<div class="gmail_quote">
<blockquote class="gmail_quote" style="PADDING-LEFT: 1ex; MARGIN: 0px 0px 0px 0.8ex; BORDER-LEFT: #ccc 1px solid">
<div class="im"><br> </div>Thanks for this information i also thought about this, but one question prevented me from doing this, i would like to clarify that with you, after distortion correction especially the voxels in frontal regions of the brain like corpus callosum ( i am not good in anatomy sorry apologize me if i am wrong) are streched or in other words restored to their original positions, so i am wondering whether this kind of voxel stretching and squeezing (undistorted) will contribute to any additional changes in b-matrix when compared to the changes in b-mtarix contributed by the original i.e distorted images. I understand that b-matrix rotation i calculated from the rotation matrix of the main affine matrix, i don't know whether the affine matrix will be different between before and after distortion correction images. I am sorry if i am not clear.<br>
<div class="im"><br> </div></blockquote>
<div>This is a good point. In a short answer, I believe that you don't have to worry about b-table or tensor reorientation for B0 distortion correction. Here is why;</div>
<div> </div>
<div>> For DWI, we first apply diffusion weighting along one axis, for example, along X axis (b = [1, 0, 0]). Then we apply imaging gradient to obtain an image; a circle object should look circle. During the data acquisition, B0 susceptibility influences the acquisition and brain signal intensity is "mislocated"; the circle object may look oval now.Important thing is, the diffusion weighting scheme and the imaging scheme are two separate process. B0 correction simply correct this mis-registered signal to the right place; the oval is corrected to a circle. For example, suppose you have two pixels side-by-side and a fiber is running along the right-left connecting the two pixels. If you apply [1, 0, 0] gradient (=x axis), both pixels look dark. If gradient is [0, 1, 0], they look bright. Tensor calculation tells that the fibers are running horizontally connecting the two pixels. Then if B0 distortion misplaces the right pixel shifted upward and now they are located at 45 degree each other. The tensor calculation still tells that the fiber is running holizontally. In this case, we want to move the right pixel shifted downward without touching the tensor calculation; now they are side-by-side and connected by the fiber. If you apply b-table rotation or tensor re-orientation, by the time the two pixels are aligned side-by-side, the fiber angles are rotated 45 degree running diagonally, which you don't want. I hope this makes sense.</div>
<div> </div>
<div>> When you are dealing with mis-registration by patient motion, it is a very different story. Suppose we run two DTI set. In the first set, the brain is upright when [1,0,0] gradient is applied and in the second set, the brain is 5 degree tilted and the same [1, 0, 0] is applied. Then we co-register these two images by rotating the second set by -5 degree. In this case, you have to rotate the second [1, 0, 0] b weighting by -5 degree to keep the consistency between the two sets. I think this is understandable.</div>
<div> </div>
<div>> Like wise, if you are trying to normalize two brains; say one is circle and the other is oval. In this case, when pixels are shifted to make the oval to the circle, you have to re-orient the b-table</div>
<div> </div>
<div>> When an object shape is transformed, you can either reorient the b-table before the tensor calculation or do the tensor calculation first and then rotate the tensor. They may provide slightly different results. Therefore, you'd better pick one of the methods and stick to it.</div>
<div> </div>
<div>I hope this helps.</div></div>