.nii.gz \
    -d ${subject}_to_regress.mat \
    -f "1,2,3,4,5,6,7" \
    -m functional_template_mask.nii.gz \
    -o ${subject}_regressed.nii.gz


Is it possible that we are missing something in the usage of fsl_glm? If this is not the case, how would you explain the difference?

Regards,
Ludovico