Semantic segmentation of cell nuclei and cytoplasms in microscopy images

Christian Eschen, Søren Blaaberg, Ole Winther, Rasmus Reinhold Paulsen

Keywords: Fully convolutional neural networks, semantic segmentation, deep learning, microscopy imaging, fluorescent imaging

TL;DR: Semantic segmentation of cell nuclei and cytoplasms in microscopy images

Abstract: Microscopy imaging of cell nuclei and cytoplasms is a powerfull technique for research, diagnosis and drug discovery. However, the use of fluorescent microscopy imaging for cell nuclei and cytoplasms labeling is time consuming and inconvenient for several reasons,thus there is a lack of fast and accurate methods for prediction of fluorescence cell nuclei and cytoplasms from bright-field microscopy imaging. We present a method for labeling bright-field images using convolutional neural networks. We investigate different convolutional neural network architectures for cell nuclei and cytoplasms prediction. Using the DeepLabv3+, we found relative impressive results with a 5-fold cross validation dice coefficient equal to 0.9503 as well as meaningful segmentation maps. This work shows proof of concept regarding microscopy fluorescence labeling of cell nuclei and cytoplasms using bright-field images

Code Of Conduct: I have read and accept the code of conduct.