Dear Charlie,

I'm not sure I fully understand what images have different acquisitions or not.  If the before and after scans still had different acquisitions then you are likely to get a bias, but if you are only interested in testing for within-subject changes over time that are _different_ between two groups (where each group has the same set of acquisitions) then this is OK.  However, if you are testing within a single-group (even if it is across time) then you should not use this data as there will probably be a non-zero bias in the between-time measurements due to the change in acquisition parameters.

I hope this is clear.
All the best,
Mark


On 8 Jan 2014, at 15:34, Charles Leger <[log in to unmask]> wrote:

Thanks for the prompt reply Mark, and I half anticipated this response. 

One of the groups was scanned twice, the second scan occurred subsequent to training and about 6 weeks after  the initial scan; both functional task-related (fMRI) and diffusion-weighted scans were taken. Perhaps I can combine the fMRI data and DTI data from this one group's before and after scans to see if there was a significant correlation between functional activation (in response to training) and fractional anisotropy change in the tracts subserving functionally activated regions. 

Charlie


On Wed, Jan 8, 2014 at 3:54 AM, Mark Jenkinson <[log in to unmask]> wrote:
Dear Charlie,

I'm afraid I have bad news.  The difference in your protocol and the fact that it corresponds exactly to the group difference means that you will not be able to separate out any effects due to the protocol versus those due to the groups.  So I would not trust any results obtained from this analysis.  Sorry to be the bearer of bad news.

All the best,
        Mark



On 7 Jan 2014, at 22:28, charlie Leger <[log in to unmask]> wrote:

> I have three questions relating to a TBSS project, and if the answer to the first is “not appropriate” then I will need to apply feedback regarding the remaining two questions elsewhere.
>
> I am tasked with comparing FA and ADC symmetry using tbss_sym  between 2 equal sized groups (n=10 controls, n=10 athletes).
>
> (1)
> First, I would  like to compare FA and ADC values . Unfortunately, differing diffusion acquisition parameters were used for each group, and I assume that this is not the recommended circumstance.
> The bvals = 1000 in both groups. However, as shown below in the fslinfo output of one (pre eddy correction) file from each group,  the resolution, mm per voxel, and number of direction gradients differ
> between the groups. Would it be inappropriate to compare these two groups using tbss_sym? Assume preprocessing is completed in fsl (eddy correction, DTIFIT)
>
> Controls
> charlies-mbp:dti$ fslinfo s1_dti.nii.gz
> data_type      INT16
> dim1           128
> dim2           128
> dim3           56
> dim4           65
> datatype       4
> pixdim1        1.5000000000
> pixdim2        1.5000000000
> pixdim3        1.9999998808
> pixdim4        6.9000000954
> cal_max        0.0000
> cal_min        0.0000
> file_type      NIFTI-1+
>
> Athletes
> charlies-mbp:EP2D_DIFF_MDDW_30_P2_0007$ fslinfo 57_KS.nii.gz
> data_type      INT16
> dim1           122
> dim2           122
> dim3           60
> dim4           31
> datatype       4
> pixdim1        1.9672131538
> pixdim2        1.9672131538
> pixdim3        2.0000000000
> pixdim4        8.3000001907
> cal_max        0.0000
> cal_min        0.0000
> file_type      NIFTI-1+
>
>
> (2)
>
> After completing TBSS steps 1-4, and prior to running randomize I know I can use fslmeants on the all_FA_skeletonized.nii.gz 4D file
> (thank you Kirstie) to extract information for histograms.  But I would like to make white matter masks for a few tracts including the CC,
> uncinate and longitudinal faciculi, coronal radiate and cingulum. I can make white matter masks from the JHU white matter tractography atlas
> and perhaps the JHU ICBM DTI white matter labels (though I am uncertain if a mask can be made from a label?). Is there any method other than simple
> visual inspection and comparison for using the Mori et al., (2008) white matter atlas to make MNI space masks?
>
> (3)
>
> I would like to add some additional supportive statistics to the randomize statistical analysis using R statistics software. Any recommendations?
> Perhaps setting some of the randomize output options (I don’t know the exact meaning of permuted vs unpermuted in the context
> of randomise options, e.g. glm_output unpermuted case only). I have only a basic understanding of FSL and a basic-to-intermediate understanding of R.
>
>
> As always, any feedback or suggestions would be very much appreciated.
>
> Charlie