I have a DWI dataset, in which most subjects had a b-value of 1000 and 130 gradient directions, and the others had a b-value of 1000 and 65 directions. All data were acquired in the same scanner. How should I deal with the dataset? Simply disgarding the data with 65 directions seems the last resort. It is acceptable to include the number of directions as a covariate in the statistical model? Maybe it is better to extract part of volumes from the 130-direction data to match the number of directions across subjects? But how to choose the part of volumes seems difficult to me. Thanks for any help!
Are the 130 directions unique directions or is it a repeat of the first 65 directions, with for instance a reversal of the phase encoding direction? What kind of analysis are you planning to run on your DWI dataset? At the very least, you may take a subject with 130 directions, cut it in two and analyse using 65 directions or 130 directions and see if the result differs for your analysis.
Hi, after checking the gradient vector, the 130-direction is a repeat of the first 65-direction. I am planning for DTI analysis, such as TBSS or constructing structural network. Since the last 65-direction data is a repeat of the first 65-direction in the 130-direction case, it seems reasonable to just use the first 65-direction data for analysis?
I would think that 65 directions with b=1000 would give you enough CNR in FSL-eddy’s definition. You could check this quantitatively but my guess would be that the results would not change between using 65 directions or 130 directions. In that case you may very well use 65 volumes for all your subjects so that they are processed the same way. Besides, eddy for instance will run faster on 65 direction than on 130 directions.
This is my opinion maybe others would have a different opinion?