An Algorithm May Soon Help People Make Babies

• Bookmarks: 67

Image: Science Photo Library — KTSDESIGN/Brand X Pictures/Getty Images
Companies are developing embryo selection algorithms aimed at boosting the success of IVF

After two years of grueling attempts at in vitro fertilization (IVF) — in which eggs and sperm are combined outside the body, and the resulting embryos are transferred to the womb — Australian couple Lorraine and James Coriakula finally got pregnant and gave birth to a son, Ebenezer, in October. A machine-learning algorithm might have helped.

In early 2020, the Coriakulas visited a Melbourne clinic that’s using new artificial intelligence software to select embryos for IVF. While the embryo selection process is usually done by humans, the software developed by startup Life Whisperer of Adelaide, Australia, scans images of embryos and judges which look healthy. The company claims Ebenezer is the first successful birth from an embryo selected by its A.I. software.

As the average age of pregnancy increases, more people are turning to assisted reproductive technology to help conceive. Birth rates from IVF are getting better, but it can still take multiple tries to get pregnant. According to the U.S. Centers for Disease Control and Prevention, 42% of patients age 35 to 37 who use their own eggs have a baby after their first round of IVF, but that rate declines significantly with age. For hopeful parents, the journey is often frustrating, not to mention expensive. An IVF cycle can cost anywhere from $12,000 to $17,000.

Some fertility specialists think artificial intelligence could help patients get pregnant faster by scanning images of their embryos and picking out the ones that have the best chance of resulting in a pregnancy. Companies like Life Whisperer say that could reduce the number of IVF rounds and ultimately save patients money.

“The selection of the embryo is really critical, and it directly relates to whether a pregnancy is going to result or not,” Michelle Perugini, PhD, CEO and co-founder of Life Whisperer, tells Future Human.

The company’s algorithm is trained on approximately 20,000 images of embryos previously used in IVF, some of which led to pregnancies and others that didn’t. Embryologists use the Life Whisperer program to upload a single image of a five-day-old embryo, and within minutes, the system provides a score from zero to 10, rating the quality of the embryo and likelihood of implantation — the first stage of pregnancy during which the embryo attaches to the wall of the uterus. A score of zero means the embryo has the lowest chance of achieving a pregnancy, whereas a 10 represents the best chance.

Life Whisperer started rolling out its software in 2020 at fertility clinics in Australia, New Zealand, Malaysia, India, Croatia, and the Czech Republic, and Perugini says the company is working with the Food and Drug Administration to get regulatory approval in the United States.

Perugini and others developing this technology see A.I. as a way to automate a process that relies heavily on human judgment. “In IVF cycles, there are often multiple embryos to choose from, and figuring out which one to transfer, which ones to save, and which ones to discard is always a decision that we have to make,” says Eric Widra, MD, chief medical officer of Shady Grove Fertility, a fertility clinic with locations across the eastern United States.

To make those decisions, fertility clinics use a grading system based on an embryo’s appearance. Embryologists inspect embryos under a microscope and assign them a grade based on size, shape, and number of cells. It’s not an exact science—two embryologists could score the same embryo differently.

Many IVF labs have tried to improve upon embryo selection with time-lapse imaging, a technique that involves taking periodic pictures of fertilized eggs before they become embryos. This allows embryologists to monitor the embryos’ health and development, rather than rely solely on the end assessment. But a 2017 review in the journal PLOS One found no significant difference in pregnancy rates between those who opted for time-lapse imaging and those who didn’t.

Another way to evaluate the health of embryos is preimplantation genetic testing, which involves extracting a few cells from an embryo and examining them for abnormalities before transferring the embryo to the uterus. Embryologists look for the correct number of chromosomes — 23 from each biological parent. Too few or too many chromosomes, known as aneuploidy, is a common factor in miscarriage. But genetic testing of embryos isn’t cheap. At around $1,500 to $3,500, on top of regular IVF costs, most patients don’t opt for it. It’s also a labor-intensive test for laboratories.

“The prep of the cells has to be pristine, otherwise there could be errors in the results,” Widra says. “It works very well, but if you could instead just put the embryo under a microscope and get the same information, that would be great.”

In a retrospective analysis using 9,000 images from 11 different IVF clinics, Life Whisperer found that its algorithm was nearly 25% better at choosing embryos that resulted in pregnancy over this standard human assessment. The results were published last year in the journal Human Reproduction.

“Where machine learning helps is in making this subjective process more objective,” says Manoj Kumar Kanakasabapathy, a research scientist at Brigham and Women’s Hospital in Boston who is part of a team developing A.I. tools for embryo assessment.

In a recent study in the journal eLife, Kanakasabapathy and his colleagues used close to 2,500 images taken of embryos at the same stage to train a deep-learning algorithm. Then they used the algorithm to assess 742 images of embryos and found that it was 90% accurate in choosing the ones likely to implant in the uterus. Among those, the system was 75% accurate at picking embryos with the normal number of chromosomes; trained embryologists were about 65% accurate. Neither study, however, looked at the probability of live births.

Though the research is promising, Eliza Curnow, PhD, senior scientist in the Department of Embryology at the University of Washington, says the new software can’t definitively detect genetic abnormalities, because 2D images don’t tell the full story about the overall health of an embryo. Some embryos that appear normal may have an abnormal number of chromosomes and lead to miscarriage. As women age, they produce fewer embryos with the normal number of chromosomes, so the software might not be best for older IVF patients, Curnow says.

On the other hand, some embryos that get lower scores based on their appearance have a normal number of chromosomes and end up making perfectly healthy babies. “Many times, the only embryos we have are of what we would consider relatively poor quality based on their appearance,” Widra says. “But we still get babies from some of those—it’s just at a much lower percentage.”

Other companies are developing similar technology. Sydney-based has introduced an A.I.-based embryo selection system called Ivy in Australia. AiVF of Tel Aviv is commercializing a similar product. Researchers at Weill Cornell Medicine, meanwhile, have developed a tool called Stork.

As more fertility clinics turn to IVF, Joyce Harper, PhD, professor of reproductive science at University College London, tells Future Human that it’s important to be up-front with patients about the limitations of the technology.

The fertility industry is notorious for offering supplementary services purporting to boost the chances of a successful pregnancy despite weak evidence that they actually work. Harper worries that A.I.-based embryo selection could become yet another “add-on” that clinics offer to people who are desperate to get pregnant.

“These sorts of technologies are not improving the embryo in any way. They are only deciding the order embryos get transferred,” Harper says. “All it could do is decrease time to pregnancy.”

Despite the flurry of interest in A.I.-based selection, there’s no published data yet on the most important outcome: whether these systems lead to more babies. We’ll need large clinical trials in which patients are randomly assigned to have their embryos analyzed the conventional way or by A.I. to determine whether the technology actually leads to faster pregnancies and higher birth rates. A trial of’s Ivy tool is currently underway in Australia.

“The software should be able to detect with certainty which embryo will lead to a high chance of pregnancy at a better rate than chance and at a rate significantly higher than that of an experienced embryologist,” Curnow says.

Until then, would-be parents should think of A.I. not as a crystal ball, but simply one more tool to help embryologists make a judgment call.


67 recommended
comments icon0 comments
0 notes
bookmark icon

Write a comment...

Your email address will not be published. Required fields are marked *