Alignment

﻿
﻿
*Different subjects' fMRI data supposedly aligned in MNI space. Overall, the general alignment quality is not very good. And notice that there is one subject is COMPLETELY off (horrific). Also, notice there are apparent intensity changes across subjects and/or RF coil bias differences. Could be due to differences in head size and fat, differences in positioning of subjects relative to coil. With regards to the quality of the images themselves, they themselves are pretty good (3.5/4)?*Different subjects' fMRI data supposedly aligned in MNI space. Overall, the general alignment quality is not very good. And notice that there is one subject is COMPLETELY off (horrific). Also, notice there are apparent intensity changes across subjects and/or RF coil bias differences. Could be due to differences in head size and fat, differences in positioning of subjects relative to coil. With regards to the quality of the images themselves, they themselves are pretty good (3.5/4)?
﻿
﻿
﻿
*Progressive improvement of the alignment of a high-res fMRI volume to a reference T2 anatomy. Be careful to discount the parts of the fMRI volume that have known artifacts (e.g. dropout). Have a clear sense of what is good and what is bad in terms of alignment quality. Also, be wary that different tissues show up in the fMRI vs. the T2 image, and so you (and the algorithm) should NOT get confused.*Progressive improvement of the alignment of a high-res fMRI volume to a reference T2 anatomy. Be careful to discount the parts of the fMRI volume that have known artifacts (e.g. dropout). Have a clear sense of what is good and what is bad in terms of alignment quality. Also, be wary that different tissues show up in the fMRI vs. the T2 image, and so you (and the algorithm) should NOT get confused.
﻿
﻿
﻿
*An example of using careful flipping to assess candidate alignments. Note that a different view (e.g. sagittal) might be much more visually effective to assess these effects. If you need to use 'tunnel vision' to focus on local aspect of the brain, that could be useful too (to avoid getting distracted by all the other voxels everywhere else).*An example of using careful flipping to assess candidate alignments. Note that a different view (e.g. sagittal) might be much more visually effective to assess these effects. If you need to use 'tunnel vision' to focus on local aspect of the brain, that could be useful too (to avoid getting distracted by all the other voxels everywhere else).
﻿
﻿
*Example of toggling T1, T2, EPI (after pre-processing and registration is done). Useful for birds-eye sanity checking.*Example of toggling T1, T2, EPI (after pre-processing and registration is done). Useful for birds-eye sanity checking.
﻿
*Assessing session-to-session stability*Assessing session-to-session stability
﻿
﻿
*The T1s are actually pretty well aligned in this MNI volumetric space.*The T1s are actually pretty well aligned in this MNI volumetric space.
﻿
﻿
*Notice the substantial misalignment across sessions and runs. By default, apparently fmriprep attempts to align each run separately to some anatomical reference (e.g. T1). This may be difficult to get good performance; and may underperform the alternative approach of aligning fMRI volumes directly to fMRI volumes. (Depicted is 7T data that has relatively high EPI spatial distortion, which may exacerbate the challenges.)*Notice the substantial misalignment across sessions and runs. By default, apparently fmriprep attempts to align each run separately to some anatomical reference (e.g. T1). This may be difficult to get good performance; and may underperform the alternative approach of aligning fMRI volumes directly to fMRI volumes. (Depicted is 7T data that has relatively high EPI spatial distortion, which may exacerbate the challenges.)
﻿
﻿
*Shown are 3 T1s after coregistration to each other and then the average of the 3. (This is shown for two loops.) Notice the apparent motion in the facial region whereas the brain itself is stable. (An ellipse was used in the alignment to ignore the facial region and instead focus on the brain.) Notice that averaging reduces noise.*Shown are 3 T1s after coregistration to each other and then the average of the 3. (This is shown for two loops.) Notice the apparent motion in the facial region whereas the brain itself is stable. (An ellipse was used in the alignment to ignore the facial region and instead focus on the brain.) Notice that averaging reduces noise.
﻿
﻿
﻿
*This is a quality check from fmriprep. While the check does provide some information, I think there are major suboptimalities. The smooth transitions actually impair quality checking. Also, the thick surface contours obscure content. Finally, stuffing so many slices on one image makes it hard to actually see fine-scale structure.*This is a quality check from fmriprep. While the check does provide some information, I think there are major suboptimalities. The smooth transitions actually impair quality checking. Also, the thick surface contours obscure content. Finally, stuffing so many slices on one image makes it hard to actually see fine-scale structure.
﻿
﻿
﻿
﻿
vanilla fmriprep -> massive instability across sessions. Also, there is a catastrophic failure.vanilla fmriprep -> massive instability across sessions. Also, there is a catastrophic failure.
﻿
﻿
﻿
﻿
first image is an anatomical reference; remaining images are aligned fMRI volumes from different runs in one scan session.   The quality of alignment here is very very good.first image is an anatomical reference; remaining images are aligned fMRI volumes from different runs in one scan session.   The quality of alignment here is very very good.
﻿
﻿
﻿
﻿
﻿