[Insight-users] Sample interface and expectation maximization mixture model estimation

Nicolas Gallego nicgallego at gmail.com
Wed Jul 14 10:47:43 EDT 2010


Dear ITK users community,

I have been using ITK library to implement an algorithm of mass detection in
mammography based on gaussian mixture modelling of texture as part of my
master thesis project.

At first we tried to used the library objects
itk::Statistics::ExpectationMaximizationMixtureModelEstimator
 with the itk::Statistics::ListSample sample interface as explained in the
software guide. We tried to pass by means of the sample interface the
measurement vectors to the EM algorithm. For our purpouse those vectors are
the gray levels of the pixels from the local neighborhoods of each target
pixel. We get those sample vectors by means of shaped neightborhood iterator
itk::ShapedNeighborhoodIterator without problems.

Our problem is then of memory allocation. We tried to allocate every mv
vector by vector inside the sample object by means of its smart pointer as:

sample->PushBack(mv) ;

but soon our system ran out of memory, as expected since mammograms are
about 12 MPixels and the mv size is 145.

We decided to implement by ourselves the method using image iterators,
linear algebra and gaussian density functions. After some programming and
debuging we have got a working prototype, but we would like to ask for ideas
to work around the mentioned problem. Is anybody supporting the EM
implementation in ITK or has anybody solved a similar issue before?. Any
advice is appreciated.

We really would like to compare our prototype's output with the ITK tested
implementation. Also let us know, if it is appropriate, how can we
contribute to the development of ITK.

Thank you very much,

Sincerely,

-- 
Nicolás Gallego Ortiz
nicgallego at ieee.org
Departamento de Ingeniería Electrónica y Telecomunicaciones
Universidad de Antioquia
Medellín, Colombia
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20100714/898907b8/attachment.htm>


More information about the Insight-users mailing list