Hi,
there is a bit on the waytotal rational in our chapter of Heidi's & Tim's diffusion book. However, the vote was to take the full Bayesian explanation out and to keep it rather simple. I'm trying to write a formal paper up and will make sure to refer to the 'waytotal' term but it make take a while because I'm about to move and will take up a new job next week.
Cheers-
Andreas
________________________________
Von: FSL - FMRIB's Software Library [[log in to unmask]] im Auftrag von cathyliu [[log in to unmask]]
Gesendet: Donnerstag, 25. Juni 2009 11:00
An: [log in to unmask]
Betreff: Re: [FSL] asymmetires
Thanks for the information!
Yan
________________________________
Date: Wed, 24 Jun 2009 08:23:13 -0700
From: [log in to unmask]
Subject: Re: [FSL] asymmetires
To: [log in to unmask]
I haven’t published anything using the waytotal normalization yet. In Rilling et al (2008) The evolution of the arcuate fasciculus revealed with comparative DTI (Nature Neuroscience), we thresholded results based on the total number of samples, but see my previous posts on the specific methods in this paper.
Peace,
Matt.
________________________________
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of cathyliu
Sent: Wednesday, June 24, 2009 7:02 AM
To: [log in to unmask]
Subject: [FSL] asymmetires
Hi,
Thank you so much for all these helpful informations on thresholding!
We are looking for some references on the waytotal-based threshold, however, we have found any article by Pubmed. We were told that Matt had published a paper on that recently, could you tell us what is the title of the paper and in which journal?
Many thanks,
Yan
PS: We have booked your diffusion book, and it will arrive in 2 weeks! We couldn't wait for reading that!
> Date: Thu, 2 Apr 2009 02:01:02 +0200
> From: [log in to unmask]
> Subject: [FSL] AW: [FSL] asymmetires
> To: [log in to unmask]
>
> Hi Yan,
>
> p<0.2 does not necessarily have to be "very low". Do you mean peak probability or mean probability over all voxels extracted in the fdt_paths?
> Actually, it will depend on the tract and population you are studying, on the normalization, you data etc. The waytotal normalization is quite resistant to SNR issues, for example (we've studied it on parallel imaging data). However, as Matt, Saad and I have previously pointed out it largely depends on your assumptions and objective which is the most appropriate.
> >And could we still believe that the probabilistic tract is correct?
> Yes.
> >I found P is quite variable in across subjects.
> Again - depending on what you normalize to. The waytotal procedure is less variable across subjects (that is what Matt was trying to say previously, I guess).
> >In reality, how could we use the information of P?
> See above - it depends on your assumptions, prior knowledge about the tract and the objective of your study.
> Cheers-
> Andreas
>
> ________________________________
> Von: FSL - FMRIB's Software Library [[log in to unmask]] im Auftrag von Matt Glasser [[log in to unmask]]
> Gesendet: Mittwoch, 1. April 2009 19:13
> An: [log in to unmask]
> Betreff: Re: [FSL] asymmetires
>
> Are you normalizing by total number of samples or waytotal?
>
> Peace,
>
> Matt.
>
> ________________________________
> From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of cathyliu
> Sent: Wednesday, April 01, 2009 5:02 AM
> To: [log in to unmask]
> Subject: [FSL] asymmetires
>
> Hi Andreas:
>
> Thank you for all the explanation!
>
> We will hold a meeting to discuss about which way of threshold is the best for our population, as you mentioned sometimes it is just a personal taste.
>
> By the way, do you have some idea on the probability, which we could also get from the fslmaths. We found that the P (probability) in lower in the assistant tracts, e.g. SLF; and usually they are the results from bigger ROIs. If P is smaller than 0.2, that means the probability of the tract is very low? And could we still believe that the probabilistic tract is correct? In reality, how could we use the information of P? I found P is quite variable in across subjects.
>
> Regards,
>
> Yan
>
>
> > Date: Mon, 30 Mar 2009 18:05:09 +0200
> > From: [log in to unmask]
> > Subject: [FSL] AW: [FSL] AW: [FSL] asymmetires
> > To: [log in to unmask]
> >
> > Hi,
> >
> > >fslmaths fdt_paths.nii.gz –div waytotal – thr % fdt_paths_norm –odt float
> > strictly spaking you would have to multiply by 100 (after -div waytotal) to use a percent threshold. Note, however, that the magnitude of peak probability in waytotal %, for example, may seem low and that you may want to retain even low probability voxels (i.e. apply a threshold much below 1%) to "clean up" your data. See Matt's previous posts.
> > >how could you threshold fdt_paths normalized to the total number of samples? Is it based on the absolute pathway intensities?
> > Same story: just replace the waytotal by the total number of samples send out (=number of samples send out from each voxels x number of voxels in yourt ROI). Again note that the waytotal is also a "total number" of samples, i.e. those which made their way from seed to target (that's why the name;).
> > >found much less volume in the ipsilateral hemisphere with respect to the contralateral hemisphere
> > Right, that is what you would expect.
> > >if we use waytotal based threshold, the asymmetries are not so evident and in some tracts they are changed to be symmetrical
> > Exactly. In a way you may think of it like atrophy studies normalize to brain size (if your skull is small your brain must be small as well but may still be "normal" for its size; the only difference here is that normally you don't measure the hemiskull - even though only half of the skull may be small;). So does the asymmetry you see >primarily< affect your tract of interest or can it be a secondary phenomenon? Hard to tell, isn't it? Then, if you have other evidence to support one or the other notion you may ask yourself if you need tractography at all to answer your particular question or if - like quite often - we end up in circulary reasoning;)
> > >However, sometimes it is difficult to say which one is false-positive or false negative.
> > ...and sometimes it remains impossible. Practising bad science you will pick whatever will meet your purpose. Being serious you think of it before you analyze your data and decide for one or the other. I.e. do you have evidence for every single case (not just in general!) that the tract must exist in the particular brain / hemisphere? Then you may want to enforce "trackablility" despite the thresholding and use the waytotal. In normals, you may want to retain variability despite thresholding and be better of using the total number of samples send out. In some regards, the thresholding itself is just a way to "clean up " your results.
> > <book of Heidi and Tim
> > See Timi's post.
> > Cheers-
> > Andreas
>
> ________________________________
> 立刻下载 MSN 保护盾,保障Messenger 安全稳定! 现在就下载!<http://im.live.cn/safe/>
________________________________
把MSN装进手机,更多聊天乐趣等你挖掘! 立刻下载!<http://mobile.msn.com.cn/>
________________________________
把MSN装进手机,更多聊天乐趣等你挖掘! 立刻下载!<http://mobile.msn.com.cn/>
|