Severe EPI distortion by using fmriprep

1

Hi, I found some severe distortion, especially in the medial wall of the brain, when using fmriprep. The fmriprep version that I used is 1.0.0-rc2. The MRIQC version that I used is 0.9.8-1. I can only attach the MRIQC image in this post because new user only allows to post one image. I also tried my old way of preprocessing by spm8 with one-step normalization (EPI to EPI template provided by spm). It looks like there is less distortion by using spm8. However, the location of ventricle is less accurate by using spm8 than using fmriprep. I am not sure why there is a huge difference between preprocessing by spm and fmriprep. One possibility is that our TE is pretty low (22.2 ms), which is lower than most of the studies (30ms). It is likely the image contrast in our EPI is lower than other studies. Please let me know if there is any good suggestions about using fmriprep. Please also let me know if you guys need info because I can not post more than one image.

Thanks,
Andy!

MRIQC
MRIQC.png1386×568 303 KB

2

This is the image by using fmriprep.

fmriprep
fmriprep.jpg1500×2000 896 KB

3

This is the image by using spm8.

spm8
spm8.png895×1122 231 KB

4

Thanks for bringing this up - could you share the full HTML report for this subject?

5

Definitely. Can I send the html to your email? Please email me which email should I send you. Thanks!

6

chrisgor@stanford.edu (please include the svg files if present)

7

Email just sent. Thanks!

8

Thanks for sending the reports. The distortions you are seeing in MRIQC are due to very rough registration that MRIQC is performing to calculate one of its metrics. It is not indicative of FMRIPREP performance since FMRIPREP uses other strategies.

Nonetheless I see that the bold-T1w coregistration in FMRIPREP reports is suboptimal. This is likely due to susceptibility distortions and low tissue contrast. There are few of things you could try (in order of most likely to work):

  1. Run it again with FMRIPREP 1.0.2 - there were many improvements since rc2
  2. Run it with FMRIPREP 1.0.2 turning on the --use-syn-sdc flag
  3. Run it with FMRIPREP 1.0.2 turning on --force-no-bbr flag
  4. Run it with FMRIPREP 1.0.2 turning on both --use-syn-sdc and --force-no-bbr flags

Please let us know what you found so we could make FMRIPREP better!

1 Like
9

Just to add here that FMRIPREP 1.0.3 came out yesterday :slight_smile:

1 Like
10

Thanks for your suggestion. I will give it a try for 1.0.2.

1 Like