[Insight-developers] Fwd: [SCIRUN-USERS] running ITK filters from Matlab: new architecture

[Luis Ibanez]

---------- Forwarded message ----------
From: Ramón Casero Cañas <ramon.casero at comlab.ox.ac.uk>
Date: 2011/5/21
Subject: [SCIRUN-USERS] running ITK filters from Matlab: new architecture
To: Mailing list SCIRun <scirun-users at sci.utah.edu>



Dear Utah group,

This is a bit off-topic, but because the function can read your NRRD
format, I thought you may be interested.

Our group (lead by Vicente Grau, Blanca Rodríguez, Peter Kohl and
David Gavaghan) is developing a function to run ITK filters from
Matlab. Like MATITK, but as far as I know, they don't provide the
source code, while ours is open source.

More details below, in the original release email.

Best regards,

Ramón.

---

Hi all,

Exciting news, after a substantial change in the architecture of
itk_imfilter(), the Matlab function that allows to run ITK filters
from Matlab has a new release

http://code.google.com/p/gerardus/source/browse/#svn%2Ftags%2Frelease-0.4.0%2Fmatlab%2FItkToolbox

Now it looks more like a software project rather than a
quick-and-dirty hack. (Despite the lack of testing units or
documentation! ;) )

First of all, many thanks to Miguel Bernabeu, Joe Pitt-Francis, Raf
Bordas and Pras Pathmanathan, for the useful discussions, pointing in
the right direction, solving compilation problems, etc. Without their
vast programming knowledge and willingness to help, I doubt this could
have been finished.


Features in the new release:
============================

- Any type of Matlab data is accepted at the input (boolean, double,
single, uint8, int8, etc). Also empty matrices.

- Output type is automatically computed depending on the filter and the image.

- 2D/3D Images can be loaded as a Matlab array, or an NRRD struct
(e.g. images produced by Seg3D), so it's possible to work in voxel
units or real world units.

- Currently, 3 filters are available: 3D skeletonization, Signed
Maurer distance map, and Danielsson distance map.

- Multiple output arguments are allowed, if the corresponding ITK
filter produces more than 1 output.

- The new architecture makes it quite easy to add new filters to the program.

- Project built with CMake, so in theory it should be multi-platform
and work with Linux, Windows and MacOS X. (But I have tested it only
with Linux.)

- Polymorphism and dynamic casting are used to have a BaseFilter class
with the default code for typical filters, while at the same time it's
possible to override the defaults for filters with special
requirements.

- Parser functions allow to map the run time input/output voxel type
and filter type variables to compilation time templates as required by
ITK without having to code a nesting nightmare.

- General code has been written for all input/output type combination.
Thanks to a macro FILTEREXCLUDE, forbidden combinations can be cleanly
prevented from compilation.

- Code distributed in several files makes compilation a lot faster and
the code easier to expand and maintain.

- A macro FILTERINST is used for clean explicit template instantiation
of the allowed input/output type combinations.

- The filtering process is clearly split up into:

  + filter->CopyMatlabInputsToItkImages();
  + filter->FilterSetup();
  + filter->RunFilter();
  + filter->CopyAllFilterOutputsToMatlab();


Best regards,

Ramon.

--
Dr. Ramón Casero Cañas

Computational Biology
Computing Laboratory
University of Oxford
Wolfson Building, Parks Rd
Oxford OX1 3QD

tlf     +44 (0) 1865 610737
web     http://web.comlab.ox.ac.uk/people/Ramon.CaseroCanas
photos  http://www.flickr.com/photos/rcasero/