Automatic Segmentation Framework for Fluorescence in Situ Hybridization Cancer Diagnosis

Abstract : In this paper we address a problem of HER2 and CEN-17 reactions detection in fluorescence in situ hybridization images. These images are very often used in situation where typical biopsy examination is not able to provide enough information to decide on the type of treatment the patient should undergo. Here the main focus is placed on the automatization of the procedure. Using an unsupervised neural network and principal component analysis, we present a segmentation framework that is able to keep the high segmentation accuracy. For comparison purposes we test the neural network approach against an automatic threshold method.