How artificial intelligence could slash miscarriage risk

The breakthrough works with IVF treatment, using a super-computer to pick out "good" embryos – or "bad" ones.

The breakthrough works with IVF treatment, using a super-computer to pick out "good" embryos – or "bad" ones.

Published Oct 10, 2018

Share

London - Women could be spared the heartbreak of miscarriage thanks to artificial intelligence that can help spot which pregnancies will fail.

The breakthrough works with IVF treatment, using a super-computer to pick out "good" embryos – or "bad" ones which could prevent a woman from giving birth or cause her to miscarry.

Researchers found the computer, nicknamed The Beast, was able to predict which embryos would lead to a live birth with 85 percent accuracy, correctly selecting 280 out of 328.

Doctors who email a picture of an embryo to the computer can expect a result within minutes.

And the technology is only five years away from being made available to couples, say the researchers. About half of miscarriages are caused by a problem with the embryo, where it has an abnormal number of chromosomes.

The results of the study, which involved Cornell University in New York and Imperial College London, will be presented at the annual meeting of the American Society for Reproductive Medicine in Denver. 

Dr Nikica Zaninovic, lead researcher in the US who was supported by Dr Zev Rosenwaks, said: "If AI can recognise embryos that are normal and those that are abnormal chromosomally that will result in a reduced miscarriage and stillbirth rate, which is really our central reason for doing this research."

At the moment, fertility clinics tend to judge if embryos are developing normally by looking at their shape, size or growing speed.

Researchers trained the super-computer using almost 700 images of five-day-old embryos. They used time-lapse photography, where a photograph of embryos in an incubator is taken every ten minutes – avoiding the risks that come with an embryo being handled.

Daily Mail

Related Topics: