AI in IVF: Startup Claims Improved Embryo Selection & Success Rates

Artificial intelligence is increasingly delivering more consistent diagnostic precision in the field of medicine each year. This trend is evident in areas like skin cancer and lung cancer detection.

Currently, an Israeli startup named Embryonics asserts that its AI technology has the potential to enhance the likelihood of a successful embryo implantation during in vitro fertilization (IVF). The company has developed an algorithm designed to forecast embryo implantation probability, utilizing time-lapse imaging of embryos as they develop during IVF.

It is still in its early stages. To date, in a trial involving eleven women between the ages of 20 and 40, six have achieved successful pregnancies, with the remaining five awaiting results, according to Embryonics.

Nevertheless, Embryonics is noteworthy for its potential to significantly impact a substantial market that has remained largely unchanged for decades, growing primarily due to demographic shifts, such as women postponing childbearing due to financial considerations.

The global IVF market is projected to expand from approximately $18.3 billion in 2019 to nearly twice that amount within the next five years, according to some projections. However, the tens of thousands of women who undergo IVF annually often face expenses ranging from $10,000 to $15,000 per cycle (at least in the U.S.), coupled with diminishing chances of success as they age.

The core motivation behind Embryonics is the possibility of reducing the number of IVF cycles and the associated costs. The company was established three years ago by Yael Gold-Zamir, M.D., who initially trained in general surgery at Hebrew University before transitioning to research in an IVF laboratory, driven by a strong interest in the science of fertility.

She subsequently connected with two individuals possessing complementary skills and experience. One was David Silver, a bioinformatics specialist who studied at the Technion-Israel Institute of Technology and spent three years as a machine learning engineer at Apple, preceded by three years as an algorithm engineer at Intel.

The other individual was Alex Bronstein, an experienced entrepreneur who previously served as a principal engineer at Intel and currently leads the Center for Intelligent Systems at Technion, while also contributing to various deep learning AI initiatives, including those at Embryonics and Sibylla, a company focused on algorithmic trading in financial markets.

Embryonics remains a relatively small organization, but the team of three, along with their thirteen full-time employees, appears to be making substantial headway.

With the support of $4 million in seed funding from the Shustermann Family Investment Office and the Israeli Innovation Authority, Embryonics anticipates receiving regulatory approval in Europe, enabling them to market their software – which the team claims can identify patterns in small cell clusters more effectively than a human – to fertility clinics throughout the continent.

Leveraging a database containing millions of anonymized patient records from various centers worldwide, Gold-Zamir states that the company is also planning future developments. Specifically, beyond introducing its embryo analysis software to the U.S. market, Embryonics aims to collaborate with fertility clinics to refine hormonal stimulation protocols.

As Bronstein explains, every woman undergoing IVF or fertility preservation receives hormone injections for 8 to 14 days to stimulate the ovaries to produce a maximum number of mature eggs. Currently, only three standard protocols are used, with a significant degree of trial and error involved in determining the optimal approach. Embryonics believes that deep learning can help identify the ideal hormone mix and timing for each individual.

Further advancements are also in development. “Embryonics’s objective is to deliver a comprehensive solution, encompassing all facets of the process,” says Gold-Zamir, who balances her role as CEO with raising four children.

It is premature to determine whether this emerging company will succeed. However, it is clearly at the leading edge of a rapidly evolving technology, contrasting with the traditional methods used by many IVF clinics globally for over 40 years, which involve visually assessing embryo health by examining days-old embryos under a microscope to evaluate cell division and morphology.

In the spring of 2019, researchers at Weill Cornell Medicine in New York City published their findings demonstrating that AI can assess embryo morphology with greater accuracy than the human eye. They trained an algorithm using 12,000 images of human embryos taken precisely 110 hours after fertilization to differentiate between embryos of good and poor quality.

The researchers initially assigned a grade to each embryo by embryologists, considering various aspects of its appearance. They then conducted a statistical analysis to correlate the embryo grade with the probability of a successful pregnancy. Embryos were classified as good quality if the probability exceeded 58% and poor quality if it fell below 35%.

Following training and validation, the algorithm was able to classify the quality of new images with 97% accuracy.