Deep learning for nuclear image analysis and application to 3D plant nuclei

International audienceThe continuous development of microscopy has led biologists to have access to large high-resolution 2D and 3D image datasets. Automatic analysis of cellular and nuclear images has become an important challenge in the bioimaging field. To help biologists extract information from these images, tools have been designed to count objects in the image, study object type, their localization or morphology. The current state-of-the-art is led by deep learning methods. Their development relies on the availability of large data sets, on enhancing graphical processing units (GPU) of computers and on developments of new methodologies such as convolutional neural networks (CNN). However, non-IT users may experience difficulties when trying to use these on their own images. This short paper contains our first results after reviewing the state-of-the-art methods in the domain. It first introduces the current difficulties when working with bioimages, then lists the existing datasets for nuclear images analysis. It then exposes some of the easy-to-use tools for bioimaging and points out the different problems related to their use. It finally presents a new dataset for 3D images of plant nuclei that is designed for benchmarking purposes. This results should shortly be published in an journal of biology