This looks like a terrific resource - thanks so much to package's authors/contributors!  Is there any more information on how the segmentations were derived?  Were they (semi-)automatically derived using the ART software and/or using a package like CCSeg?  Or were they manually segmented by  expert raters?  I've looked in all of the downloadable files and I haven't found more info on this point.

Thanks so much again.

Thanks for your comment.  These images were segmented using a new automated multi-atlas-based segmentation method.  We intend to include a description of the technique here in the future.  We also have an abstract accecpted on this in the ISMRM 2012 in Melbourne and plan to write a full paper about it.  The success rate of the algorithm depends on the quality of the images, ranging from 75-95% of cases.  In the remaining cases, minor manual touch-ups are required.  

By the way, the x1 factor in the speardsheets is subject's age and the x2 factor is the eTIV (as provided by OASIS) raised to the power of (2/3) in order to be used a covariate in the analysis of the area.  I will include a description of this as well in the near future.


Best,
Babak

Hello,

I have now released my Linux program 'yuki' that I have developed for automatic CC segmentation.   It can be downloaded under ART's Corpus Callosum Segmentation package.  

The previous results were obtained using an earlier version of the program.  This new version has a much better success rate.  

If you would like to use the program, read the 'gettingstartedwithyuki.pdf' documentation first which provides installation and running instructions as well as a citation for the program for publications.

Best,
Babak Ardekani

Thank you very much for this very useful tool, I tested yuki2 on a few subjects and it did very well to delineate the CC and to compute the different areas of the CC. I wonder if there is a paper published (that could be cited) detailling the different steps of the technique used, or if there is data about the reliability of the delineation (or any kind of comparison with a manual delineation) ? Anyway, many thanks for this tremendous tool !
Best regards,

Samuel

Originally posted by Babak Ardekani:

Originally posted by Samuel SARRAZIN:
Dear Samuel,

Thanks for your message.  I am glad that you are finding the program useful.  I have written a full paper about this method but it has not been published yet.  There I include some quantitative assessment of the methods reliability.

Best,
Babak 

2 questions (sorry if they are answered elsewhere):
1) Does yuki depend on art?
2) Is this truly linux only (i.e. won't run on an intel mac?)

Thanks,

Dianne

Hi Dianne,

yuki is independent of all other program in ART.  The package is self contained.  Although it uses many of the algorithms developed in other ART modules. 

For the time being, it only runs on Linux.  But I believe it can be compiled for Mac without much difficulty.  

Best,
Babak

Thanks for these explanations. I have a supplementary question. Indeed, I do not understand how the parcellation of the CC (into subregion following the Witelson's scheme) is performed in yuki ? Does this rely on the same method than your multi-atlas-based segmentation method ?
Thanks,
Samuel
Originally posted by Babak Ardekani:

Originally posted by Samuel SARRAZIN:


Hi Samuel

The subdivision is performed after the segmentation process is completed using purely geometrical methods according to the two references I provide in yuki documentation.

Babak

Dear Dr Ardekani,
Thank you for your reply.
Best,
Samuel

Hi
Thank you for the great program on CC segmentation, worked perfect for me. I was wondering if you could tell me something about how the CC is segmented into 99 points in thickness profile? How can it be normalized between a group of subjects who have varying CC areas? How are the 99 points derived considering these inter subject differences?

Thank you in advance.

Regards,
Keerthana

Originally posted by Keerthana Karunakaran:
Hi Keerthana,

I am glad the program is working well for you.  

To briefly answer your question, I used a method to find 'a medial axis' to the CC. This axis is shown as the red line in the *_cc.ppm output (also shown on Figure 1 of the documentation for version 2.1).  The medial axis is then divided into 100 equally-distance arc segments by placing 99 points on the medial axis.  Then at each point, a line is drawn that is perpendicular to the tangent line to the medial axis at that point.   The intersection points of this perpendicular line with the upper and lower CC boundaries are found and the length of the resulting line segment is recorded as the CC thickness at that point.   This method, therefore, automatically normalizes across subjects with different CC lengths.  These 99 points can be directly compared across subjects to find group difference in thickness.   

Btw, the thicknesses go from the posterior to anterior direction from the most inferior point of the splenium to the tip of the rostrum along a medial axis.

I hope this helps and let me know if you have any other questions and good luck with your work.

Babak

Babak Ardekani,

Thank you very much for your response. I have one more question.

If CC length of subject 1 is 'X' cm and subject 2 is 'Y' cm and Y>X. After 100 thickness points are obtained, when i want to compare thickness point 87 of X with point 87 of Y. I may not be looking at the same region of the brain that particular point of CC is linked to.

For example point 87 in CC of Y might include prefrontal and mid frontal gyrus, where as 87 in CC of X might only include prefrontal gyrus because of smaller CC length and segment area. 

This is where I am having problem understanding. I would very much appreciate it if you could please help.

Thank you once again.

Regards,
Keerthana

Originally posted by Keerthana Karunakaran:
Hi,

Yes, although there is no guarantee that the tracts that go through point #87 on both subjects connect the same cortical regions in both subject, there is a good chance that they do because in both subjects the distance from the inferior point of the splenium to the 87th point equal 87% of the total length of their medial axes.  

So this method assumes that for example the isthmus of the CC in different subjects is located at approximately the same location along the length of the CC if the length is normalized to 1.  

When you plot the thickness profiles for multiple subjects, you will see that they more or less have the same pattern which supports the above model.  

However, there will be cases that this assumption will not hold so this method is only good for comparing groups.   

Babak