Research groups at Keio University, Kinki University, Tokyo University, and Sanyo Onoda City Yamaguchi Tokyo Science University have succeeded in developing an algorithm that accurately identifies cell nuclei from three-dimensional fluorescence microscope images of fertilized mouse eggs using deep learning. bottom.
In vitro fertilization (IVF), which is one of the treatments for infertility, is not effective, and the pregnancy success rate by assisted reproductive technology in Japan is only 12.6%.The evaluation of human embryos in conventional IVF treatment was manual based on the morphological analysis of skilled embryo culture specialists, and there was variability between the evaluations.On the other hand, with the improvement of microscope technology and imaging technology, it has become possible to acquire time-series 3D fluorescence microscope images of the development process, but the image processing accuracy is low and it is not possible to obtain accurate quantitative indicators in the development process. It is still difficult.
In the research, we proposed a segmentation algorithm (QCANet) using the "convolutional neural network" of the deep learning algorithm, and accurately identified the cell nucleus even in a state where more than 50 cells are densely packed like a mouse embryo, and quantified it during mouse development. We aimed to obtain a target index.As a result, we succeeded in automatically evaluating the cell nuclei with the world's highest accuracy in the three species of mice, nematodes, and Drosophila.It was found that even fertilized eggs that seem to occur in the same way have "individuality" in the state of each division.
The developed algorithm has succeeded in nuclear identification for a wide range of developmental stages (hundreds to thousands of cells), and may be a very useful tool that forms the basis of developmental biology.In addition, this time, fluorescent substances were introduced into each embryo for model organisms to stain cell nuclei, but in preparation for application to human embryos, the company plans to develop an algorithm for cell identification from unstained images in the future. ..
