[Mrtrix-discussion] csd equivalent of FA

Thu May 28 13:51:07 PDT 2015

Hi Dorian,

May I pop in a couple of related questions. First, has anyone obtained
> correlation between AFD and FA? I extracted once AFD/HMOA with another
> software (not MRtrix) and the correlation with FA was very high 0.9+. I was
> susprised by this at the time but didn't have time to follow up.
>

I'm not all that familiar with HMOA, but from what I understand, it would
definitely be correlated with FA. First off, while strongly related to AFD,
HMOA differs in that the FOD it's derived from is normalised to the b=0
signal (I think) - this immediately increases any correlations with FA,
since areas of high b=0 signal (i.e. CSF) will tend to have low FA, and
also low HMOA (but not necessarily low AFD, or at least not to the same
extend, since AFD is not normalised to the b=0).

Second, if you only look at the HMOA of the dominant fibre orientation,
then it will clearly be lower in crossing fibre regions: the FOD is now
split between multiple orientations, so purely from a partial volume /
density argument, the FOD amplitude (and hence HMOA) will be lower. FA will
also be reduced in crossing fibre regions...

So I don't think there's anything unexpected there, it just depends what
you correlate with what and what you interpret each measure to report on. I
prefer the term AFD since I really think it is a good description of what
it corresponds to: the density of WM along that orientation in that voxel
(although you could argue that it should be called apparent fibre *volume*,
since it's proportional to the partial volume of the fibre population, but
that's another story). It is explicitly *not* a measure of microstructure -
at least in the sense that it does not report on axonal properties as such.

Second, pardon if I am not updated with recent development, but is
> multi-tissue CSD similar to NODDI? What is the difference of the two?
>

I guess multi-tissue CSD is similar to NODDI, in the same way that it is
similar to bedpostx or CHARMED. They are all multi-compartment models, and
all require multi-shell data (note that I consider b=0 to be a shell in its
own right). The main difference as I see it between NODDI and MT-CSD is
that MT-CSD is a more general framework (the tissue types can be anything
you can estimate a distinct response for), data-driven (we estimate the
responses from the data), and properly models crossing fibres. But at
heart, all these methods are indeed strongly related.

Cheers,
Donald.

>
> Thank you.
> Dorian
>
> On Thu, May 28, 2015 at 3:42 PM, J-Donald Tournier <jdtournier at gmail.com>
> wrote:
>
>> Hello again,
>>
>>
>>> I will read AFD related documentation in mrtrix3; you are right, I
>>> should move to new release, but we have data already processed with old
>>> version and doing csd fittings again would take quite long time...
>>>
>>
>> Well, the version of CSD in MRtrix3 is much faster - I process a typical
>> dataset in ~15s. And with the changes we have coming up, it'll go down
>> further to ~6s... :)
>>
>> After a quick look at afd paper, I have only another question for you
>>> related to rational of AFD. If you have one fiber population in a voxel its
>>> intepretation is straightforward. But, provided you have two fiber bundles
>>> "estimated" within a voxel, what could be a proper afd measure? Could
>>> average of two fod amplitudes be the answer?
>>>
>>
>> From my point of view, the interpretation of AFD is straightforward
>> whether you have one or more fibre populations - it relates explicitly to
>> each individual fibre population. I guess the issue you're having is that
>> it's no longer a simple scalar per voxel - but then that is a clearer
>> reflection of the reality than a simple scalar could give you.
>>
>> However, one option that you have available to you is to use the *total *AFD
>> - i.e. the sum of the AFDs for all fibre populations. It is clearly less
>> informative than the AFD per fibre population (i.e. the *fixel-wise*
>> AFD), but if you must have a scalar per voxel, it would definitely be
>> better than the average AFD (if you have 2 fibre populations in a voxel,
>> their average AFD will be half that of the voxel next door that contains
>> only one of the fibre populations, which is a very artificial difference).
>> The total AFD is trivial to compute since it's the l=0 term of the CSD
>> output - the first volume in the file (all other harmonics have zero
>> integral over the sphere). This measure is actually a pretty good surrogate
>> for neurite density - with the caveat that the CSD output is not very well
>> normalised, so care would be needed to ensure data are comparable across
>> subjects, as for the AFD itself. If performing non-linear registration to a
>> common space, you will also have the issue of modulation to deal with,
>> which is not trivial since ideally you should modulate according to the
>> change in cross-sectional area across the fibre axis (as is done in AFD).
>>
>> So yes, there are options, but none are as simple as one might like...
>>
>> Cheers,
>> Donald.
>>
>>
>>
>>>
>>> Thanks again,
>>> Alessandro
>>>
>>>
>>>
>>> 2015-05-28 18:54 GMT+02:00 J-Donald Tournier <jdtournier at gmail.com>:
>>>
>>>> Hi Alessandro,
>>>>
>>>> Well, at heart AFD really is only the amplitude of the FOD along the
>>>> fibre direction of interest - actually, technically it's the integral of
>>>> the FOD over the FOD peak. You can compute the FOD amplitude in the old
>>>> version, but all the tools to compute the integral are in MRtrix3...
>>>>
>>>> Besides, to really do AFD justice requires some fairly complex
>>>> preprocessing steps to ensure values can be compared across subjects, which
>>>> we still haven't documented properly. They're all currently implemented in
>>>> MRtrix3 - this is where all the development effort is going. While you
>>>> probably could hack something together using the old version, my
>>>> recommendation would be to upgrade and wait for the documentation to be
>>>> updated (hopefully not too long after the ISMRM...).
>>>>
>>>> Even if you do want to stick with the old version, the steps required
>>>> will at least be documented on the MRtrix3 wiki, so you'll be able to see
>>>> for yourself how feasible this might be...
>>>>
>>>> Cheers,
>>>> Donald
>>>> On 28 May 2015 5:41 pm, "Alessandro Calamuneri" <
>>>> alecalamuneri at gmail.com> wrote:
>>>>
>>>>> Hi Donald,
>>>>> thanks for quick answering..
>>>>> Given I am using old mrtrix version (0.2), what could be a pipeline
>>>>> for calculating AFD?
>>>>>
>>>>> Thanks,
>>>>> Alessandro
>>>>>
>>>>> 2015-05-28 18:03 GMT+02:00 J-Donald Tournier <jdtournier at gmail.com>:
>>>>>
>>>>>> Hi Alessandro,
>>>>>>
>>>>>> Yes, that particular question has come up many times over the
>>>>>> years... I think we have a suitable answer now: apparent fibre density (
>>>>>> http://www.ncbi.nlm.nih.gov/pubmed/22036682). And this will probably
>>>>>> be improved further using multi-tissue CSD (
>>>>>> http://www.ncbi.nlm.nih.gov/pubmed/25109526)...
>>>>>>
>>>>>> Hope this answers your question.
>>>>>> Cheers,
>>>>>> Donald
>>>>>> On 28 May 2015 4:00 pm, "Alessandro Calamuneri" <
>>>>>> alecalamuneri at gmail.com> wrote:
>>>>>>
>>>>>>> Hi mrtrix experts,
>>>>>>>
>>>>>>> I would like to know whether there exists a csd based measure that
>>>>>>> might be considered the equivalent of Fractional Anisotrpy.
>>>>>>>
>>>>>>> Best Regards,
>>>>>>> Alessandro
>>>>>>>
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> Mrtrix-discussion mailing list
>>>>>>> Mrtrix-discussion at www.nitrc.org
>>>>>>> http://www.nitrc.org/mailman/listinfo/mrtrix-discussion
>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>
>>
>>
>> --
>> *Dr J-Donald Tournier (PhD)*
>>
>> *Senior Lecturer, **Biomedical Engineering*
>>
>> *Division of Imaging Sciences & Biomedical EngineeringKing's College
>> London*
>>
>>
>> *A: Department of Perinatal Imaging & Health, 1st Floor South Wing, St
>> Thomas' Hospital, London. SE1 7EH*
>> *T: +44 (0)20 7188 7118 ext 53613
>> <%2B44%20%280%2920%207188%207118%20ext%2053613>*
>> *W: http://www.kcl.ac.uk/medicine/research/divisions/imaging/departments/biomedengineering
>> <http://www.kcl.ac.uk/medicine/research/divisions/imaging/departments/biomedengineering>*
>>
>> _______________________________________________
>> Mrtrix-discussion mailing list
>> Mrtrix-discussion at www.nitrc.org
>> http://www.nitrc.org/mailman/listinfo/mrtrix-discussion
>>
>>
>

-- 
*Dr J-Donald Tournier (PhD)*

*Senior Lecturer, **Biomedical Engineering*

*Division of Imaging Sciences & Biomedical EngineeringKing's College London*

*A: Department of Perinatal Imaging & Health, 1st Floor South Wing, St
Thomas' Hospital, London. SE1 7EH*
*T: +44 (0)20 7188 7118 ext 53613*
*W: http://www.kcl.ac.uk/medicine/research/divisions/imaging/departments/biomedengineering
<http://www.kcl.ac.uk/medicine/research/divisions/imaging/departments/biomedengineering>*
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.nitrc.org/pipermail/mrtrix-discussion/attachments/20150528/5e214e47/attachment.html>