I would like to ask a question regarding the preprocessing results from fMRIPrep. Based on those results, I performed smoothing, scaling, and first-level analysis using AFNI (the screenshots below show some of the relevant files in my func folder).
I noticed that the brain activation imaging generated from the first-level analysis in AFNI (stats.sub-501) perfectly aligns with the *space-MNI152NLin6Asym_desc-brain_mask.nii.gz generated by fMRIPrep (as shown below).
However, when I place the MNI152NLin6Asym template (which I downloaded from TemplateFlow) underneath the activation imaging, the alignment is not as accurate (as shown below). You can see that some of the activated areas do not correspond to the regions on the brain.
Nothing looks wrong with the registration, but It looks like the brain mask from fmriprep includes some skull above the brain, and the GLM results largely look like noise.
I agree with @Steven that alignment might likely be fine. Checking alignment from each stage of processing is an important step for processing quality control (QC), and there are several cases it occurs: EPI-to-anatomical, and anatomical-to-template are two typical ones.
However, I don’t think you can look at the given image and judge this adequately. The statistics don’t carry anatomical information per see directly. Alignment might be good, or it might be off, or off in part. You should look at overlay/underlay images where the datasets with structure are present for each alignment.
Hi @Steven,
Thank you very much for your reply and for all your ongoing support! Regarding your mention that fMRIPrep includes some skull above, I have a question related to my fMRIPrep output. I noticed that in the EPI-to-anatomical registration, both imaging outputs are not skull stripped. I don’t quite understand why, as I thought fMRIPrep automatically performs skull stripping, but it seems that my images did not undergo this process (see the image below).
However, when I checked the anatomical-to-template alignment, it appears to be quite good, so I’m unsure where the issue may lie.
About your comment regarding the GLM results looking like noise, are you referring to the areas of brain activation in the image appearing chaotic and lacking significant clusters? The reason for this is that I haven’t adjusted the p-value to 0.001 and performed clustering yet. If I implement these settings, it results in no brain activation areas for this subject, making it difficult for me to showcase the brain activation images and demonstrate that my template is not perfectly aligned.
Hi @ptaylor ,
Thank you so much for your detailed response! I completely agree with you and @Steven that QC check is indeed a crucial step.
Regarding your mention of checking EPI-to-anatomical and anatomical-to-template, I have a question about the EPI-to-anatomical output from my fMRIPrep. I noticed that both the BOLD and T1W images are not skull stripped. I am using the latest version of fMRIPrep (24.0). Does this indicate that there might be an issue with my fMRIPrep preprocessing?
I’m also very excited that you shared the wonderful AFNI toolbox with me—I am just starting to learn both fMRIPrep and AFNI, and I truly appreciate your insights. It has been incredibly helpful for my learning process!
