||The major goal of the whole project, to which thesis also belongs, is making automated fluorescence microscope system for entirely diagnostic purpose, while this paper refers on estimating the ratio of HER-2 and CEN-17 signals per nucleus. Program itself cannot give a final diagnose or tell whether a subject patient is healthy or needs to be treated against breast cancer. The basic requirement was creating software that would work independently from MatLab functions available in various Toolboxes, in order to obtain lower cost of the whole system.
After autofluorescence has been removed from original image, segmentation of spectrally unmixed image was done by using Ridler’s isodata thresholding algorithm which is where mask of the nucleus comes from. Some binary morphological operations have also been used in order to make nuclei look more regular in shape, the next step, which is separating the touching nuclei, is achieved by using watershed segmentation (CBA, 2002). For purpose of dot detection, modified “dot label” algorithm (Netten et al, 1996) which uses a variable threshold level is developed.