Researchers from Weill Cornell Medicine have developed an AI-powered model that could help couples undergoing in vitro fertilization (IVF) and guide embryologists in selecting healthy embryos for implantation. Recently published in Nature Communications, the study presents the Blastocyst Evaluation Learning Algorithm (BELA). This state-of-the-art deep learning model evaluates embryo quality and chromosomal health using time-lapse imaging data and maternal age.
By offering a non-invasive and cost-effective supplement to standard genetic testing known as preimplantation genetic testing for aneuploidy (PGT-A), BELA could streamline embryo selection and reduce costs for families.
“This is a fully automated and more objective method compared to prior approaches, and the larger amount of time-lapse imaging data it uses can generate greater predictive power” said study senior author Dr. Iman Hajirasouliha, an associate professor at Weill Cornell Medicine.
Since its introduction in 1978, IVF has been responsible for over 8 M successful births, providing a solution for individuals and couples facing infertility worldwide. Embryo selection is a crucial step in the process, significantly influencing pregnancy success rates. Traditional methods, such as PGT-A, require cell extraction for chromosomal analysis, which can be expensive and risky to embryo viability.
The researchers developed BELA, which automates the embryo evaluation process. By analyzing time-lapse imaging data collected over five days of development, combined with maternal age, the AI-powered model predicts the chromosomal health of embryos and ranks them by quality. Timing and speed are crucial indicators of embryo viability and are core to the model’s analysis.
BELA was trained on Cornell’s high-performance BioHPC computing cluster using NVIDIA A40 GPUs and a diverse dataset of over 2,800 embryo time-lapse sequences, capturing stages of developing cells. The infrastructure enabled efficient data processing, averaging just 5.23 minutes for training and taking about 30 seconds per embryo prediction.
Figure 1. The STORK-V is a clinical tool that uses automation to assist embryologists in providing a comprehensive assessment of the embryos (credit: Rajendran, S., Brendel, M., Barnes, J. et al.)
To make the model usable in clinical settings, the team also developed STORK-V, a web-based platform powered by BELA. Embroloyigsts upload time-lapse imaging data and get real-time embryo quality and chromosomal health predictions.
BELA outperforms current AI-based models, achieving an AUC (a measure of model accuracy) of 0.82 when distinguishing normal from abnormal embryos. It also delivers reliable, automated predictions that match or exceed the accuracy of traditional methods involving manual evaluation by embryologists.
While the researchers don’t aim to replace PGT-A, BELA can help streamline IVF workflows. It prescreens embryos, assisting embryologists to decide which ones to analyze further. This reduces costs and ensures that efficient and reliable embryos are chosen.
The source code is available open source on GitHub.
Read the research Automatic ploidy prediction and quality assessment of human blastocysts using time-lapse imaging.
Photo courtesy of Freepik
