A fully automated AI-based system to assess IVF embryo quality

The new system, called “BELA”, and described in an article published on September 5 in Nature Communicationsis the team’s latest AI-based platform for assessing whether an embryo has a normal (euploid) or abnormal (aneuploid) number of chromosomes – a determining factor in the success of IVF. Unlike previous AI-based approaches, BELA does not need to consider subjective assessments of embryos made by embryologists. It thus offers an objective and generalizable measure and, if its usefulness is confirmed in clinical trials, it could one day be widely used in embryology clinics to improve the efficiency of the in vitro fertilization process.

“This is a fully automated and more objective approach compared to previous approaches, and the greater amount of imaging data it uses may yield greater predictive power,” said study senior author Dr. Iman Hajirasouliha, associate professor of physiology and biophysicist and member of the Englander Institute for Precision Medicine at Weill Cornell Medicine.

The study’s first author was Suraj Rajendran, a doctoral student in Dr. Hajirasouliha’s lab. The study’s embryology work was led by Dr. NewYork-Presbyterian Medical Center/Weill Cornell. Zev Rosenwaks, director and physician-in-chief of CRM and Revlon Distinguished Professor of Reproductive Medicine in Obstetrics and Gynecology at Weill Cornell Medicine, co-authored the study.

Embryologists typically assess the quality of an IVF embryo by examining it under a microscope. If it appears relatively normal but there is reason to suspect possible problems, such as in cases of advanced maternal age, they may test your chromosomal status more directly. The “gold standard” test is a somewhat risky, biopsy-like procedure called preimplantation genetic testing for aneuploidy (PGT-A). In recent years, embryologists have teamed up with IT/AI experts to find ways to automate some of this workflow and improve results. Hajirasouliha and colleagues developed an AI-based system called STORK-A, which uses a single microscopic image of an embryo, plus maternal age and embryologist scores, to predict the embryo’s ploidy status with about 70% accuracy. .

Researchers developed BELA to generate accurate ploidy predictions, independent of embryologists’ assessments. The heart of the system is a machine learning model that analyzes nine time-lapse video images of an embryo under a microscope at a key interval about five days after fertilization to generate an embryo quality score. The system then uses this score and maternal age to predict euploidy or aneuploidy.

The researchers trained the model on a de-identified Weill Cornell Medicine CRM dataset with image sequences of nearly 2,000 embryos and their ploidy status tested by PGT-A. They then tested the model on new CRM datasets from Weill Cornell Medicine and separate large IVF clinics in Florida and Spain. They found that the model predicted ploidy status with moderately greater accuracy than previous versions and performed well for both the external and internal data sets.

The next step, the researchers say, is to prospectively test the predictive power of BELA in a randomized, controlled clinical trial, which they are currently planning.

“BELA and AI models like this could expand the availability of IVF to areas that lack access to cutting-edge IVF technology and PGT testing, improving equity in IVF care around the world,” said the Dr.

The fact that BELA is configured to process a large amount of image data for each embryo also suggests to researchers that it could be used for more than ploidy prediction.

“Our hope is that this model can also be useful for general estimation of embryo quality, prediction of embryo developmental stage, and other functions that an embryology clinic can adapt to its own needs,” said Rajendran.

Many physicians and scientists at Weill Cornell Medicine maintain relationships and collaborate with external organizations to advance scientific innovation and provide expert guidance. The institution makes these disclosures public to ensure transparency. For this information, see the Dr. Iman Hajirasouliha and Dr. Nikica Zaninovic.

The research reported in this story was supported in part by the National Institute of General Medical Sciences, part of the National Institutes of Health, through grant number R35GM138152.