[Insight-developers] UNC's WA3

[Stephen R. Aylward]

Hi,

These haven't been approved yet, but I wanted to post a summary of our
tentative WA3 work.  Note, we were off on our deliverables this year -
we did stuff we hadn't planned and didn't complete all that we expected
(surprise surprise).   So, please take these for what they're worth
(bidding starts at $1... :) ).  Note, much of the stuff due Feb 1 is
nearly complete but needs documentation and approval by the group...

Any suggestions on how else we can coordinate algorithm development
between groups?   Descriptions in the document directory that we all
edit to reach a design consensus?   Should one group assume
responsibility for statistics classes, another for active surface
classes, etc?   Or is this stuff suppose to be all kitware's as has been
suggested?  Or....

UNC's Tentative WA3
===================

3. Architecture: Delivery, in source code form, of implementations of
the following methods (Software will be implemented according to the
coding conventions, styles, methodologies, and requirements document
agreed on by SRT members.)

	a. Simple C++ Interface
		- Alpha: Delivered
		- Beta: September 1, 2001.   Once an alpha release of itk has been
delivered, it will be used as the platform upon which a "simple" C++
interface (functional-style interface, not requiring knowledge of many
advance C++ conventions, short and intuitive names, etc.).
(Luis/Aylward)

	b. Adaptors
		- Alpha: Delivered an adaptor for pixel-type transforms
		- Beta: July 1, 2001 - to be included in alpha release. New adaptors
will be added including coordinate transforms via matrices with
arbitrary interpolation and an implementation for generating
scale-space. (Luis/Aylward)

	c. Statistical Classes: The goal is to provide basic capabilities that
handle the most common statistical tasks and that support the use of
other, more complete, statistical packages.
		- Alpha: February 1, 2001: Support pixel-level classification and more
general data classification. Also supports the training and testing of
density-function based classification systems (Sungwook)
		- Beta: June 1, 2001 (Sungwook)

	d. Active Surface Classes: We propose a class hierarchy that supports
implicit surface evolution in 2D and 3D. The hierarchy supports
different internal, external, and user-control functions.
		- Alpha: February 1, 2001: Supports 2D and 3D implicit surface
evolution with optimizations for speed (e.g., annulus to limit image
energy calculations) and support for derivation to implement different
cost functions (Parag)
		- Beta: June 1, 2001 (Parag)

	e. Registration Classes: We propose a class hierarchy that supports a
variety of rigid and affine registration techniques. The goal is to
support experimentation with different optimization functions.
		- Alpha: Delivered
		- Beta: February 1,2001: Affine registration via landmarks. Rigid
registration via least square error. Affine registration via curves.
(Julien)

	f. Image IO Classes
		- Alpha: Delivered
		- Beta: Further development (acceptance/rejection) will be decided by
the itk group. It seems likely that Kitware will continue any IO work.

4. Segmentation: 

	a. Data Labeling
		- Alpha: February 1, 2001: Image to Histogram, Histogram to Histogram,
Binning, Goodness of Fit functions using base statistic classes
(Sungwook)
		- Beta: June 1, 2001: Integrate LDA and PCA and classification based
on PDFs into base statistic classes (Sungwook)

	b. Active Surfaces
		- Alpha: February 1, 2001: Implicit snakes using three different cost
functions: Schlegel and Turello; Malladi, Sethian and Vemuri; Caselles,
Kimmel, Sapiro. (Parag)
		- Beta: June 1, 2001 (Parag)

	c. Model-Based
		- Gradient-directed boundary profiles: March 31 (Stetten)
		- Fuzzy medial-based segmentation: June 30 (Stetten)
		- Classification of patterns in relative wall velocity: September 30
(Stetten)
		- Incorporation of "connector segments": December 31 (Stetten)

5. Registration

	a. Support Functions
		- MRI Intensity Inhomogeneity Correction: June 1: As published in IEEE
TMI, June 2000 by Styner et al. (Parag)

	b. Distance Minimization
		- Danialson Distance: February 1, 2001: Calculates error-surface for a
set of  points, curves, and surfaces (Julien)
		- Curve-based Affine Transforms: February 1, 2001 (Julien)
		- Curve-based Non-Rigid Transforms: December 31, 2001 (Julien)

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