Hi,

It appears that the rest.nii files in the Dallas data set are left-right flipped.

I noticed this for two subjects I examined so I assume it's true for all of them.

Now, I could be wrong (left-right issues are tricky), of course. But if you look at sub16493 slice 141 for mprage_skullstripped and 21 for the rest.nii file, examine the posterior horn of the lateral ventricles, where the asymmetry is reversed for the two images. I believe the mprage is correctly oriented (RPI) and the functional image is actually LPI.

also, I flipped the rest image (and then manually resetted the orientation in the header) and reran "align_epi_anat.py" and got much better registration.

Brad Buchsbaum

there's also something odd about the voxels sizes for the Dallas data set.

rest.nii is 3.75mm isotropic
mprage_skullstripped.nii is .938mm isotropic


but if you overlay the two you'll the epi image is much larger than the mprage.

Hi Brad,

Thank you for bringing this to our attention. We have reviewed the anatomical and EPI images for the Dallas site, and believe that you are correct in noting a disparity in their left-right orientations. We will contact the original contributors to verify the correct orientation for anatomical and resting-state scan and post the updated information and files ASAP. We will also forward your query concerning their voxel size.

We'll keep you informed... Many thanks!

Maarten

Hi Brad,

together with the PIs of the Dallas dataset we found out that the EPI was L/R flipped compared to the MPRAGE (which were in the correct L/R orientation). We corrected this issue (reoriented the EPIs) and uploaded the correct data. The Dallas dataset is now available for download again.

Thanks for pointing us to this issue!
Maarten

Hi Brad,

turns out you are correct on this one too. The correct voxel sizes are

rest: 3.44 x 3.44 x 4 mm
mprage: 1 x 1 x 1 mm

It seems that something went wrong when converting the original data. Unfortunately the error was not big enough to throw of our batch processing analyses, alerting us to go back and check. This is always a danger when processing a lot of datasets through the same processing path. Fortunately (and hopefully) such errors are rather infrequent and do get picked up eventually. Thanks a lot for pointing us to this problem!

We uploaded the corrected data, they are available for download now.

Maarten