Hello folks,
I have a simple question, but couldn’t find proper math explaination in the doc of 3dDeconvolve to be clear on this, and very few exemples on internet.
I am regressing a breathhold task from multi-echos dataset.
Following previous work on this kind of task, I would like to regress the BH task on 11 deffirent delays (-2trs : 8trs) and select the best t-test score as the best physiological delay for every voxel.
I am doing with one big regression through 3dDeconvolve :
3dDeconvolve -num_stimts 11
-polort A
-input $fname
-stim_times 3 “BH_regressor_0TR.1D” ‘TWOGAMpw(3,6,0.2,10,12,20.5)’
-stim_label 3 breathhold
-stim_times 2 “BH_regressor_-1TR.1D” ‘TWOGAMpw(3,6,0.2,10,12,20.5)’
-stim_label 2 breathhold_-1TR
-stim_times 1 “BH_regressor_-2TR.1D” ‘TWOGAMpw(3,6,0.2,10,12,20.5)’
-stim_label 1 breathhold_-2TR
-stim_times 4 “BH_regressor_1TR.1D” ‘TWOGAMpw(3,6,0.2,10,12,20.5)’
-stim_label 4 breathhold_offset_1TR
-stim_times 5 “BH_regressor_2TR.1D” ‘TWOGAMpw(3,6,0.2,10,12,20.5)’
-stim_label 5 breathhold_offset_2TR
-stim_times 6 “BH_regressor_3TR.1D” ‘TWOGAMpw(3,6,0.2,10,12,20.5)’
-stim_label 6 breathhold_offset_3TR
-stim_times 7 “BH_regressor_4TR.1D” ‘TWOGAMpw(3,6,0.2,10,12,20.5)’
-stim_label 7 breathhold_offset_4TR
-stim_times 8 “BH_regressor_5TR.1D” ‘TWOGAMpw(3,6,0.2,10,12,20.5)’
-stim_label 8 breathhold_offset_5TR
-stim_times 9 “BH_regressor_6TR.1D” ‘TWOGAMpw(3,6,0.2,10,12,20.5)’
-stim_label 9 breathhold_offset_6TR
-stim_times 10 “BH_regressor_7TR.1D” ‘TWOGAMpw(3,6,0.2,10,12,20.5)’
-stim_label 10 breathhold_offset_7TR
-stim_times 11 “BH_regressor_8TR.1D” ‘TWOGAMpw(3,6,0.2,10,12,20.5)’
-stim_label 11 breathhold_offset_8TR
-tout -fout -bout -rout
-fitts PW_fitts.nii -bucket PW_func.nii -errts PW_residuals_errts.nii
So, I am a little confused regarding the amount of overlapping regressors. Regressors are regressed independantly I suppose, but what does the residuals represent in such regression ? I use them to measure the tSNR as analysis of pilot tests.
As I understand, “Full Model” has all regressors, but I want to test regressors against each other for every voxel, so I don’t want to include every regressor for every voxel right ?
Thanks for any advice, or some explicationn page that could have some math.
edit : from their article : “For each subject and dataset, mean GM tSNR was computed on a voxelwise basis and defined as the mean signal divided by the standard deviation of the noise. Noise was defined as the residual between each voxel’s best fit to the model and the signal itself”
How do we get residuals for only one regressor while the model contains 11. Can we have a errts dataset with 11 more dimenssions ?