[Mrtrix-discussion] AFD-weighted connectivity

[Robert Smith]

Hi Timo,

We have discussed this internally a couple of times. Although the track
mapping code is capable of sampling the FOD amplitude along each streamline
for the purpose of generating a TWI, I haven't implemented something
similar for the connectome code. This is somewhat deliberate; we were
concerned that if we made such an option available, then people would
assume that this is a valid alternative to AFD / SIFT, which it most
certainly would not be.
When you look at what AFD / SIFT are quantifying at the fixel /
connectivity level, then consider what you would actually be quantifying by
taking the e.g. mean FOD amplitude along each streamline followed by the
e.g. mean value across streamlines connecting a pair of nodes, the latter
is not equivalent.
For instance, in this particular case (i.e. mean FOD amplitude along
streamline, mean value across streamlines for each edge) what you would
actually get is approximately proportional to the mean partial volume
fraction of the pathway of interest, averaged across the volume of the
pathway of interest (average weighted by streamline count & includes any
bundle interfaces where the partial volume fraction decreases), potentially
also confounded by streamline / FOD orientation dispersion if you sample
the FOD amplitude along the streamline tangent rather than doing FOD
segmentation & fixel assignments.
This can be implemented if there is enough demand for it, but it's not
something that we would necessarily advocate doing, and much care would
need to be taken in the interpretation of any significant results.

The afdconnectivity command is provided as what we believe to be a better
alternative for quantifying connection density between regions where SIFT
is not applicable. By using the streamlines to define a fixel mask rather
than having each streamline make an equal contribution to the edge, the
effect of non-biological regional fluctuations in streamlines density is
somewhat mitigated; by taking the integral of segmented FOD lobes rather
than the tangential FOD amplitude, effects of fibre dispersion are
mitigated; and by summing these integrals and dividing by the mean
streamline length, you get an approximation of the cross-sectional area of
the connecting pathway; a 'poor-man's SIFT' if you will. Probably the main
confound with this approach is 'partial volume' with other connections
within the fixels defined by the mask.

Regarding tcknodeextract, turns out a bug snuck into the code while we were
optimising the tractography file handling code a few months back. It only
affects tcknodeextract, hence why it hasn't been spotted until now.
(By default, we use a buffered tractography file writing class to maximise
efficiency in writing to HDD; but because tcknodeextract can potentially
generate a huge number of output .tck files at once and could therefore
require more than the system's available memory, it uses a slower
un-buffered class; the bug was causing it to write infinity over and over
again instead of the track data, hence why your matrices were empty :-/ .)
I've pushed the fix to the repository, so you can re-download or pull the
changes using Git.

P.S. If only there were a method that could provide a
biologically-meaningful measure of regional connectivity that doesn't
require the removal of streamlines... ;-P

Rob


--

*Robert Smith, Ph.D*
Research Officer, Imaging Division

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 Wed, May 28, 2014 at 5:54 PM, Roine Timo <timo.roine at uantwerpen.be>wrote:

>  Hello,
>
>  Thanks for the great software and the recent prerelease of the new
> functionalities!
>
>  I am doing structural connectivity analyses and plan to also use AFD as
> the network weight (among other measures). However, I noticed that the
> tract-level (i.e. fixel-, not voxel-wise) AFD is not implemented in the
> tck2connectome-function yet. Are there any plans to include AFD-weighting
> to tck2connectome?
>
>  An alternative, but much more complicated approach, is to use
> tcknodeextract to extract streamlines between all pairs of labels and then
> use afdconnectivity and output the AFD-values into a text file. However, it
> seems that tcknodeextract does not work properly. It saves the tck-files
> and everything looks correct (e.g. some file-sizes are larger than others,
> so the number of streamlines varies), but when loading the files (e.g.
> matlab or mrview), no streamlines exist in the files (the data-field has
> empty matrices for each streamline). I tested this also with a very simple
> parcellation image and multiple different whole brain tck-files, without
> effect. Any idea what could be causing this? I managed to circumvent the
> problem partly by using tckedit instead, but the result will be different,
> as any part of the tract may now pass through the parcellated region, not
> just the end-point (i.e. a pass-through behavior instead).
>
>  Thanks in advance!
>
>  Kind regards,
> Timo
>
>  P.S. I know that the SIFTed tract-count may be a better measure for the
> connectivity, but due to a huge amount of data (about 380 datasets), I
> would prefer to keep all of the tracts instead.
>
>  --
> Timo Roine
> timo.roine at uantwerpen.be
> iMinds-Vision Lab
> University of Antwerp, Belgium
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.nitrc.org/pipermail/mrtrix-discussion/attachments/20140529/6ed387cd/attachment-0001.html>