open-discussion
open-discussion > RE: DWI QC Report interpretation
Apr 19, 2018 04:04 PM | Martin Styner
RE: DWI QC Report interpretation
Originally posted by Susanne Meinert:
Hi everyone,
we are trying to implement DTIPrep in our study. Unfortunately we are unsure of how to interpret the output of the Diffusion Gradient Check. 5-10% of subjects are excluded due to a mismatch in diffusion gradients in some dw-directions. What does that mean? When visually inspecting the images, we found no changes or artifacts. How should we proceed? Exclude these participants from analysis alltogether, exclude the affected directions or ignore the error in absense of observable artifacts?
Thanks in advance!
Susanne
Hi Susannewe are trying to implement DTIPrep in our study. Unfortunately we are unsure of how to interpret the output of the Diffusion Gradient Check. 5-10% of subjects are excluded due to a mismatch in diffusion gradients in some dw-directions. What does that mean? When visually inspecting the images, we found no changes or artifacts. How should we proceed? Exclude these participants from analysis alltogether, exclude the affected directions or ignore the error in absense of observable artifacts?
Thanks in advance!
Susanne
The Diffusion Gradient Check is not well named (we should probably change that), as it detects residual motion that the prior motion & eddy current correction step was not able to correct.
The default settings of that Diffusion Gradient Check step are quite aggressive (another thing to definitely change), and thus it could be that DTIPrep is removing acceptable images. One way to diagnose whether that is the case is to turn this step off and look at the otherwise-rejected DWIs and see whether they show motion compared to the other DWIs. If so, you should increase the thresholds. Alternatively, you could also just increase the thresholds and look at the output and see whether you see increased residual motion in the output DWI. If not, then you are fine with the increased thresholds.
In our own studies, we commonly use larger thresholds than the defaults. We use 1.5 - 2 x voxelsize for the translation and 1 - 1.5 degrees for rotation.
Martin
Threaded View
| Title | Author | Date |
|---|---|---|
| Giuseppe Cabras | Mar 6, 2018 | |
| Susanne Meinert | Apr 19, 2018 | |
| Martin Styner | Apr 19, 2018 | |
| Martin Styner | Mar 6, 2018 | |