Title: "Cancer detection by spectral methods"
Speaker:Samy Tindel, Departement de Mathmatiques Universit Henri Poincar (Nancy) , Institut Elie Cartan (Nancy), France
Place: LILY G126; April 7, 2009, Tuesday, 4:30pm


The possibility of detecting cancer by means of a spectroscopic instrumentation has emerged during the last past years. This concerns mainly hollow organs such as pancreas, oesophagus skin, or bladder. In those cases, one expects in the end to produce a quick, sure and non invasive cancer diagnosis.

The current study is focused on an experimental protocol for bladder cancer detection realized on rats. The data we are faced with consists in two different spectrums (called autofluorescence and diffuse reflectance spectrums), made of many points. As usual in this kind of experiment, the size of the sample is rather small, and in particular smaller than the data dimension.

In this difficult context, our aim is twofold: (1) Select a reasonable number of spectrum variables allowing to discriminate between cancer and non cancer tissues. (2) According to the variables we have selected, try to get a good sensibility and specificity on our data. We shall see how to use some modern tools of data analysis (regularized discriminant analysis, logistic regression, support vector machines) in order to reach this goal.

Associated reading:
E. Pery, W.C.P.M. Blondel, J. Didelon, A. Leroux, and F. Guillemin. 2002. Simultaneous Characterization of Optical and Rheological properties of Carotid Arteries via Bimodal Spectroscopy: Experimental and Simulation Results. JOURNAL OF LATEX CLASS FILES, VOL. 1, NO. 11, NOVEMBER.

Click here for a full schedule of BIOINFORMATICS SEMINARS, past and present.