Dear Stephen and Co.,
I would like to reincarnate discussion on this patch, so I know where
are we standing and either it would be adopted in the submitted shape or
needs adjustment.
Best regards,
Yaroslav
On Wed, 29 May 2013, Stephen Smith wrote:
> HI - I think Dave's on holiday this week, but I expect that hopefully it
> should be fine for us to include this patch.
> Thanks, Steve.
> On 26 May 2013, at 21:38, Yaroslav Halchenko <[1][log in to unmask]>
> wrote:
> Dear FSL(view) author(s),
> As a part of NeuroSynth ([2]http://neurosynth.org) hackathon
> ([3]http://hackathon.neurosynth.org) I have developed a patch to fslview
> to
> support atlases providing various, no necessarily "probability",
> statistic. I would like to include this patch at least in a
> (Neuro)Debian distribution of fslview, and ideally I would appreciate if
> you adopt it upstream for your future releases of fslview. Although
> patch changes fslview's API, it should (I hope) not effect how existing
> Label and Probabilistic atlases are handled, exposed to the user.
> Please find patch attached, and the description excerpt below.
> I would greatly appreciate your feedback (e.g. "it will not be adopted
> upstream, because ..."/"we will adopt it with following
> changes..."/etc.).
> FSLView is capable of consuming various atlases, but unfortunately
> Probabilistic atlas is too restrictive -- it assumes values from 0 to
> 100,
> which forbids construction of atlases which might provide a different
> statistic
> describing any given voxel.
> Usecase:
> E.g. [4]neurosynth.org provides posterior probabilities for a set of
> terms
> harvested from publications. Since those are not "population percents",
> statistic values expressed as z-scores could go as high as 20, and
> visualization of corresponding p-values is not visually appealing.
> Moreover z-scores are easier to comprehend in this context, and they
> could be
> as well negative (to signal areas which were unlikely to be reported for
> any
> given location).
> Changes:
> Generalize ProbabilisticAtlas into StatisticAtlas, which gets
> parameterized
> with additional parameters previously hardcoded in the "Probabilistic"
> logic.
> Corresponding atlas's .xml file could still be "Probabilistic" but also
> could
> be of type "Statistic" with additional parameters:
> upper, lower (float) -- values for thresholding of the maps ( 5, 100 for
> Probabilistic)
> precision (int) -- to which decimal precision (10^precision) to
> consider/report
> the statistic value (0 for Probabilistic, where
> int % are reported)
> statistic (string)-- name of the statistic which would prepend its
> reports (empty for
> Probabilistic, "z" in case of neurosynth's
> z-score)
> units (string)-- reported units ("%" in Probabilistic, "" in
> neurosynth)
> Such methods as "getProbability" were left with old names for now but
> now would
> provide float instead of unsigned int.
--
Yaroslav O. Halchenko, Ph.D.
http://neuro.debian.net http://www.pymvpa.org http://www.fail2ban.org
Senior Research Associate, Psychological and Brain Sciences Dept.
Dartmouth College, 419 Moore Hall, Hinman Box 6207, Hanover, NH 03755
Phone: +1 (603) 646-9834 Fax: +1 (603) 646-1419
WWW: http://www.linkedin.com/in/yarik
|