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

A122
QUALITY CONTROL OF AN AUTOMATIC IMAGE ANALYSIS PROCEDURE FOR QUANTIFICATION OF IMMUNOLABELLED CANCER CELL NUCLEI : EVALUATION OF SEVERAL STRATEGIES DEDICATED TO COLOUR CHARACTERISATION OF LABELLED NUCLEI IN MAMMARY CARCINOMA.
Schüpp S 1, Herlin P 2, Huet G 2, Duigou F 2, Lebeau C 2, Chasle J 2, Elmoataz A 1, Bloyet D 1

1) GREYC - ISMRA, UPRES-A CNRS 6072, Caen, 2) Laboratoire d'Anatomie Pathologique, Centre de Lutte Contre le Cancer François Baclesse, Caen, 1-2) Pôle Traitment et Analyse d'Images de Basse Normandie, France

Automatic quantification of immunolabelled nuclei by image analysis in breast carcinoma is based on three sequences: segmentation of all nuclei, identification of cancer cell clumps and characterisation of immunolabelled nuclei inside these clumps, in order to determine numerical density and volume fraction of positive nuclei. Several strategies can be used for characterisation of immunolabelled nuclei; they are more or less sophisticated and time consuming, the simplest and most popular one is based on a simple thresholding of the blue component of the colour image, the most sophisticated methods use true colour segmentation. The final choice of the most efficient stategy must be guided by the results of a quality control procedure. Such a quality control has been carried out in our laboratory, referring to a test base of two hundred and ninety eight images, from 18 invasive and in situ breast carcinomas and for the segmentation of cancer cell clusters, which have been obtained using five different strategies..