<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 3.2//EN">
<HTML>
<HEAD>
<META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=US-ASCII">
<META NAME="Generator" CONTENT="MS Exchange Server version 5.5.2658.24">
<TITLE>RE: [Insight-users] ITK ROAD MAP 2005-2006 : Call for feedback</TITLE>
</HEAD>
<BODY>
<P><FONT SIZE=2>dear david:</FONT>
<BR><FONT SIZE=2>we have an A2D2 contract to implement graph cuts, and hope to have something to show soon.</FONT>
</P>
<P><FONT SIZE=2>jim</FONT>
<BR><FONT SIZE=2> </FONT>
</P>
<P><FONT SIZE=2>> -----Original Message-----</FONT>
<BR><FONT SIZE=2>> From: insight-users-bounces+gee=rad.upenn.edu@itk.org </FONT>
<BR><FONT SIZE=2>> [<A HREF="mailto:insight-users-bounces+gee=rad.upenn.edu@itk.org">mailto:insight-users-bounces+gee=rad.upenn.edu@itk.org</A>] </FONT>
<BR><FONT SIZE=2>> Sent: Tuesday, May 31, 2005 2:59 PM</FONT>
<BR><FONT SIZE=2>> Cc: insight-users@itk.org</FONT>
<BR><FONT SIZE=2>> Subject: RE: [Insight-users] ITK ROAD MAP 2005-2006 : Call </FONT>
<BR><FONT SIZE=2>> for feedback</FONT>
<BR><FONT SIZE=2>> </FONT>
<BR><FONT SIZE=2>> hi gang,</FONT>
<BR><FONT SIZE=2>> </FONT>
<BR><FONT SIZE=2>> one additional comment--if MRF support in ITK is to be </FONT>
<BR><FONT SIZE=2>> enhanced, i'd strongly recommend adding a class to implement </FONT>
<BR><FONT SIZE=2>> the graph cuts approach to MRF optimization, which published </FONT>
<BR><FONT SIZE=2>> articles, and my own experience, have shown to be </FONT>
<BR><FONT SIZE=2>> consistently better than the old greedy updating methods.</FONT>
<BR><FONT SIZE=2>> </FONT>
<BR><FONT SIZE=2>> -dh</FONT>
<BR><FONT SIZE=2>> -david haynor (haynor@u.washington.edu)</FONT>
<BR><FONT SIZE=2>> department of radiology</FONT>
<BR><FONT SIZE=2>> box 356004</FONT>
<BR><FONT SIZE=2>> university of washington</FONT>
<BR><FONT SIZE=2>> seattle, WA 98195</FONT>
<BR><FONT SIZE=2>> (206) 543-3320</FONT>
<BR><FONT SIZE=2>> </FONT>
<BR><FONT SIZE=2>> On Tue, 31 May 2005, Sayan Pathak wrote:</FONT>
<BR><FONT SIZE=2>> </FONT>
<BR><FONT SIZE=2>> > Hi Zach,Martin and other interested users/developers, It is </FONT>
<BR><FONT SIZE=2>> great to </FONT>
<BR><FONT SIZE=2>> > hear about emerging interest in enhancing support for MRF </FONT>
<BR><FONT SIZE=2>> filters in </FONT>
<BR><FONT SIZE=2>> > ITK. I would like to add a couple of things to Jim's mail.</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > The itk::MRFImageFilter class implements Besag's classical </FONT>
<BR><FONT SIZE=2>> MRF filter </FONT>
<BR><FONT SIZE=2>> > where the classification labels are iteratively updated . </FONT>
<BR><FONT SIZE=2>> This class </FONT>
<BR><FONT SIZE=2>> > was envisioned to be a base implementation, which could be </FONT>
<BR><FONT SIZE=2>> extended to </FONT>
<BR><FONT SIZE=2>> > other MRF realizations. One idea would be to capture the different </FONT>
<BR><FONT SIZE=2>> > classes of MRFs the ITK users community would like to have. </FONT>
<BR><FONT SIZE=2>> It would </FONT>
<BR><FONT SIZE=2>> > be very helpful to hear from someone in the community who </FONT>
<BR><FONT SIZE=2>> has worked </FONT>
<BR><FONT SIZE=2>> > in this area and is willing to share their experiences.</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > Thanks,</FONT>
<BR><FONT SIZE=2>> > Sayan</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > Date: Tue, 31 May 2005 11:56:07 -0400</FONT>
<BR><FONT SIZE=2>> > From: "Miller, James V (Research)" <millerjv@crd.ge.com></FONT>
<BR><FONT SIZE=2>> > Subject: RE: [Insight-users] ITK ROAD MAP 2005-2006 : Call for</FONT>
<BR><FONT SIZE=2>> > feedback</FONT>
<BR><FONT SIZE=2>> > To: "Zachary Pincus" <zpincus@stanford.edu>, "ITK mailing"</FONT>
<BR><FONT SIZE=2>> > <insight-users@itk.org></FONT>
<BR><FONT SIZE=2>> > Message-ID:</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > <FA26BEF1EA775E4584FB34B91E14A1C4B419A2@SCHMLVEM01.e2k.ad.ge.com></FONT>
<BR><FONT SIZE=2>> > Content-Type: text/plain; charset="Windows-1252"</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > Zachary,</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > Is the current itk::MRFImageFilter not sufficient? Does it </FONT>
<BR><FONT SIZE=2>> need to be </FONT>
<BR><FONT SIZE=2>> > refactored to accomadate different MRF algorithms?</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > I only used the MRFImageFilter once. It seemed to perform </FONT>
<BR><FONT SIZE=2>> as I would </FONT>
<BR><FONT SIZE=2>> > have expected (from reading some of the literature).</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > One thing I think ITK probably needs are more techniques </FONT>
<BR><FONT SIZE=2>> for learning </FONT>
<BR><FONT SIZE=2>> > the pdf's for each class of material.</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > Jim</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > -----Original Message-----</FONT>
<BR><FONT SIZE=2>> > From: insight-users-bounces+millerjv=crd.ge.com@itk.org</FONT>
<BR><FONT SIZE=2>> > </FONT>
<BR><FONT SIZE=2>> [<A HREF="mailto:insight-users-bounces+millerjv=crd.ge.com@itk.org">mailto:insight-users-bounces+millerjv=crd.ge.com@itk.org</A>]On </FONT>
<BR><FONT SIZE=2>> Behalf Of </FONT>
<BR><FONT SIZE=2>> > Zachary Pincus</FONT>
<BR><FONT SIZE=2>> > Sent: Saturday, May 28, 2005 7:37 PM</FONT>
<BR><FONT SIZE=2>> > To: ITK mailing</FONT>
<BR><FONT SIZE=2>> > Subject: Re: [Insight-users] ITK ROAD MAP 2005-2006 : Call for </FONT>
<BR><FONT SIZE=2>> > feedback</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > Hi all,</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > After looking over the 2005-2006 ITK roadmap, I've also got </FONT>
<BR><FONT SIZE=2>> a couple </FONT>
<BR><FONT SIZE=2>> > of questions/comments on the machine learning aspects.</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > Specifically, to what ends are classification algorithms (e.g. </FONT>
<BR><FONT SIZE=2>> > gaussian mixture models, k-nearest neighbors, putative </FONT>
<BR><FONT SIZE=2>> neural networks </FONT>
<BR><FONT SIZE=2>> > or SVMs) present in ITK? It strikes me that one major use of such </FONT>
<BR><FONT SIZE=2>> > algorithms in medical imaging is for classification of image pixels </FONT>
<BR><FONT SIZE=2>> > into various tissue types, e.g. grey matter vs. white matter.</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > If this is the case, I would think that adding Markov Random Field </FONT>
<BR><FONT SIZE=2>> > capabilities to ITK would be a big win. Basically, MRFs </FONT>
<BR><FONT SIZE=2>> allow users to </FONT>
<BR><FONT SIZE=2>> > add priors about the *spatial* distribution of various pixel types </FONT>
<BR><FONT SIZE=2>> > into the classification process. For example, a single </FONT>
<BR><FONT SIZE=2>> isolated pixel </FONT>
<BR><FONT SIZE=2>> > initially labeled as "grey matter" in a blob of white matter might </FONT>
<BR><FONT SIZE=2>> > (depending on the priors) be considered an unlikely </FONT>
<BR><FONT SIZE=2>> configuration and </FONT>
<BR><FONT SIZE=2>> > thus be re-labeled in the final MRF configuration. Such spatial </FONT>
<BR><FONT SIZE=2>> > considerations are ignored by traditional classifiers.</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > Because spatial information is so important, and MRFs are a </FONT>
<BR><FONT SIZE=2>> relatively </FONT>
<BR><FONT SIZE=2>> > easy way to add simple spatial priors, they have become </FONT>
<BR><FONT SIZE=2>> quite popular </FONT>
<BR><FONT SIZE=2>> > in the image processing literature. I think filters to estimate the </FONT>
<BR><FONT SIZE=2>> > MAP MRF given an input "label images" (e.g the results of pixel-wise</FONT>
<BR><FONT SIZE=2>> > classification) would be a very valuable addition, </FONT>
<BR><FONT SIZE=2>> especially if the </FONT>
<BR><FONT SIZE=2>> > stable of pixel classification methods in ITK is to expand.</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > Now, I haven't described in too much detail how MRF models actually </FONT>
<BR><FONT SIZE=2>> > work. A Google Scholar search for "Markov random field image" will </FONT>
<BR><FONT SIZE=2>> > show the breadth of utilization of MRFs in the imaging literature. </FONT>
<BR><FONT SIZE=2>> > Here is a good introduction to MRF segmentation, with specific </FONT>
<BR><FONT SIZE=2>> > reference to MRI</FONT>
<BR><FONT SIZE=2>> > images:</FONT>
<BR><FONT SIZE=2>> > Segmentation of brain MR images through a hidden Markov </FONT>
<BR><FONT SIZE=2>> random field </FONT>
<BR><FONT SIZE=2>> > model and the expectation-maximization algorithm.</FONT>
<BR><FONT SIZE=2>> > Y Zhang, M Brady, S Smith - IEEE Trans Med Imaging, 2001 </FONT>
<BR><FONT SIZE=2>> > <A HREF="http://www.cvmt.dk/~hja/teaching/cv/HMRF_EM_BRAIN.pdf" TARGET="_blank">http://www.cvmt.dk/~hja/teaching/cv/HMRF_EM_BRAIN.pdf</A></FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > I would be happy to discuss at (much) more length how a MRF </FONT>
<BR><FONT SIZE=2>> > "classification cleanup" filters could be implemented in </FONT>
<BR><FONT SIZE=2>> ITK, if there </FONT>
<BR><FONT SIZE=2>> > is any interest in these methods.</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > Zach</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> > On May 27, 2005, at 2:46 PM, Lino Ramirez wrote:</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> >> Hi Luis and ITK Users/Developers,</FONT>
<BR><FONT SIZE=2>> >></FONT>
<BR><FONT SIZE=2>> >> I had a brief look at the ITK roadmap 2005-2006. It looks quite </FONT>
<BR><FONT SIZE=2>> >> impressive. I cannot wait until having available all these tools in</FONT>
<BR><FONT SIZE=2>> > one</FONT>
<BR><FONT SIZE=2>> >> single package ;-)</FONT>
<BR><FONT SIZE=2>> >></FONT>
<BR><FONT SIZE=2>> >> I have some small comments/questions about functionalities I would </FONT>
<BR><FONT SIZE=2>> >> like to see in the toolkit.</FONT>
<BR><FONT SIZE=2>> >></FONT>
<BR><FONT SIZE=2>> >> I noticed that Neural Networks will be added to the toolkit. Are </FONT>
<BR><FONT SIZE=2>> >> there</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> >> any</FONT>
<BR><FONT SIZE=2>> >> plans for adding a Support Vector Machines (SVM) [1] </FONT>
<BR><FONT SIZE=2>> implementation?</FONT>
<BR><FONT SIZE=2>> >> SVM</FONT>
<BR><FONT SIZE=2>> >> have been used successfully in a variety of applications </FONT>
<BR><FONT SIZE=2>> that could </FONT>
<BR><FONT SIZE=2>> >> be</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> >> of</FONT>
<BR><FONT SIZE=2>> >> interest to the ITK community (see [2] for some sample </FONT>
<BR><FONT SIZE=2>> applications).</FONT>
<BR><FONT SIZE=2>> >> Moreover, it is always good to have a machine learning </FONT>
<BR><FONT SIZE=2>> approach that</FONT>
<BR><FONT SIZE=2>> > is</FONT>
<BR><FONT SIZE=2>> >> similar to the neural networks in architecture but that uses a </FONT>
<BR><FONT SIZE=2>> >> different learning strategy. In this way, one could try the two of </FONT>
<BR><FONT SIZE=2>> >> them and determine which one is more appropriate for a particular </FONT>
<BR><FONT SIZE=2>> >> dataset.</FONT>
<BR><FONT SIZE=2>> >> Sometimes, in datasets in which the neural networks fail the SVM </FONT>
<BR><FONT SIZE=2>> >> succeed and vice versa.</FONT>
<BR><FONT SIZE=2>> >></FONT>
<BR><FONT SIZE=2>> >> Are there any plans (even in the very long term) to add </FONT>
<BR><FONT SIZE=2>> support for </FONT>
<BR><FONT SIZE=2>> >> Fuzzy Sets [3], Fuzzy Geometry [4], and Fuzzy Spatial </FONT>
<BR><FONT SIZE=2>> Relations [5] </FONT>
<BR><FONT SIZE=2>> >> between objects in an image. I think these concepts would be </FONT>
<BR><FONT SIZE=2>> >> invaluable in the future of medical image analysis. For </FONT>
<BR><FONT SIZE=2>> example, when </FONT>
<BR><FONT SIZE=2>> >> we want to measure geometric properties in objects in an image, we </FONT>
<BR><FONT SIZE=2>> >> find that generally</FONT>
<BR><FONT SIZE=2>> > the</FONT>
<BR><FONT SIZE=2>> >> objects are not crisply defined (due to errors during the</FONT>
<BR><FONT SIZE=2>> > segmentation,</FONT>
<BR><FONT SIZE=2>> >> errors in the acquisition of the image, or errors in the </FONT>
<BR><FONT SIZE=2>> definition </FONT>
<BR><FONT SIZE=2>> >> of</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> >> the</FONT>
<BR><FONT SIZE=2>> >> object -where do the ribs start and the vertebrae end in a spine </FONT>
<BR><FONT SIZE=2>> >> X-ray).</FONT>
<BR><FONT SIZE=2>> >> In this case, fuzzy geometry could be used to compute the object </FONT>
<BR><FONT SIZE=2>> >> properties. Another example would be in the identification </FONT>
<BR><FONT SIZE=2>> of objects </FONT>
<BR><FONT SIZE=2>> >> in the images. For instance, in the internal brain structures, the </FONT>
<BR><FONT SIZE=2>> >> right caudate nucleus should be closer to the right </FONT>
<BR><FONT SIZE=2>> lateral ventricle </FONT>
<BR><FONT SIZE=2>> >> than</FONT>
<BR><FONT SIZE=2>> > to</FONT>
<BR><FONT SIZE=2>> >> the left lateral ventricle. Fuzzy spatial relations with </FONT>
<BR><FONT SIZE=2>> the help of </FONT>
<BR><FONT SIZE=2>> >> fuzzy logic [6] could be used to develop a system that </FONT>
<BR><FONT SIZE=2>> makes use of </FONT>
<BR><FONT SIZE=2>> >> that piece of information to identify right lateral ventricle.</FONT>
<BR><FONT SIZE=2>> >></FONT>
<BR><FONT SIZE=2>> >> Well, those are my two picks ;-)</FONT>
<BR><FONT SIZE=2>> >></FONT>
<BR><FONT SIZE=2>> >> I am looking forward to any comment</FONT>
<BR><FONT SIZE=2>> >></FONT>
<BR><FONT SIZE=2>> >> Take care</FONT>
<BR><FONT SIZE=2>> >></FONT>
<BR><FONT SIZE=2>> >> Lino</FONT>
<BR><FONT SIZE=2>> >></FONT>
<BR><FONT SIZE=2>> >> [1] C. Cortes and V. Vapnik, "Support-Vector Networks," Machine </FONT>
<BR><FONT SIZE=2>> >> Learning, vol. 20, pp. 273-297, 1995 [2] </FONT>
<BR><FONT SIZE=2>> >> <A HREF="http://www.clopinet.com/isabelle/Projects/SVM/applist.html" TARGET="_blank">http://www.clopinet.com/isabelle/Projects/SVM/applist.html</A></FONT>
<BR><FONT SIZE=2>> >> [3] L.A. Zadeh, "Fuzzy sets," Information and Control, vol. 8, pp.</FONT>
<BR><FONT SIZE=2>> >> 38-352,</FONT>
<BR><FONT SIZE=2>> >> 1965</FONT>
<BR><FONT SIZE=2>> >> [4] A. Rosenfeld, "Fuzzy geometry: An updated overview," </FONT>
<BR><FONT SIZE=2>> Information </FONT>
<BR><FONT SIZE=2>> >> Sciences, vol. 110, pp. 127-133, 1998 [5] I. Bloch, "Fuzzy spatial </FONT>
<BR><FONT SIZE=2>> >> relationships for image processing and</FONT>
<BR><FONT SIZE=2>> >> interpretation: a review," Image and Vision Computing, vol. 23, pp.</FONT>
<BR><FONT SIZE=2>> >> 89-110, 2005</FONT>
<BR><FONT SIZE=2>> >> [6] L.A. Zadeh, "Outline of a new approach to the analysis </FONT>
<BR><FONT SIZE=2>> of complex </FONT>
<BR><FONT SIZE=2>> >> systems and decision processes," IEEE Transactions on Systems, Man,</FONT>
<BR><FONT SIZE=2>> > and</FONT>
<BR><FONT SIZE=2>> >> Cybernetics, vol. SMC-3, no. 1, pp. 28-44, 1973</FONT>
<BR><FONT SIZE=2>> >></FONT>
<BR><FONT SIZE=2>> >>> A first draft of the road map for ITK development/maintenance has </FONT>
<BR><FONT SIZE=2>> >>> been crafted for the period of September 2005 - September 2006.</FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>> You will find this draft as a link to the Oversight Committee page</FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>> <A HREF="http://www.itk.org/Wiki/ITK_Oversight_Committee" TARGET="_blank">http://www.itk.org/Wiki/ITK_Oversight_Committee</A></FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>> More specifically at</FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>> <A HREF="http://www.itk.org/Wiki/ITK_Roadmap_2005_2006" TARGET="_blank">http://www.itk.org/Wiki/ITK_Roadmap_2005_2006</A></FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>> The purpose of this road map is to plan for features and </FONT>
<BR><FONT SIZE=2>> >>> functionalities to be included in ITK in the near/medium </FONT>
<BR><FONT SIZE=2>> term (1 to </FONT>
<BR><FONT SIZE=2>> >>> 2 years).</FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>> The addition of these features should make of ITK a </FONT>
<BR><FONT SIZE=2>> better tool for </FONT>
<BR><FONT SIZE=2>> >>> supporting your efforts in medical research, and development of </FONT>
<BR><FONT SIZE=2>> >>> medical applications.</FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>> The road map also includes the maintenance tasks to be </FONT>
<BR><FONT SIZE=2>> undertaken in </FONT>
<BR><FONT SIZE=2>> >>> ITK. This may involve refactoring of classes, deprecation of </FONT>
<BR><FONT SIZE=2>> >>> classes, additional testing, additional coverage, improvements on </FONT>
<BR><FONT SIZE=2>> >>> tutorials</FONT>
<BR><FONT SIZE=2>> > and</FONT>
<BR><FONT SIZE=2>> >>> so on.</FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>> Please let us know of the features that you would like to </FONT>
<BR><FONT SIZE=2>> see in ITK </FONT>
<BR><FONT SIZE=2>> >>> in the upcoming future, and what points of the toolkit </FONT>
<BR><FONT SIZE=2>> you consider </FONT>
<BR><FONT SIZE=2>> >>> that can be improved in order to better server the community.</FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>> Thanks</FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>></FONT>
<BR><FONT SIZE=2>> >>> Luis</FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> ></FONT>
<BR><FONT SIZE=2>> _______________________________________________</FONT>
<BR><FONT SIZE=2>> Insight-users mailing list</FONT>
<BR><FONT SIZE=2>> Insight-users@itk.org</FONT>
<BR><FONT SIZE=2>> <A HREF="http://www.itk.org/mailman/listinfo/insight-users" TARGET="_blank">http://www.itk.org/mailman/listinfo/insight-users</A></FONT>
<BR><FONT SIZE=2>> </FONT>
</P>
</BODY>
</HTML>