Hello,
First let me declare that I have searched the forum archive and found a few related topics, which greatly helped me to understand the issue I’m going to ask. However, I have some questions that haven’t been asked/answered in those previous topics therefore I figured maybe it’s still worthwhile to ask again.
Similar to what other researchers have encountered, some of my preterm neonatal images (about 20% of total) failed to register to the target image I picked. The failure rate was even higher when I used the program picked target or FMRIB58_FA. When registration failed, the *_FA_to_target.nii.gz image looks like some random magnetic field map, nothing resembling the original FA image (I can upload these failed output images if necessary). The orientation of all my input images was aligned correctly with the target (confirmed from index.html). These images were all from preterm babies so they have low contrast and may miss anatomic features (due to illness or under-development) here and there. I assume these images are very challenging to the registration program but interestingly all my term controls successfully registered to the target (a term baby brain as well). The TBSS program, when using –n option, picked a pre-term brain as target and that resulted in a even higher registration failure rate.
According to some previous posts, tbss_2_reg uses non-linear FNIRT to register all input FA images to the target. So I tried FNIRT outside of TBSS using default settings and to my surprise, the failed images mentioned above registered well with the target. Therefore my questions are:
Why FNIRT inside TBSS failed? What setting TBSS uses with FNIRT? Is it possible to change these settings used by tbss_2_reg script (maybe tbss_3_postreg as well) and how?
If it’s impractical to change TBSS scripts, can I register those failed images outside of TBSS and merge them with all_FA.nii.gz (containing only successfully registered FA’s) and continue with subsequent TBSS steps?
Thank you for you time.
Xintian Yu
|