[ome-devel] State of the art pattern analysis comes to OME

[Tom Macura]

Over the course of three years, Nikita Orlov, Josiah Johnston, Tom  
Macura, Lior Shamir and Mark Eckley in Ilya Goldberg's group at the  
NIA/NIH have been hard at work developing generalized pattern  
analysis for cell biology applications.  The result of this work is a  
robust and accurate image classifier called WND-CHARM that can be  
applied to many types of images (even ones that you know for sure  
could never be processed analytically).

Other than happily classifying scary images from DIC, phase-contrast,  
cytoskeletal morphologies, sub-cellular organelles and other  
seemingly intractable things cell biologists see every day, WND-CHARM  
is also quite accidentally one of the top ranked classifiers for face  
recognition.  Whoops!

At this time, we have a preliminary version integrated into OME that  
represents a complete automated solution - feature extraction,  
training, and classification.  It leverages the UI developed for  
Categories/Category groups to define training classes, and uses the  
same data model for classification results.  An important aspect of  
the design is that the computationally intense feature extraction  
(CHARM) can be used by classifiers other than WND-5 without having to  
recompute them.

This is our first successful attempt at this, so we wanted to  
announce it to allow people to try it out.  In its present form, WND- 
CHARM relies entirely on MATLAB, so to use it you will need a MATLAB  
license and install the OME-MATLAB connector.  Lior has completely  
reimplemented WND-CHARM in C/C++, and this is the version that will  
be in the release, but its not fully integrated into OME yet.

This is very much a preview to get comments and feedback from the  
community, so this functionality is only available from CVS-HEAD  
(also our paper is still in review).  Tomasz has written up a howto,  
and provided handy scripts to make it all work:
http://users.openmicroscopy.org.uk/~tmacur1/OME-WEBSITE/howto/wnd- 
charm-ome.html

It should be noted that the feature extraction is very computationaly  
intensive, so if you want to do this seriously, you really should be  
doing it on a distributed analysis system (DAE) with as many high-end  
cores as you can manage.  We've been testing this quite a bit lately  
as well, and its become quite robust and performant.  We would also  
greatly appreciate feedback on the DAE from the community.

We're very excited because this application really puts all of the  
pieces of OME through their paces, and we think will eventually be  
seen as a minor revolution in cell biology.

Enjoy!
The OME Team