Automatic prediction of embryo cell stages using artificial intelligence convolutional neural network

Abstract

OBJECTIVE: The objective of this study is to identify the times of embryo cell division (up to 8-cell) by applying an artificial intelligence-based (AI) approach using time-lapse microscopy (TLM) images. Our hypothesis is that the unbiased AI approach would identify embryo division times without relying on human intervention.DESIGN: We built a stand-alone framework with a convolutional neural network (CNN) as the core to predict cell division times for mouse and human embryos, respectively, based on raw time-lapse digital images.MATERIALS AND METHODS: This study used 31,120 images of 100 mouse embryos from a public dataset. We also used 661,060 images of 11,898 human embryos cultured in the TLM system (EmbryoScope, Vitrolife, Sweden) at various cell stages post-insemination/ICSI. Images were separated randomly into training, validation, and test sets with a ratio of 70:15:15 for mouse and 80:10:10 for human. Separation was done on an embryo basis, Data are presented as n (%), risk ratio (RR) and 95% confidence interval (CI) *Adjusted for age at retrieval, number of oocytes retrieved during fresh cycle, BMI, embryo quality, and PGT-A e360

Fast Multi-Focus Fusion Based on Deep Learning for Early-Stage Embryo Image Enhancement

“…Embryo assessment based on convolutional neural networks (CNNs) allows to achieving high accuracy results >0.98 classifying them into two classes: good- and poor-quality embryos [ 4 , 5 ]. However, even CNN based approaches encounter difficulties for a multi-class prediction problem, since the embryo quality assessment in a >2-cell stage is still challenging [ 6 , 7 , 8 ]. Reasons for this may be due to the noise in the image, highly overlapped cells or poor quality of the image.…”

Towards the automation of early-stage human embryo development detection

“…Notably, the application of artificial intelligence focusing on multi-class prediction remains scarce. The recent study proposed a standalone framework based on Inception-V3 CNNs as the core to classify individual TL images up to the 4-cell stage for mouse and human embryos, respectively [18]. In their work, 31,120 images of 100 mouse embryos and 661,060 images of 11,898 human embryos cultured in the TL monitoring system were analysed.…”

Machine Learning Reveals The Effect of Maternal Age on The Mouse Pre-Implantation Embryo Developmental Timing

“…Yet, our understanding of the interplay between embryos' morphology, viability, and maternal age is limited, as manual approaches to infer embryo morphokinetics are timeconsuming, subjective, and prone to errors. Machine learning [15][16][17][18] was recently harnessed to predict embryo developmental potential 19,20 , however, with limited success.Here, we develop an artificial intelligence (AI) platform that infers the embryos' developmental stage and captures tens of morphological properties and developmental dynamics. We show that developmental timing is the most informative and predictive morphokinetic property, particularly for embryos from maternally aged females.…”

“…Yet, our understanding of the interplay between embryos' morphology, viability, and maternal age is limited, as manual approaches to infer embryo morphokinetics are timeconsuming, subjective, and prone to errors. Machine learning [15][16][17][18] was recently harnessed to predict embryo developmental potential 19,20 , however, with limited success.…”