Candice Reed – Australia’s first baby conceived by in vitro fertilisation (IVF) – is now in her late 30s. Since her birth in Melbourne in 1980 more than 200,000 children have been born here as a result of IVF-type treatments.
Around 13,000 babies were born in Australia from related treatments in 2014 alone – roughly one in every 22 children.
Despite those figures, and decades of research, IVF still has limited success rates, confounded by age and a multitude of other factors.
One of the biggest challenges of the often heartbreaking process is selecting a viable embryo to begin with. At present that is usually done by eye, and based on the good judgement of the embryologist: it’s manual, subjective, and imprecise.
Could AI do better? An Adelaide company founded earlier this year believes that artificial intelligence could achieve higher success rates than human clinicians in the selection of viable embryos resulting in live births.
“IVF is an incredibly difficult process for people. We feel for people who have been through the devastating emotions of unsuccessful IVF,” co-founder of Life Whisperer Dr Michelle Perugini told Computerworld. “We’re involved here in something that actually has a positive impact on people’s lives.”
Life Whisperer was launched in February this by husband and wife team Dr Michelle Perugini and Dr Don Perugini – who sold predictive analytics firm ISD Analytics to Ernst & Young in 2015 – and co-founder Dr Jonathan Hall.
In July the start-up – based at the University of Adelaide’s ThincLab incubator – announced a commercial agreement with the ASX-listed fertility clinic Monash IVF Group and its subsidiary Repromed, and gained access to thousands of embryo images in order to train and validate its image analysis techniques.
“That was a huge bonus because they obviously are the custodian of the data that we need to do this analysis and build up these models, so that really expedited the process for us,” Michelle Perugini explains.
There are a number of methods to select embryos for the IVF process. The least invasive and most common is by identifying certain tell-tale features of an image of the embryo to assess its health and viability.
“There are some things we know from past research that have been shown to be important features. And we've codified algorithms that represent those features. The other part is, via a deep learning system that basically trawls through thousands of past images of embryos and knows what the outcomes of those embryos are, actually identifying new features and classifying them to come up with a selection model that you can apply,” Perugini explains.
Sign up for Computerworld eNewsletters.