Hi Win,
Thanks for sending the data so clearly organised.
The answer is very clear - you have two different acquisitions in
this dataset, as seen in the image header information (e.g. slightly
different voxel dimensions) and even in the image naming conventions.
It seems that whatever has changed between the two acquisitions, the
FA is not very sensitive to it, but the MD is hugely sensitive.
Changes in voxel size, b-value, correction for imaging gradients etc
can have a large effect on diffusion data, in particular MD could
change a lot. So it is quite dangerous to combine different
acquisitions in the same analysis, as proven here in your data!
Hope this helps - Cheers, Steve.
On 28 Oct 2006, at 21:39, Win Gongvatana wrote:
> Hi Steve,
>
> Thank you very much for the clarification. So it seems I can rule
> out the MD
> projection issue as a source of our problematic statistical results
> (i.e.,
> the significant group differences in most voxels even with the null
> validation data set). Do you have some ideas about other potential
> sources
> of problem? The same design matrices for randomise were used for
> the MD
> analyses as the FA analyses; and the latter yielded results
> consistent with
> expectation.
>
> Win
>
> On Sat, 28 Oct 2006 08:13:22 +0100, Steve Smith
> <[log in to unmask]> wrote:
>
>> Hi - if you are following the tbss_non_FA script and section in the
>> manual, the projection vectors themselves are _still_ being estimated
>> from the FA data - we know that this marks the tracts out well, with
>> highest values at the tract centres, and then these projection
>> vectors are being used to decide which voxels to take the MD value
>> from to put onto the skeleton. Hence for MD you won't necessarily get
>> a simple relationship between the source and destination values, but
>> you should be achieving the goal of taking the MD from the tract
>> centres (as defined by the FA) to put onto the skeleton.
>>
>> I hope this clarifies this? Cheers, Steve.
>>
>>
>>
>> On 28 Oct 2006, at 00:56, Win Gongvatana wrote:
>>
>>> Hi Steve,
>>>
>>> Thank you for your reply. I examined the values by overlaying (in
>>> afni)
>>> "all_MD_skeletonized" on "all_MD", and comparing the overlay and
>>> underlay
>>> MD
>>> values of the same subject at each pixel on the skeleton. My
>>> understandin
>>> g
>>> is that the pixel values on the skeleton in "all_MD_skeletonized"
>>> are
>>> derived from searching perpenticularly from the original skeleton
>>> for the
>>>
>>> maximum value, representing the center of the track. These should
>>> therefo
>>> re
>>> be >= the same pixels on "all_MD", which represent the values
>>> before th
>>> e
>>> projection. Our FA data are consistent with this, although MD are
>>> not. Do
>>> es
>>> this sound like I have an incorrect understanding of the process
>>> somehow?
>>>
>>> Thank you in advance for your valueble time.
>>>
>>> Win
>>>
>>> On Fri, 27 Oct 2006 09:03:23 +0100, Steve Smith
>>> <[log in to unmask]> wr
>>> ote:
>>>
>>>> Hi Win,
>>>>
>>>> It sounds like something's very wrong - in particular any null
>>>> validation of the permutation testing should definitely give the
>>>> right results. How does the all_MD (the 4D nonlinear aligned data)
>>>> look - view as a movie loop in FSLView ?
>>>>
>>>> How are you judging whether values go up or down after projection?
>>>> The projected values stay exactly the same (there's no
>>>> interpolation
>>>> involved) so I'm not sure what measure you're using.
>>>>
>>>> Cheers, Steve.
>>>>
>>>>
>>>> On 26 Oct 2006, at 21:16, Win Gongvatana wrote:
>>>>
>>>>> Hi all,
>>>>>
>>>>> I was wondering what has been your experiences with tbss and mean
>>>>> diffusivity (or other non-FA) data. We've successfully used
>>>>> tbss on
>>>>> the F
>>>>> A
>>>>> data in our lab, but still have some trouble with MD. The
>>>>> recommended ste
>>>>> ps
>>>>> were followed through to tbss_non_FA to get the projected MD
>>>>> skeletons fo
>>>>> r
>>>>> individual subjects (based on the original FA skeleton, as
>>>>> recommended).
>>>>> Performing t-tests on our various data sets processed with this
>>>>> technique
>>>>> ,
>>>>> "randomise" yielded wildly significant group differences (1-alpha
>>>>>> .99 f
>>>>> or
>>>>> the majority of voxels) even with our "control vs. control"
>>>>> analyses.
>>>>>
>>>>> One thing that I find unusual is that the projected MD values
>>>>> on the
>>>>> skeleton are sometimes lower than the values before projection,
>>>>> which is
>>>>> counter-intuitive to my understanding of the "perpendicular
>>>>> search"
>>>>> strat
>>>>> egy
>>>>> for the projection. Inspecting the projected FA values shows them
>>>>> to alwa
>>>>> ys
>>>>> be >= the values before projection.
>>>>>
>>>>> If this matters, my scaling factor for MD at step 1 was 2000 (our
>>>>> values
>>>>> range from about 0 to 5, with no negatives).
>>>>>
>>>>> Any advice/suggestions would be greatly appreciated.
>>>>>
>>>>> I'd also like to thank the FSL/TBSS team for the incredibly useful
>>>>> tools
>>>>> you
>>>>> made available.
>>>>>
>>>>> Win
>>>>
>>>>
>>>> -------------------------------------------------------------------
>>>> --
>>>> ---
>>>
>>>> ---
>>>> Stephen M. Smith, Professor of Biomedical Engineering
>>>> Associate Director, Oxford University FMRIB Centre
>>>>
>>>> FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
>>>> +44 (0) 1865 222726 (fax 222717)
>>>> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>>>> -------------------------------------------------------------------
>>>> --
>>>> ---
>>>
>>>> ---
>>>> ========================
>>> =========================
>>> ========================
>>
>>
>> ---------------------------------------------------------------------
>> ---
>> ---
>> Stephen M. Smith, Professor of Biomedical Engineering
>> Associate Director, Oxford University FMRIB Centre
>>
>> FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
>> +44 (0) 1865 222726 (fax 222717)
>> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>> ---------------------------------------------------------------------
>> ---
>> ---
>> =====================================================================
>> ====
------------------------------------------------------------------------
---
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
------------------------------------------------------------------------
---
|