A cutting-edge artificial intelligence system, dubbed “BELA,” has been developed to accurately evaluate the chromosomal status of in vitro-fertilized (IVF) embryos using only time-lapse video images and maternal age. This breakthrough comes from researchers at Weill Cornell Medicine and was detailed in a study published on September 5 in Nature Communications. BELA represents the latest innovation in AI-based platforms designed to determine whether an embryo has a normal (euploid) or abnormal (aneuploid) chromosome count, a critical factor in the success of IVF treatments.
Unlike previous AI models that relied on embryologists’ subjective evaluations, BELA provides an objective, generalized assessment. If validated in clinical trials, this system could be widely adopted in embryology clinics, enhancing the efficiency of the IVF process. “This fully automated and objective approach offers increased predictive power due to the extensive image data it utilizes,” explained Iman Hajirasouliha, the study’s senior author and associate professor at Weill Cornell Medicine’s Englander Institute for Precision Medicine.
The study’s lead author, Suraj Rajendran, is a doctoral student in Hajirasouliha’s lab. The embryology aspect of the research was led by Nikica Zaninovic, associate professor of embryology in clinical obstetrics and gynecology and director of the Embryology Laboratory at the Ronald O. Perelman and Claudia Cohen Center for Reproductive Medicine (CRM) at Weill Cornell Medicine. Co-author Dr. Zev Rosenwaks serves as the director and physician-in-chief of the CRM and is a distinguished professor in Reproductive Medicine at Weill Cornell Medicine.
Traditionally, embryologists assess the quality of IVF embryos through microscopic examination. If an embryo appears relatively normal but maternal age raises concerns, they may opt for more direct chromosomal testing. The “gold standard” for this assessment is preimplantation genetic testing for aneuploidy (PGT-A), which involves a biopsy-like procedure with inherent risks. In recent years, collaborations between embryologists and AI experts have aimed to automate aspects of this workflow to enhance outcomes. A previous AI system, STORK-A, created by Hajirasouliha and colleagues, predicted embryo ploidy status with about 70% accuracy using a single microscopic image, maternal age, and embryologists’ evaluations.
BELA was designed to generate accurate ploidy predictions independently of embryologist input. Its core is a machine-learning model that evaluates nine time-lapse video images of an embryo taken five days post-fertilization, generating an embryo quality score that, along with maternal age, predicts whether the embryo is euploid or aneuploid.
The researchers trained the BELA model using a de-identified dataset from Weill Cornell Medicine’s CRM, encompassing nearly 2,000 embryos and their PGT-A-tested ploidy statuses. They then validated the model using new datasets from Weill Cornell and large IVF clinics in Florida and Spain. The results indicated that BELA predicted ploidy status with improved accuracy compared to earlier models and performed well across both internal and external datasets.
Looking ahead, the research team plans to prospectively test BELA’s predictive capabilities in a randomized, controlled clinical trial. Zaninovic noted that “BELA and similar AI models could broaden access to IVF in areas lacking advanced technology and PGT testing, enhancing equity in IVF care globally.”
Additionally, BELA’s ability to process extensive image data for each embryo suggests it may have applications beyond ploidy prediction. Rajendran remarked, “We hope this model can also assist with general embryo quality estimation, development stage prediction, and other customizable functions suited to individual embryology clinics.”
Related topics:
