Artificial intelligence holds the key to improving embryo selection for in vitro fertilization.
Jonathan James | | Quick Read
In vitro fertilization (IVF) technology has made enormous strides in recent decades, resulting in millions of successful pregnancies. Nevertheless, disparity in visual morphology assessment results between embryologists has raised serious questions about the efficacy of embryo selection (1). Attempting to solve this problem, Iman Hajirasouliha (Assistant Professor of Computational Biology) and his team at Cornell University the Englander Institute for Precision Medicine at Weill Cornell Medicine have developed a new tool – the aptly-named STORK – which is capable of driving robust assessment and selection of human blastocysts (2).
“Often, to overcome uncertainties in embryo quality, many more embryo’s than required are implanted into the patient,” says Hajirasouliha. “Often, this leads to undesired multiple pregnancies and serious complications.” Convinced there must be a more efficient way to screen for healthy, viable embryos, the team trained a deep neural network to select the best embryo’s using time-stamped images of embryos from a large IVF center. “Our network – STORK – is capable of predicting blastocyst quality with over 98 percent accuracy,” says Hajirasouliha. “It also generalizes well to images from other centers. And it outperforms individual embryologists in similar assessments.”
STORK certainly builds on the groundwork of earlier studies - for example, (3), but what makes it unique, according to Hajirasouliha, is its mathematical complexity. “From an algorithmic point of view, we really had to study and evaluate a wide range of different deep neural network architectures before we identified a suitable methodology.”
Despite its complexity and success to date, STORK has the potential to deliver more (pun intended). “We need to train our AI method on the whole series of time-lapse data and the development of embryos from day 0 to day 5, instead of single images,” says Hajirasouliha. “We also hope to integrate several different clinical factors into our model – a process that we’ll perform in tandem with further work to improve accuracy when dealing with data from different clinics.”
Subscribe to The Translational Scientist Newsletters
- CL Curchoe & CL Bormann., “Artificial Intelligence and machine learning for human reproduction and embryology presented at ASRM and ESHRE 2018,” J Assist Reprod Genet, 36, 591-600 (2019). PMID: 20690654.
- P Khosravi et al., “Deep learning enables robust assessment and selection of human blastocyst after in vitro fertilization,” npj Digital Medicine, 2,
- B Ting et al., “Recent Advances of Deep Learning in Bioinformatics and Computational Biology,” Front Genet, 10, 214-220 (2019). PMID: 30972100.