Dear Katharina,
after a discussion with Ian I shared his concerns about the use of TIV as scaling effect and decided to rather use TIV as nuisance variable in CAT12. Only if TIV is correlated (orthogonal) to your parameters of interest (see the CAT12 manual for more information) I would recommend to use TIV with global scaling.
Ian has not listed his own paper which might give a good overview about this topic:
https://doi.org/10.1016/j.neuroimage.2014.09.034
Best,
Christian
On Thu, 13 Jul 2017 11:57:17 +0100, Ian Malone <[log in to unmask]> wrote:
>On 11 July 2017 at 11:20, Katharina W.
><[log in to unmask]> wrote:
>> Dear CAT12 experts,
>>
>> since we used the option "modulation for non-linear transformations only" in the VBM8 toolbox and it is now in CAT12 only available in the expert mode, I was looking for the reasons why it is not recommended anymore.
>>
>> The CAT12 manual refers to Malone et al. 2015.
>>
>> Are there more papers in this direction that point to more precise results using a modulation of all transformations and adjusting for TIV later on?
>>
>> Thank you in advance and best wishes,
>> Katharina
>
>Hi Katharina,
>
>I wont claim to be a CAT12 expert, but I can explain some of the
>thinking behind that. Essentially overall brain size and the sizes of
>different structures are not (across populations) proportional to TIV,
>so directly scaling by TIV can actually introduce a variation due to
>TIV that wasn't there before, or overcorrect and simply change the
>direction. On top of that you have noise from the TIV estimation that
>you're introducing into the ratio.
>
>How much it matters depends exactly what comparison you intend to do.
>Are your groups significantly different in mean or variance of TIV (or
>TIV correlated with your EV)? If yes, you may be introducing that
>significant difference into the experiment. If not then less of a
>problem, but still not the best way to deal with noise. I used to keep
>some references on this area, it may not be completely up to date, but
>here are a few:
>Nordenskjöld 2015
>http://www.sciencedirect.com/science/article/pii/S0925492714003084
>doi:10.1016/j.pscychresns.2014.11.011
>Pintzka 2015 http://journal.frontiersin.org/article/10.3389/fnins.2015.00238/abstract
>doi:10.3389/fnins.2015.00238
>Voevodskaya 2014
>http://journal.frontiersin.org/article/10.3389/fnagi.2014.00264/abstract
>doi:10.3389/fnagi.2014.00264
>Hansen 2015 http://www.ajnr.org/content/early/2015/04/09/ajnr.A4299
>doi:10.3174/ajnr.A4299
>Liu 2014 http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4222222/
>doi:10.3389/fnins.2014.00356
>And on the reliability of ratio measures:
>Arndt 1991 http://www.sciencedirect.com/science/article/pii/092549279190031K
>doi:10.1016/0925-4927(91)90031-K
>
>--
>imalone
|