Convolutional Neural Networks for Dot Counting in Fluorescence in Situ Hybridization Imaging

During breast cancer a small tissue sample is extracted and evaluated to estimate the malignancy of the growth and the possible treatment. When a difficult case is diagnosed, a more accurate diagnosis based on fluorescence in situ hybridization imaging is performed where HER2 and CEN-17 reactions are determined. In this paper we address a problem of immunohistochemistry reaction detection that are often tested when it is difficult to decide on the type of treatment the patient should undergo. Here we describe a segmentation framework adopting convolutional neural networks that are able to classify image pixels into HER2 and CEN-17 reactions respectively. Using the above mentioned framework, the proposed system is able to keep the high segmentation accuracy.