[Mrtrix-discussion] TDI map differs from FA map

[Robert Smith]

Todd

Thanks for sending those images; it makes the discussion a whole lot more
clear.

Firstly, I don't see anything fundamentally 'wrong' with your images; your
processing appears to be fine.
I would however like to discuss a few points with regards to the
observations you have made, for anyone interested enough to continue
reading:


* Although you haven't provided an example image, you noted that the
discrepancy between the TDI and anatomical image appears most prominent in
the genu of the corpus callosum.
        Most likely this is caused by susceptibility-induced geometric
distortions. For a standard EPI acquisition with the phase-encode direction
running anterior-posterior, the genu is shifted posteriorly by as much as a
couple of voxels; it's not enough to make the DWIs look 'wrong' when viewed
independently, but it's highly noticeable when you overlay them on a
non-EPI image.
        In my experience, the best results for susceptibility distortion
correction (in terms of both acquisition time and accuracy) are obtained
using the reversed-phase-encode method
(this<http://www.sciencedirect.com/science/article/pii/S1053811909012294>methodis
probably easiest to read and understand). Our lab now acquires a
reversed-phase-encode b=0 pair for this purpose, for every subject scanned.
I had done some experimentation with methods for estimating the
inhomogeneity field from such an image pair, in the hope of including
something in MRtrix, but the tools included in FSL5 ('topup' and
'applytopup') work better than anything I've come up with thus far. You do
NOT need to acquire a reversed phase-encode image for every single DW
direction (although it can yield better results); a single echo-planar
image pair is enough to estimate the inhomogeneity field and 'shift' the DW
signal back to its correct spatial position.

* The overlap between the TDI and the ventricles is commonplace, and occurs
due to a combination of factors.
        The default FOD amplitude threshold for streamline termination is
0.1; maximal FOD values are around 3.0. Now imagine a voxel that is half
corpus callosum, half ventricle, that has an FOD peak amplitude of 1.5, and
the voxel adjacent to it is in the ventricle and has zero FOD amplitude. As
the streamlines algorithm interpolates the FODs as it tracks, streamlines
can go almost all the way out to the centre of the ventricle voxel before
the (interpolated) FOD amplitude threshold is disobeyed; this is why the
tracking algorithm substantially over-estimates the cross-section of the
corpus callosum (clearer when viewed on a mid-sagittal slice of the T1 /
TDI).
        Correction of reconstruction errors such as this are the target of
my Anatomically-Constrained
Tractography<http://www.sciencedirect.com/science/article/pii/S1053811912005824>framework,
which makes use of a segmented T1 image during the tracking
process. Streamlines entering the ventricles are not only terminated, but
rejected entirely (as we don't expect white matter connections to enter the
ventricles). This will be included in the next major release of MRtrix; but
it relies on very good correspondence between the DWIs and the T1, which is
why I recommend acquisition of reversed-phase-encode image pairs as soon as
possible (so you can use this capability on retrospective data).

* The final issue you raise is the contrast present in the TDIs, where the
intensity is highest in the corpus callosum (particularly its inferior
edge) and doesn't necessarily match the FA map.
        This is not surprising at all, as the two images draw their
contrast from very different sources. Neither the streamline density nor
the FA are accurate markers of 'white matter connection density' (this
paper<http://www.sciencedirect.com/science/article/pii/S1053811912007306>by
Derek Jones is worth a read), so care needs to be taken in the
interpretation of both sources.
        But since I've already strayed down the path of self-proclamation,
I might as well mention my recently-published
SIFT<http://www.sciencedirect.com/science/article/pii/S1053811912011615>method.
This tries to match the streamline densities with the tissue
volumes estimated by CSD; as a result, the TDIs look very much like the DC
term of the relevant FOD image. To me this is a much better quantification
of 'white matter connection density' than FA, TDI, or the *ad hoc* metrics
being used by the connectomics community. So I'm looking forward to seeing
what people do with it when it's released.



Anyway, I hope that was all at least somewhat interesting / informative for
yourself or others out there in the æther...

Rob

--

Robert Smith
PhD Candidate

The Florey Institute of Neuroscience and Mental Health
Melbourne Brain Centre - Austin Campus
245 Burgundy Street
Heidelberg Vic 3084
Ph: +61 3 9035 7128
Fax: +61 3 9035 7301
www.florey.edu.au


On Sat, Jan 5, 2013 at 7:31 PM, Todd Jolly <todd.jolly at uon.edu.au> wrote:

>  Hi Rob,
> No problem, I have attached a couple of images to show you what I mean.
> These show the TDI overlayed on the FA map.
> Thanks
> Todd
>  ------------------------------
> *From:* Robert Smith [r.smith at brain.org.au]
> *Sent:* Saturday, 5 January 2013 2:13 PM
> *To:* Todd Jolly
> *Subject:* Re: [Mrtrix-discussion] TDI map differs from FA map
>
>   Todd
>
>  Could you forward me a couple of example images to demonstrate exactly
> what you mean? It's difficult to establish what the problem may be from
> your description alone.
>
>  Rob
>
>
> --
>
> Robert Smith
> PhD Candidate
>
> The Florey Institute of Neuroscience and Mental Health
> Melbourne Brain Centre - Austin Campus
> 245 Burgundy Street
> Heidelberg Vic 3084
> Ph: +61 3 9035 7128
> Fax: +61 3 9035 7301
> www.florey.edu.au
>
>
> On Sat, Jan 5, 2013 at 1:57 PM, Todd Jolly <todd.jolly at uon.edu.au> wrote:
>
>>  Hi everyone,
>> I have been performing whole-brain tracking and using this to generate
>> track density images. I have noticed that the track density images differ
>> from my anatomical or FA images. The TDI appear to be including regions
>> that the anatomical images are classifying as being the lateral ventricle.
>> This is most noticeable for the genu of the corpus callosum.
>> I am concerned as these regions that are classified as CSF in the
>> anatomical and FA images appear have the highest track density in the TDI?
>> Is this normal for there to be a discrepancy or have I made an error
>> somewhere in the processing of these data?
>> Thanks in advance
>> Todd
>>
>> _______________________________________________
>> Mrtrix-discussion mailing list
>> Mrtrix-discussion at www.nitrc.org
>> http://www.nitrc.org/mailman/listinfo/mrtrix-discussion
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.nitrc.org/pipermail/mrtrix-discussion/attachments/20130105/756abc4f/attachment.html