AI Could Help Increase the IVF Success Rate

Algorithms help doctors create better treatment plans for aspiring parents

4 min read
A photo of a display on a tablet and woman’s hand pointing to something.

Embryonic’s algorithm uses deep learning to classify images of the embryos and predict which ones will result in a successful pregnancy.

Embryonics

More than 8 million people have been born worldwide with the help of in vitro fertilization since 1978. In IVF, an egg is fertilized by sperm in the lab; the resulting clump of cells is transferred into a patient’s uterus.

Although IVF techniques have advanced significantly in recent decades, the average success rate is still fairly low: around 45 percent. The percentage steadily declines as women age; a 40-year-old woman has a likely success rate of about 12 percent, according to Pregnancy & IVF Clinics Worldwide.


Embryonics, a startup in Haifa, Israel, aims to raise the IVF success rate with its suite of AI algorithms. The company’s system uses machine learning to help doctors create personalized treatment plans.

“Technology can help fertility doctors make data-driven decisions and answer complex questions in a smarter way,” says Dr. Yael Gold-Zamir, CEO and cofounder. She launched the company in 2018 with David Silver and IEEE Fellow Alex Bronstein.

Gold-Zamir has a medical degree from the Hebrew University of Jerusalem. Silver is a machine learning engineer who previously worked for Apple and Intel. Bronstein is a computer science professor at the Technion.

“Embryonics is tackling very unique problems—the quality of human analysis and how to analyze big data so that it is clinically relevant,” Bronstein says.

IVF PRIMER

In IVF, several mature eggs are retrieved from the patient’s ovaries. The eggs are then mixed with sperm in a clinic. The developing embryos grow in the lab for several days until an embryologist chooses one or two to be implanted. (The term embryo technically refers to the developmental stage, when the amniotic sac forms inside the uterus, around two weeks after fertilization. But fertility clinics typically refer to the clusters of cells that they evaluate and implant as embryos.)

Doctors typically choose which embryos to implant based on chromosomal testing and appearance, Silver says. Each is graded based on the number and size of its cells and its rate of development.

But there are several problems with that approach, Silver points out.

“One is that the embryologists’ ability to collect data is limited,” he says. “The amount of data about embryos, past patients, and successful live births available to any single doctor is very small, so it’s hard for them to generalize [about] what indicates that a fertilized egg is viable.”

Another problem is that not all clinics have the same grading system, so two facilities might rate the same embryo differently.

“Technology can help doctors in fertility make data-driven decisions and answer complex questions in a smarter way.”

One of the startup’s algorithms uses deep learning to classify images of the embryos and predict which ones will result in a successful pregnancy. It compares the patient’s medical information, such as age and underlying health conditions, along with images of her embryos to the same data from past patients who had successful or unsuccessful implantations.

Silver and Bronstein used thousands of medical images from around the world to train the AI system. But while developing the algorithm, the engineers found that clinics don’t have the same equipment or use the same settings on microscopes and other tools. The variation affected how the platform classified the embryos.

To overcome that problem, Bronstein and Silver developed their own data-augmentation system for the images. It cancels out environmental factors such as lighting and removes irrelevant parts of the images.

“The system only extracts information that is biologically meaningful, such as cellular structures,” Silver says.

The algorithm is currently being tested in clinics in several countries including Lithuania, Malaysia, and Spain. Doctors were hesitant to use the platform at first, Gold-Zamir says, but since testing it with patients, they have given the company positive feedback. The system has increased the success rate by more than 15 percent, Silver says.

The company has submitted its embryo-classification system to the U.S. Food and Drug Administration for approval. It already has been approved in Europe.

Embryonics is developing an algorithm to help doctors prescribe the best hormone-replacement treatment for patients who require it to increase their chances of successful implantations. There are currently no definitive guidelines to help doctors decide which medication is best for patients, Silver says.

“We found that sometimes the same patient goes to several clinics and is prescribed completely different hormone treatment plans,” he says.

To improve decision-making for the treatment plan, the Embryonics team is developing an algorithm that uses machine learning to provide customized recommendations. The algorithm is learning from information about patients as well as a collection of past treatment plans and their outcomes.

“Based on similarities among patients we can do simulations,” Silver says, “and estimate what would have happened if another treatment protocol was chosen.”

“IVF is complicated,” Gold-Zamir says. “It’s not just one decision doctors have to make; it’s a process of sequential decisions. And we need to maximize the potential for the success of all of those decisions.”

FOUNDING

The startup emerged from Gold-Zamir’s belief that technology can help doctors make better decisions and therefore increase IVF success. She says most fertility specialists make decisions about a patient’s treatment options the same way experts did 40 years ago.

“Many complicated decisions are made based on the doctor’s gut feeling, which is based on all the cases they have seen in their career,” she says. The decisions include which embryos are viable, how many should be implanted, and what kind of hormone treatment is most appropriate.

Gold-Zamir was introduced to Bronstein and Silver through a colleague. Although their original goal was simply to publish a research paper, the trio wanted to improve fertility outcomes and decided to commercialize their first algorithm.

Initially, funding for the company came from friends and family, but the team later received a grant from the Israel Innovation Authority, a government agency that helps fund technology startups. Gold-Zamir says the grant enabled them to launch the company.

The founders also participated in the Google for Startups program, which provides companies with funding, mentoring, and networking.

Embryonics now has 17 employees including doctors, bioinformaticians and computer scientists.

Its next goal is to develop algorithms to help doctors choose which embryos to freeze for future IVF cycles as well as noninvasive genetic screening and analysis.

“I love being able to apply the latest and greatest technologies to something that impacts human life in one of the greatest ways possible: starting a family,” Silver says.

The Conversation (0)

Get unlimited IEEE Spectrum access

Become an IEEE member and get exclusive access to more stories and resources, including our vast article archive and full PDF downloads
Get access to unlimited IEEE Spectrum content
Network with other technology professionals
Establish a professional profile
Create a group to share and collaborate on projects
Discover IEEE events and activities
Join and participate in discussions

Tony Fadell: The Nest Thermostat Disrupted My Life

The Nest founder tells of years in pursuit of a thermostat he actually likes

7 min read
A man holds a circular device in front of a blue wall that says nest on it.

Tony Fadell shows off the Nest thermostat in 2012.

Karsten Lemm/picture-alliance/dpa/AP

The thermostat chased me for 10 years.

That is pretty extreme, by the way. If you’ve got an idea for a business or a new product, you usually don’t have to wait a decade to make sure it’s worth doing.

Keep Reading ↓ Show less
{"imageShortcodeIds":[]}