[Insight-users] Question about SymmetricForcesDemonsRegistrat ionFilter and LevelSetMotionRegisrationFilter

[Miller, James V (Research)]

Corinne already answered the question about the SymmetricForcesDemons, so I
will answer the 
LevelSetMotion question.

The level set motion paper referenced describes two separate PDE's.  The
first is used to evolve one
image towards another image where you are only interested in the "warped"
image as an output.  This is equation 8.

The second PDE is used to evolve one image towards another image where you
are interested in the 
deformation field as an output. This is equation 9.  

Only equation 9 is implemented.  To obtain an image similar to the result of
equation 8, you take the 
output deformation field and an image an run them through a WarpImageFilter.
Doing this is not exactly 
equivalent to solving equation 8.  The paper mentions that they have fully
analyzed the stability and solution
of equation 8 but have not completed the analysis of equation 9.

Equation 9 was implemented because it produces an output like the
implementaitons of Demons, i.e. a
deformation field.

Equation 23 is used to solve equation 8 and equation 27 is used to solve
equation 9.

Jim



-----Original Message-----
From: Guorong Wu [mailto:grwu at sjtu.edu.cn]
Sent: Thursday, July 15, 2004 3:13 AM
To: ITK Mailling List
Subject: [Insight-users] Question about
SymmetricForcesDemonsRegistrationFilter and LevelSetMotionRegisrationFilter


Hello all:
	In the directory /Insight/Examples/Registration there are three
demons like non-rigid registration examples: DeformableRegistration2.cxx,
DeformableRegistration3.cxx DeformableRegistration5.cxx respectively.
DeformableRegistration2.cxx is based on Thirion's original demons algorithm,
and DeformableRegistration3.cxx use symmetric forces which can avoid the
non-uniformity of deformation field, while DeformRegistration5.cxx
implements a deformable registration algorithm that align a fixed image and
a moving image under level set motion. 
	I'v tested those three programs for sevel MR brain images finding
that levelsetmotionfilter is the most efficient one. Symmetricforcefilter
does great better result than Thirion's but the computation time is three
times than demonsregistrationfilter. The match result of
levelsetmotionfilter is about 10% better than symmetricforcefilter in every
test while takes much less (about 30%) time than demonsregistraionfilter. 
	I am so interested in itkSymmetricForcesDemonsRegistrationFilter and
itkLevelSetMotionRegistrationFilter that I look into the source code of
these algorithms, but I am got confused in some implementations in the souce
code. So I will be very appreciated if someone would give me some help.
Thanks in advance.

	QUESTION1:  This is about LevelSetMotionRegistrationFilter. I'v read
the paper mentioned in
/Code/Algorithms/itkLevelSetMotionRegistrationFilter.h. In this paper, the
process of registration is controled by two PEDs (equation 23 and equation
27). But in itkLevelSetMotionRegistrationFunction I could only find the
implementation of equation 27 in function ComputeUpdate(...) at line 399.
Could someone tell me where is the implementation of equation 23?
	QUESTION2:	SysmmetricForcesRegistrationFilter is the variant of
triditional demons which is contributed by Corinne Mattmann. I could see the
modification through the source code but cann't tell why does it take the
effect. Would you like to recommmend some paper?
	QUESTION3: Most of the registraion methods is proposed in the
multi-resolution framework. But the example programs does not use
multi-resolution. If I try to add it to the examples what kind of work I
should do or what should I take careful?

	Good Luck. 
			

????????Guorong Wu
????????grwu at sjtu.edu.cn
??????????2004-07-15

=============================================================
Guorong Wu	
The Department of Computer Science and Technology
Shanghai Jiao Tong University
1954 Huashan RD
Shanghai China    
200030
Tel:  +86 21 3226 1236  , 6293 2089
=============================================================