6th ESACP Congress, Heidelberg, April 7-11, 1999

A127
MULTIPARAMETRIC CELL SORTING FOR DNA MEASUREMENT OF SOLID TUMORS. PART II : SELECTION OF THE MOST INFORMATIVE PARAMETERS.
Signolle JP 1, Herlin P 2, Plancoulaine B 1, Masson E 3, Elmoataz A 3, Lefranc F 3, Boudry C 4, Hémery A 1, Bloyet D 3

1) Départment Mesures Physiques IUT Caen, 2) Laboratoire d'Anatomie Pathologique, Centre de Lutte Contre le Cancer François Baclesse, Caen, 3) Greyc - ISMRA, UPRES-A CNRS 6072, Caen, 4) Lermat - ISMRA, UPRES-A CNRS 6004, Caen, 1-4) Pôle Traitement et Analyse d'Images de Basse Normandie, France

The automatic sorting of nuclei populations for DNA analysis of solid tumors can be obtained by the Principal Component Analysis of 38 parameters, measured or calculated on each nucleus (cf Part I). The use of so many paramaters may not be justified, and could alter the quality of the analysis. So, in order to improve the classification method, parameters have been arranged according to their degree of importance, by using two different approaches : the Knock Out method (*) and the critical analysis of parameters, taken one by one during the Principal Component Analysis (looking for redundant parameters, aberrantly distributed parameters and parameters notably implied only in high rank components). The results obtained by the two approaches are in accordance with each other and put in light 19 out of 38 pertinent parameters.
* Goresnic A, Rotman SR: Texture classification using the cortex transform. Graphical Models and Image Processing 54:329-392(1992)