An end-to-end pipeline based on open source deep learning tools for reliable analysis of complex 3D images of Medaka ovaries

Computational analysis of bio-images by deep learning (DL) algorithms has made exceptional progress in recent years and has become much more accessible to non-specialists with the development of ready-to-use tools. The study of oogenesis mechanisms and female reproductive success in fish has also recently benefited from the development of efficient three-dimensional (3D) imaging protocols on entire ovaries. Such large datasets have a great potential for the generation of new quantitative data on oogenesis but are, however, complex to analyze due to imperfect fluorescent signals and the lack of efficient image analysis workflows. Here, we applied two open-source DL tools, Noise2Void and Cellpose, to analyze the oocyte content of medaka ovaries at larvae and adult stages. These tools were integrated into end-to-end analysis pipelines that include image pre-processing, cell segmentation, and image post-processing to filter and combine labels. Our pipelines thus provide effective solutions to accurately segment complex 3D images of entire ovaries with either irregular fluorescent staining or low autofluorescence signal. In the future, these pipelines will be applicable to extensive cellular phenotyping in fish for developmental or toxicology studies. Summary statementAn accessible image analysis method for biologists, which includes easy-to-use deep learning algorithms, designed for accurate quantitative measurement of ovarian content from complex 3D fluorescent images.