Hi Matt and Greg, Thanks for the feedback! I've looked at the various fix .m files from the current release; based on fix_3_clean.m, I tried the following for a single resting-state run:
# highpass filter; sigma of 1000.08 = FWHM of 2355 per Smith et al 2013 NeuroImage, also consistent with comments in the fix_3_clean.m script fslmaths rfMRI_REST1_LR.nii.gz -bptf 1000.08 -1 REST1LR_bp # format movement parameters (manually corrected header after paste step, not shown) Text2Vest Movement_Regressors.txt Movement_Regressors.mat Text2Vest Movement_Regressors.txt Movement_Regressors_dt.mat paste Movement_Regressors.mat Movement_Regressors_dt.mat > Movement_Regressors_all.mat # regress movement parameters out of timeseries and re-add mean fsl_glm -i REST1LR_bp.nii.gz -d Movement_Regressors_all.mat --out_res=REST1LR_bp_mc_demeaned.nii.gz --demean fslmaths REST1LR_bp.nii.gz -Tmean REST1LR_bp_mean fslmaths REST1LR_bp_mc_demeaned.nii.gz -add REST1LR_bp_mean.nii.gz REST1LR_bp_mc # regress movement parameters out of melodic mix fsl_glm -i filtered_func_data.ica/melodic_mix -d Movement_Regressors_all.mat --out_res=melodic_mix_mc --demean # regress unique variance from bad components (taken from .fix file) out of timeseries fsl_regfilt -i REST1LR_bp_mc.nii.gz -d melodic_mix_mc -o REST1LR_bp_mc_softICA -f "1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 28, 29, 30, 31, 32, 33, 34, 36, 38, 39, 40, 41, 42, 43, 44, 45, 46, 49, 50, 51, 52, 53, 54, 55, 56, 57, 61, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 81, 82, 83, 84" # compare against HCP's released FIX-denoised file fslmaths rfMRI_REST1_LR_hp2000_clean -sub REST1LR_bp_mc_softICA diff_REST1LR_bp_mc_softICA Visual inspection and fslstats indicate reasonably good agreement between my denoised file and the HCP's denoised file; the mean difference is about 0.84 units (compared to a mean signal intensity of around 10,000), and the "robust" range of the difference is about +/- 72 units. More worryingly, though, the maximum difference is around 2000 units, and around 6000 voxels show differences greater than 500 units, so I'm not sure machine precision can account for the differences. Does the above denoising scheme seem consistent with what FIX is doing? I plan to use FIX going forward, rather than trying to replicate it using the FSL command-line, but I'd like to understand any discrepancies between the two. Thanks again, -Ely _______________________________________________ HCP-Users mailing list HCP-Users@humanconnectome.org http://lists.humanconnectome.org/mailman/listinfo/hcp-users