Genetic abnormalities in embryos are the leading cause of miscarriage and non-implantation among women choosing to undergo in vitro fertilization (IVF). Current methods for assessing embryo quality include noninvasive microscopic examination by an embryologist as well as preimplantation genetic testing for aneuploidy (PGT-A), an invasive biopsy-like procedure that is not without risk. In a new study published in Nature Communications, physicians and researchers from NewYork-Presbyterian and Weill Cornell Medicine report on the value of an innovative noninvasive artificial intelligence (AI)-based technique known as BELA (Blastocyst Evaluation Learning Algorithm) that is highly accurate for assessing the chromosomal status of IVF embryos.
Unlike prior AI-based approaches, BELA does not use embryologists' subjective assessments of embryos. If its utility is confirmed in clinical trials, BELA could be used widely in embryology clinics to improve the efficiency of the IVF process. Below, Nikica Zaninovic, PhD, director of the embryology laboratory at the Ronald O. Perelman and Claudia Cohen Center for Reproductive Medicine (CRM) at Weill Cornell Medicine and one of the study's senior authors, explains the findings and their implications for IVF care.
Research Background
Visual examination of an embryo by an embryologist is not an exact science. One embryo can look "prettier" than another, but we cannot accurately tell if it is normal or abnormal. If it looks relatively normal but there are reasons to suspect possible problems, such as in cases of advanced maternal age, we may test its chromosomal status more directly using PGT-A.
During PGT-A, we remove a few cells from the embryo and analyze them to see if there is euploidy (the correct number of chromosomes) or aneuploidy (an abnormal number of chromosomes). This approach is more accurate than visual assessment, but it is very invasive and can potentially damage the embryo. We also haven't had effective means to determine which embryos to biopsy.
In addition to being invasive, PGT-A is very expensive. It is also not permitted in some U.S. states and in other countries. My colleagues and I — including senior author Iman Hajirasouliha, PhD, a member of the Englander Institute for Precision Medicine at Weill Cornell Medicine, Zev Rosenwaks, MD, director and physician-in-chief of the CRM and a reproductive endocrinologist at NewYork-Presbyterian and Weill Cornell Medicine, and lead author Suraj Rajendran, a doctoral student in Dr. Hajirasouliha’s laboratory — sought to develop a method to accurately assess embryos in a noninvasive manner without human intervention.
Research Methods
In 2022, our team developed an AI-based system called STORK-A, which uses a single microscopic image of an embryo plus maternal age and embryologists’ scoring, to predict an embryo's ploidy status with about 70% accuracy. In the current study, we report on the development of BELA, which differs from STORK-A because it is entirely automated. There is no human intervention.
To create BELA, we supplied the computer with raw images, drawing from the CRM's vast repository of images of normal and abnormal embryos and data on their PGT-A ploidy status, and asked the computer to learn by itself. BELA's machine-learning model analyzes time-lapse video images of an embryo under a microscope in a key interval four to six days after fertilization to generate an embryo quality score. The system then uses this score and maternal age to predict euploidy or aneuploidy.
BELA and AI models like it could expand the availability of IVF to areas that do not have access to high-end IVF technology and preimplantation genetic testing, improving equity in IVF care across the world.
— Dr. Nikica Zaninovic
Key Findings
BELA was able to predict embryonic ploidy status with 70% to 80% accuracy. By achieving an area under the receiver operating characteristic curve of 0.76 for discriminating between euploidy and aneuploidy embryos on the Weill Cornell Medicine dataset, BELA matched the performance of models trained on embryologists' manual scores. The next step is to test BELA’s predictive power prospectively in a randomized, controlled clinical trial, and which is required by the FDA to achieve approval of this technique for clinical practice.
Future Implications
While not a replacement for PGT-A, BELA demonstrates how AI-based ploidy prediction models can streamline the embryo evaluation process. This method could be very helpful in states and countries where embryonic biopsies are not permitted. In cases where we can do PGT-A, BELA may help physicians and patients better identify which ambiguous embryos — those that are not obviously normal or abnormal — to biopsy.
This study shows the potential of AI for enhancing IVF. I believe that AI can be used in every step of the IVF process — from patient evaluation through treatment in the laboratory to predict the chance that a patient will have a healthy baby. It can also improve our workflow to make it more efficient. AI is not making the decision of what embryo to choose or telling patients and physicians what to do, but rather is a tool that can help them make better decisions.
BELA is the product of merging high-level clinical IVF care and high-level computer science expertise from the largest academic IVF program in the country to create a tool that may benefit patients everywhere.
