Differences in tckgen output

Hello everybody,

I have a somewhat weird issue. Due to having DWI data with only one phase-encoding and no fieldmap, I could not use FSL’s topup to correct for distortions, and so I used Brainsuite, which corrects them by registering the diffusion image to the structural one (more details about this method here).

However, when generating tracts from both the data with distortions and distorsion-corrected data with the exact same ROIs and command, I get better results from the data that did not go through the Brainsuite analysis (left).

Why could this be? Brainsuite uses the nii.gz format which I later convert back to mif. Could something be happening along the way with the b-vectors or…? Any input would be much appreciated!


I’ve no experience with brainsuite, so I can’t comment on what it does and how it does it. But things to watch out for include:

  • checking that each and every volume has been appropriately registered. I’d suggest going into orthoview and keeping your finger on the right arrow. You can also place the crosshairs on clear boundaries and check for misalignment.

  • check for intensity variations that might have been introduced somehow. You’ll want the intensities in the original and corrected images to match volume by volume, at least within non-distorted areas.

  • just in case: estimate the response function from the corrected data. Don’t estimate it from the original data, then apply to the corrected data. It should work either way, unless the correction introduced differences in image scaling.

I tried to use the BrainSuite method myself a couple of years back but didn’t have any success… It would be worth at least running the distortion-corrected data through dwigradcheck. If you do manage to get decent results out of it I’d be curious to see any advice you can share from your experience.