A joint research group from Keio University, Kinki University, and the University of Tokyo has successfully developed an AI that predicts the birth of mouse fertilized eggs with an accuracy of 83.87%.
In vitro fertilization (IVF), one of the fertility treatments, relies on the visual judgment of embryologists to evaluate the quality of fertilized eggs. Accurate evaluation of the fertilized eggs that lead to each other has become difficult.In fact, the efficacy of IVF is low, and the pregnancy success rate with assisted reproductive technology in Japan is only 12.6%.
On the other hand, by analyzing live-cell imaging images that continuously capture the state of cell division in mouse fertilized eggs, our group is working to acquire indicators that lead to birth prediction, such as chromosomal segregation abnormalities, synchrony of cleavage, and developmental speed. It's here.In particular, the Quantitative Criteria Acquisition Network (QCANet), developed in 2020, is a unique image processing technology that uses a convolutional neural network, a deep learning algorithm, to efficiently extract only the cell nucleus from images of developing embryos. , has contributed to the acquisition of numerous quantitative indices in mouse developmental processes.
This time, using this QCANet, we developed a new AI algorithm Normalized Multi-View that predicts births by machine learning multivariate time-series data such as morphological features extracted from mouse embryos that lead to pregnancy and embryos that have aborted. Constructed an Attention Network (NVAN). NVAN achieves 83.87% birth prediction accuracy for mouse fertilized eggs, surpassing existing machine learning methods (74.19%) and visual inspection by embryologists (64.87%).In addition, by retrospectively clarifying the morphological characteristics of embryos that contributed to NVAN's birth prediction, we found that the shape of the nucleus and the timing of cell division at the morula stage are important for the birth of mouse embryos. It says.
This method is expected to be applied to human fertilized eggs in the future as a new basic technology for embryo evaluation of in vitro fertilization, and it is hoped that it will contribute to the improvement of pregnancy rates with assisted reproductive technology.
Paper information:【Artificial Intelligence in Medicine】An explainable deep learning-based algorithm with an attention mechanism for predicting the live birth potential of mouse embryos