Hi, Clément–
Welcome to the other side of the primate tracks…
Firstly, for ease of readability, I put end of line characters in your AP command to space it vertically a bit:
afni_proc.py \
-subj_id imapt1s080 \
-script Imapt1s080/proc.imapt1s080 \
-scr_overwrite \
-out_dir imapt1s080.results \
-blocks despike tshift align tlrc volreg blur mask regress \
-dsets Rest_BOLD.nii.gz \
-copy_anat anatSS.imapt1s080.nii \
-anat_has_skull no \
-tcat_remove_first_trs 5 \
-regress_RSFC \
-volreg_align_to first \
-volreg_align_e2a \
-volreg_tlrc_warp \
-align_opts_aea \
-epi_strip 3dAutomask -cost lpc+zz -giant_move -check_flip \
-tlrc_base MNI152_2009_template_SSW.nii.gz \
-tlrc_NL_warp \
-tlrc_NL_warped_dsets \
anatQQ.imapt1s080.nii \
anatQQ.imapt1s080.aff12.1D \
anatQQ.imapt1s080_WARP.nii \
-regress_motion_per_run \
-regress_bandpass 0.01 0.1 \
-blur_size 4 \
-regress_est_blur_detrend yes \
-regress_run_clustsim no \
-execute
Short answer about your run time: yes, I’m surprised it is that long.
The necessary longer answer (plus questions for you):
-
You should not need to pre-resample your BOLD data (2.8mm isotropic is a pretty standard FMRI voxel size, certainly not “high-res”).
-
Just to be clear about the spatial resolution and number of time points in your EPI, as well as of your anatomical, what is the output of:
3dinfo -ad3 -n4 -prefix Rest_BOLD.nii.gz anatSS.imapt1s080.nii
?
- Do you have multiple cores on your computer? Are you leveraging the OpenMP capabilities of the AFNI alignment programs by using multiple threads? What is the output of:
afni_check_omp
… which tells the setting of the OMP_NUM_THREADS env variable (on my laptop with 8 CPUs, it is 6, for example).
-
mALFF should be a voxelwise value… so I don’t understand this statement:
“Also, to calculate mALFF, the result is strongly dependent on the mask used. Is there a way with afni_proc to add this mask in -regress_RSFC?”
-
Taking a step back: if you don’t censor your data, using “-regress_RSFC” is OK… However, in most cases with resting state analyses, you will likely want to censor your data to reduce effects of subject motion as much as possible. This is done by estimating locations of big motion through the volume-to-volume motion parameters (which are collapsed into a single number, the Euclidean norm or “enorm”) and outlier spikes; to provide censoring thresholds for each to afni_proc.py, as here:
-regress_censor_motion 0.3 \
-regress_censor_outliers 0.05 \
Once you are censoring time points, you should not use “-regress_RSFC”, which depends on having uniformly sampled (=non-censored) time series. The way to appropriate estimate RSFC parameters when you have censored data is use 3dLombScargle (yeeees, really its name) and 3dAmpToRSFC. This is a more appropriate framework in this case.
Please see this discussion thread for more about this discussion:
https://afni.nimh.nih.gov/afni/community/board/read.php?1,161671,161676
4b) Do you need to bandpass?
- I notice in your AP command, you use:
-volreg_align_to first \
… which would mean that the first datapoint in your FMRI time series is always used for registration, eeeeeven if motion/badness occurred there. Probably it would be preferable to select a “most appropriate” one in each case, by having AFNI determine which volume has fewest outliers; so, replace that line with:
-volreg_align_to MIN_OUTLIER \
-
Please use the python HMTL review for nicer QC output. Add this option:
-html_review_style pythonic \
-
Given your input EPI voxel size (if it is 2.8mm isotropic), blurring with 4mm radius will be pretty small. You might want something more in the range of 5-6 mm if you are doing voxelwise analysis later (if you are doing ROI-based analysis later, you would not want to blur spatially).
-
For estimating a useful mask for combining results at the group level/etc., you likely would want to include this opt:
-mask_epi_anat yes \