Computer-aided Evaluation of Her-2 Status in Fluorescent in Situ Hybridization Images

The analysis of Her-2/neu status is an effective indicator for the diagnosis of several types of breast carcinomas. Conventional evaluation is a difficult task since it involves manual counting of dots in multiple fluorescent in situ hybridization (FISH) images. In this paper we present a multistage algorithm for the automated evaluation of Her-2/neu status by the analysis of FISH images from breast carcinomas. The algorithm focuses on the detection of FISH spots and on the cell nuclei segmentation in order to perform overall case classification as positive or negative. Spots detection includes mainly a top-hat filtering stage, a binary thresholding, a 3D template matching and a grey level contrast evaluation. Nuclei segmentation consists of a non-linear blue channel correction step, a global thresholding by Otsu algorithm, a grey level hole classification by a geometric rule and of the marked watershed transform using local h-dome maxima as markers. By the measurement of the FISH signals ratio per cell nucleus we perform the classification of cases. The performances of the algorithm were evaluated with receiver operating characteristic (ROC) analysis.