IXI dataset diffusion data - number of directions


The IXI dataset openly provides scans of diffusion MRI from a large number of subjects. There is however a strange discrepancy in the data. The b-vectors and b-values that are provided have 16 entries each. This includes a zero entry in the b-values, which corresponds to the first volume of the acquisition. The first b-vector is [0, 0, 1], but there is also a [0, 0, -1] in the table further down, so I suspect that this is also supposed to be zero. However, there are 17 volumes of diffusion data provided. They are provided as 17 different nifti files, named “XXX-DTI-00.nii.gz”, “XXX-DTI-01.nii.gz”… “XXX-DTI-16.nii.gz”, where “XXX” is the individual subject identifier – something like “IXI640-Guys-1106”.

In the past when I analyzed this data, I scratched my head, tried a couple of things and then came to the conclusion that dropping the last volume makes sense. This was done based on looking at FA, MD and color FA maps for a few individuals and based on the fact that dropping that volume results in the best model fit for the DTI model. But now I am wondering whether anyone out there knows what is actually going on here.

Why are there 17 volumes in this data, but only 16 b-values/b-vectors? If the last volume really shouldn’t be used for fitting the data, what is that last volume for?



Aha! Upon further inspection (and with thanks to Jason Yeatman for suggesting this!), it seems that this last volume is the mean diffusivity or trace, or some-such. In other words, if you want to analyze the data from scratch, dropping that last volume is sensible.

1 Like