🚀 go-pugleaf

In article <x4ohdr1wh6.fsf@sunecho.ifc.pi.cnr.it>,
Geir Torheim <geir@sunecho.ifc.pi.cnr.it> wrote:
:Max F Puille wrote:
:
:> Jeffrey Tsao wrote:
:[..]
:> >         The problem I want to solve, however, is a lot *dumber* than the
:> > knowledge-based approach. I want to know how to find the best match between
:> > two images, which are acquired from THE SAME 3D structures. (i.e. they
:> > SHOULD HAVE INHERIT SIMILARITIES.) 
:> 
:> They should. But you don't know which.
:
:I found an article recently which looked interesting to me. In [1] they
:have developed a method based on 'mutual information'. They do this in
:terms of the entropies in the two images. Comparing it to [2] they write:
:
:"Woods technique can break down when there is a bi-modal or multi-modal
:distribution of test volume values. This is a common occurrence
:when matching CT and MR: indistinguishable tissue in CT can map to
:significantly different tissues in MR. In addition, differing levels of
:imaged activation may normally occur in brain compartments. In contrast
:our mutual information can easily handle data that are conditionally
:multi-modal."
:
:Could this technique partly solve the problems?
:Or is this 'mutual information' method flawed in some other respect?

The correct calculation of any "information" requires some averaging,
e.g., over P lnP, where P is some distribution.  Mutual information is
broken down more than global information, but when applied to
nonlinear/oscillatory stuctures, this too may lose sensitivity that can
be obtained in other ways.  This certainly is true for EEG, e.g., in
smni97_cmi.ps.Z
          %A L. Ingber
          %T Statistical mechanics of neocortical interactions: Canonical
             momenta indicators of electroencephalography
          %J Physical Review E
          %P (to be published)
          %D 1997
          %O URL http://www.ingber.com/smni97_cmi.ps.Z

:They do mention that in some experiments they ended up in local minima,
:giving one possible explanation:
:
:"One reason this problem arises is because the MRI data is anatomical,
: while the PET data is functional."

This is just because of some gross averaging over long time and large
spatial regions typical of most PET imaging.  If more temporally or
spatially precise imaging techniques are performed that measure
neuronal function/dynamics, the quite nonlinear stochastic physiology
of cortical function presents lots of such local minima.

Lester
--
 /*             RESEARCH                            ingber@ingber.com *
  *       INGBER                                 ftp://ftp.ingber.com *
  * LESTER                                     http://www.ingber.com/ *
  * Prof. Lester Ingber __ PO Box 857 __ McLean, VA 22101-0857 __ USA */