Hi,
 
Hi Chenxi,
 
You're correct that whenever you rotate the images during co-registration, the gradient tables should also be rotated, and with the coreg_DKI.m script you can get the applied affine transformation as output. If you want to apply this, just discard the translation and dilation / contraction parts of the matrix so you're just rotating the corresponding gradient vectors. I should probably add this to the preprocessing as default if you choose not to do signal averaging... 
 
It is currently a little cumbersome to do this in DKE. The best way to accommodate this is to keep all of the DWIs separate during the tensor estimation, instead of signal averaging. By design, DKE assumes signal averaging of the b0 and repeat DWIs. If you feel that rotating the gradients is necessary, you can rearrange your data so it's average b0 followed by all b1000 and then all b2500 (without averaging the DWIs) and then make a 128 direction table after rotating the last half of the gradient vectors (this works because your b0s are not interleaved). Alternatively, you can opt not to co-register the DWIs. However, I have done a post-hoc analysis on some of my datasets and found the rotation applied during co-registration to result in a maximum angular deviation of only like a degree so I stopped worrying about it, but if there is a lot of motion during your scan then this could be more. The bigger concern for me is corrections from the image_orientation_patient field because these can be quite large and result in huge deviations of the gradient tables. Hopefully in the future, we will have a format to just supply the gradient vector for each image volume in the 4D nifty, with out the 'block' gradient table expectation, which will make these sort of changes much easier.
 
My strategy for figuring out the orientation of the gradient tables has almost always been trial and error! which is why I didn't try to do anything automated for figuring it out (it is just supplied as a text file of numbers). In my experience, if you get the tables directly from the dicoms of from DSI studio, then the table will be in LPS, whereas if you get it from mricron, the table will be in LAS, but it can also be RAS in some animal data I have seen. It's just super confusing, especially when you have data coming from different places or use different programs to extract the tables. Fortunately, you should only have to figure it out once :)
 
Best,
 
Russell