Hi

You do not need to normalise the data, normalisation with the b=0 image is part of the process anyway of generating the DTI maps.

In terms of the ROI, you could define it in template space, to ensure consistency across your subjects. To do that, you need to obtain transformations for every subject to a template space (look e.g. the first steps of TBSS for an automated way to do that).

Cheers
Stam



> On 13 Nov 2016, at 14:12, John anderson <[log in to unmask]> wrote:
>
> Dear FSL experts,
> I have DTI data for 20 subjects, (patients and 10 healthy controls). I want to compare FA, MD, AD and RD values between patients and controls in a region of interest ROI
>
> This ROI is known to develop a pathology in patients due to neurodegenration. If I measure the mean of FA, MD, AD and RD in this ROI for the patents and the healthy controls. Then the values will be biased by the number of voxels inside each ROI.
>
> I think the number of voxels is lower in the ROI derived from patients as a comparison with the controls due to neurodegenration, as a result this will affect the mean FA, MD, AD, RD values derived from that ROI.
>
> In order to take into account the possible effect of voxels on the FA, MD, AD and RD values between patients and healthy controls. I am thinking of normalizing the DWI images before the analysis of ROIs by dividing the DWI maps by it is mean using the command "fslmaths" and the flage "-inm 1"
>
>
> My questions are
> 1. Is this procedure correct? Can DWI data be normalized that way?
> 2. instead of normalizing the DWI images. Ca I include the number of voxels a covariate in GLM to adjust for the possible effect of the number of voxerls on the FA, MS, AD and RD values derived from these ROI?
>
> Thank you very much for any advice
>
> Jon