Nuclei detection and segmentation of fluorescence microscopy images using three dimensional convolutional neural networks

David Joon Ho, Chichen Fu, Paul Salama, Kenneth W. Dunn, Edward J. Delp

Abstract: Recent advance in fluorescence microscopy enables acquisition of 3D image volumes with better quality and deeper penetration into tissue. In this paper, we describe a 3D method which can detect and segment nuclei in fluorescence microscopy images using convolutional neural networks (CNN). For nuclei detection, a 3D adaptive histogram equalization, a 3D distance transform, and a 3D classification CNN are used to find centers of nuclei. For nuclei segmentation, a 3D segmentation CNN is used which is trained from automatically generated synthetic microscopy volumes and their synthetic ground truth volumes. Our method outperforms other 3D segmentation methods and can detect nuclei successfully on multiple data sets.