[Statlist] Seminar ueber Statistik

[Christina Kuenzli]

                       ETH and University of Zurich 
                                 Proff. 
A.D. Barbour - P. Buehlmann - F. Hampel - H.R. Kuensch - S. van de Geer

**************************************************************************     
           We are pleased to announce the following talks
**************************************************************************

Friday, January 13, 2005, 15.15, HG E 41

       Wandering and outlying birds: A statistical analysis
             Special lecture in honor of Frank Hampel

       Chris Field from Dalhousie University, Halifax
**************************************************************************
Friday, January 20, 2005, 15.15, LEO C 15

       Automated analysis of magnetic resonance spectroscopic images

       B.M. Kelm and F.A. Hamprecht
       Multidimensional Image Processing, IWR, University of Heidelberg
     
Magnetic Resonance Spectroscopic Imaging (MRSI) allows to detect cell
metabolites in vivo.  As metabolic concentrations change
characteristically in pathologic tissue, MRSI can be used for the
detection and localization of tumors.

The common approach to the evaluation of MRSI data comprises two
steps. First, metabolite concentrations are deduced from the MR signal
based on a physical model, i.e. the spectrum is quantified.  In the
second step, a decision rule is applied to the estimated concentrations
in order to answer a diagnostic question. The main problem with this
approach is that the quantification is very sensitive to noise, thus
requiring human supervision. We therefore apply pattern recognition
methods directly to the high-dimensional and noisy spectral pattern, 
i.e. without the prior quantification step. Superior results are obtained 
with (generalized) linear models which also allow a consistent 
interpretation. Nonlinear approaches such as SVMs and random forests 
can lead to a further slight improvement of results, at the expense 
of interpretability.

In this talk, we present different approaches to the feature extraction
in MRSI data and compare various machine learning techniques for the
prediction of local tumor probability.

We will also give a brief overview of the other types of problems
addressed -- mainly in the field of industrial quality control -- in the
group for multidimensional image processing at the IWR.       

________________________________________________________
Christina Kuenzli            <kuenzli at stat.math.ethz.ch>
Seminar fuer Statistik      
Leonhardstr. 27,  LEO D11      phone: +41 (0)44 632 3438         
ETH-Zentrum,                   fax  : +41 (0)44 632 1228 
CH-8092 Zurich, Switzerland        http://stat.ethz.ch/~