[Insight-developers] [Fwd: ITK Registration and general remarks]

[Stephen R. Aylward]

Hi,

Luis' poster at ISBI was a huge success.   There were more people 
crouded around the ITK poster than any other.   The session was suppose 
to last 2 hours, but people stayed for 3.5 hours.   The good news is 
that most people had heard of ITK, had at one point tried to download 
itk, and are very very very supportive of the effort.

It was a very easy sell...

Also, I've been working with a few other groups on itk.   Just wanted to 
forward a list of suggestions from one of those groups.   Basically, 
this is just one of several emails from different people asking for the 
Insight Book!

-------- Original Message --------
Subject: ITK Registration and general remarks
Date: Mon, 08 Jul 2002 10:17:10 +0200

I would like you to point me to some documents in the documentation
that give an overview over the highlights of algorithms that are or will be
  in ITK if there is any such document.
I started looking at http://www.itk.org/Doxygen/html/modules.html
but became somewhat overwhelmed by the vast number of things on offer.

...snip...

For each module we would like to know,

1:  What are parameters the user
  must  / should / can / should never   change

2: For those in the first three categories:
what are reasonable values and admissible / sensible ranges for those
parameters ?

3: Is there a simple way to specify reasonable default values derived
e.g. from
basic data properties like image dimensions, data type and actual range
(or maybe a simple histogram)

4: Are those parameters to be changed on a linear or rather on a
logarithmic scale ?

5: What is the scaling behavior of the parameters, i.e. which
modification (if any) of the parameter
is required to obtain similar results when using the algorithm on an
upsampled  or downsampled
resp. rescaled or shifted version of the data.

6:  Does the algorithm need some sort of normalized data
(e.g. zero background, specific range ...)

7: Maybe some formula explaining the parameters role within the
algorithm exactly

In a perfect ITK the algorithms would be able to tell part or this (1,
2, 4 ?)  themselves when the user  calls
some standard function of a base class, but building this into ITK is
probably too much work.

Giving links to the original literature in the detailed description of
the doxygen documentation
is a very good thing in this direction but following those links by
physically going to a library
( to possibly learn that required journal is not available)
  only to find out about the parameters of one algorithm might become
very cumbersome / time consuming if you think about using various ITK
algorithms.

It would be of great help to potential users if  at least some of the
information required above
was readily available.
...
But especially with some of the more complex algorithms in ITK that have
quite a number of parameters and  choices of sub algorithms
(like interpolation and transformation in the registration which in turn
have further parameters)
having no clue on the parameter chooses turns  those well designed
powerful algorithms into
unusable code for the user.

I hope I don't sound too critical  asking all these questions, but
I fear a major  obstacle to the deserved success of ITK might be its
complexity
and lack of a simple user interface to get started.
I assume few potential users will  master C++ on the level used in ITK,
most will require a much more basic interface.


-- 
===============================================
Dr. Stephen R. Aylward
Assistant Professor of Radiology
Adjunct Assistant Professor of Computer Science
http://caddlab.rad.unc.edu
aylward@unc.edu
(919) 966-9695