Neural cell segmentation in large-scale 3D color fluorescence microscopy images for developemental neuroscience

The cells composing brain tissue, neurons, and glia, form extraordinarily complex networks that support cognitive functions. Understanding the organization and development of these networks requires quantitative data resolved at the single cell level. To this aim, we apply novel large-scale 3D multicolor microscopy methodologies in combination with 'Brainbow', a transgenic approach enabling to label neural cells with diverse combinations of spectrally distinct fluorescent proteins. In this paper, we present a pipeline based on Convolutional Neural Network (CNN) to detect and segment individual astrocytes, the main type of glial cells of the brain, and map the domains occupied by their fine processes. This bioimage analysis approach successfully handles the challenging variety of astrocyte shape, color, size and their overlap with background elements. Our method shows significant improvement compared with classical techniques, opening the way to varied biological inquiries.