Clinical Review

2020 Update on fertility

Author and Disclosure Information



Artificial intelligence and embryo selection

Tran D, Cooke S, Illingworth PJ, et al. Deep learning as a predictive tool for fetal heart pregnancy following time-lapse incubation and blastocyst transfer. Hum Reprod. 2019;34:1011-1018.

With continued improvements in embryo culture conditions and cryopreservation technology, there has been a tremendous amount of interest in developing better methods for embryo selection. These efforts are aimed at encouraging elective single embryo transfer (eSET) for women of all ages, thereby lowering the risk of multiple pregnancy and its associated adverse neonatal and obstetric outcomes—without compromising the pregnancy rates per transfer or lengthening the time to pregnancy.

One of the most extensively studied methods for this purpose is preimplantation genetic testing for aneuploidy (PGT-A, formerly known as PGS), but emerging data from large multicenter randomized clinical trials (RCTs) have again cast significant doubt on PGT-A's efficacy and utility.5 Meanwhile, alternative methods for embryo selection are currently under investigation, including noninvasive PGT-A and morphokinetic assessment of embryo development via analysis of images obtained by time-lapse imaging.

The potential of time-lapse imaging

Despite the initial promising results from time-lapse imaging, subsequent RCTs have not shown a significant clinical benefit.6 However, these early methods of morphokinetic assessment are mainly dependent on the embryologists' subjective assessment of individual static frames and "annotation" of observed spatial and temporal features of embryo development. In addition to being a very time-consuming task, this process is subject to significant interobserver and intraobserver variability.

Considering these limitations, even machine-based algorithms that incorporate these annotations along with such other clinical variables as parental age and prior obstetric history, have a low predictive power for the outcome of embryo transfer, with an area under the curve (AUC) of the ROC curve of 0.65 to 0.74. (An AUC of 0.5 represents completely random prediction and an AUC of 1.0 suggests perfect prediction.)7

A recent study by Tran and colleagues has employed a deep learning (neural network) model to analyze the entire raw time-lapse videos in an automated manner without prior annotation by embryologists. After analysis of 10,638 embryos from 8 different IVF clinics in 4 different countries, they have reported an AUC of 0.93 (95% confidence interval, 0.92-0.94) for prediction of fetal heart rate activity detected at 7 weeks of gestation or beyond. Although these data are very preliminary and have not yet been validated prospectively in larger datasets for live birth, it may herald the beginning of a new era for the automation and standardization of embryo assessment with artificial intelligence—similar to the rapidly increasing role of facial recognition technology for various applications.

Improved standardization of noninvasive embryo selection with growing use of artificial intelligence is a promising new tool to improve the safety and efficacy of ART.

Continue to: Environmental toxicants: The hidden danger...


Next Article: