general-discussion > resampling - coregistration issue
Showing 1-7 of 7 posts
Results per page:
Nov 28, 2014  06:11 PM | Kangjoo Lee
resampling - coregistration issue
I'm currently running niak fmri preprocessing using niak-boss-0.12.15 to analyze a set of public RS fMRI data from NYU ( They provide three T1-images and three EPI images for each subject, obtained from three different sessions respectively. I converted the NIFTI files to MINC files by using nii2mnc.

When I run the script for 25 subjects from this dataset,
from some of the subjects I found that there's a resampling error, while it is not for other subjects. so I take a look at one of the subjects.
When I check the intermediate files generated in the folder /intermediate/,
the slice-timing corrected functional image was normal,
motion_target_subject2_session1_run1.mnc was normal, (the target for coregistration)

fmri_subject2_session1_run1_a_res.mnc was failed in /intermediate/resample/ (file attached),
fmri_subject2_session1_run1_cor.mnc was failed in /intermediate/regress_confounds/, 

indicating that resampling in stereotaxic space was failed.

When I take a look at the folder /anat/subject2/, the images in the following files are not normal, just like as shown in the resampling failure.

anat_subject2_nativefunc_hires. mnc

Also looking at the quality control directory, for example I can see that the relative overlap of individual brain functional masks with the group mask in those subjects are extremely low (ex 0.22, 0.04, etc), shown from the file func_tab_qc_coregister_stereonl.csv.

what can I check now to solve this problem?

Thank you,
Dec 10, 2014  02:12 PM | Pierre Bellec
RE: resampling - coregistration issue
Dear Kangjoo,

Sorry for the late answer. We have observed such high failure rate for data where the voxel-to-world information in the header is either lost or compromised, and consequently the functional/anat volumes are completely misrealigned in native space. A quick fix is to set:
opt.slice_timing.flag_center = 1;

You may also have to do some centering "surgery" on the anatomical scans themselves. Please have a look at the QC&QA manual


Dec 14, 2014  08:12 AM | Kangjoo Lee
RE: resampling - coregistration issue
Dear Pierre,

The QC&QA manual is extremely helpful, and I could made the centering surgery on the header files of the anatomical scans thanks to the manual.
I'm using NYU "Classic" TRT dataset, for me it's a bit surprising that you told me that you haven't had any problem to run the preprocessing pipeline for the dataset:
If you have done it without any centering surgery, I'm wondering what's the difference between what you've done and mine. 
Since there were indeed significant centering problems in most of the subjects.

Dec 14, 2014  06:12 PM | Pierre Bellec
RE: resampling - coregistration issue
Dear Kangjoo,

I may have re-centered some volumes at the time. As I said this is quite a common step when dealing with the 1000 functional connectome project, ADHD200, ABIDE, etc. I hope your processing is now running smoothly. Best,

Dec 18, 2014  07:12 AM | Kangjoo Lee
RE: resampling - coregistration issue
Dear Pierre,

After the centering correction, niak preprocessing for most of the dataset went well, though I had to change 'slice_timing.arg_nu_correct'  from 200 to '-distance 100' in order to remove significant partial distortions in 4 subjects. I have now an another issue from two sessions, each from two different subjects, that the coregistered functional image is rotated significantly, although it was given an appropriate center.
In this case, what can I check or do? I couldn't find the info from the QC manual.
Thank you your advice, always.

Best regards,
Dec 19, 2014  12:12 AM | Kangjoo Lee
RE: resampling - coregistration issue
Problem solved. Never mind. Thanks!
Jan 2, 2015  02:01 PM | Pierre Bellec
RE: resampling - coregistration issue
Dear Kangjoo,

First of all, happy new year ! For the cases where coregistration between fMRI and T1 fail, you can try to tweak the parameters of nu_correct for the T1 (for example 50 instead of 75). It does not necessarily improve the T1 nu-correct per say, but it just perturbates the fMRI-T1 coregistration process a bit, and it ends up converging. You should be able to correct problematic cases in one or two iterations.