Artificial intelligence detects aneupoloidy non-invasively

Artificial intelligence (AI) could be used in future as a non-invasive tool for predicting aneuploidy in embryos new research demonstrates.

Aneuploidy is thought by some to be a significant factor contributing to whether an IVF embryo will fail to implant or result in a healthy pregnancy. The current test PGT-A is an invasive technique that requires a physical biopsy of cells taken from the embryo to analyse its chromosomes. Researchers at Weil Cornell Medicine in New York have developed non-invasive and automated method of embryo evaluation which they hope will make embryo selection more accurate, more affordable and less risky.

'Our hope is that we'll ultimately be able to predict aneuploidy in a completely non-invasive way, using artificial intelligence and computer vision techniques,' said Dr Iman Hajirasouliha, senior author of the paper published in The Lancet Digital Health.

To create the algorithm, named STORK-A, data on 10,378 embryos from 1385 patients were analysed. All the embryos had undergone PGT-A.

Microscope images of the embryo, as well as information about maternal age and the IVF clinic's scoring of the embryo's appearance, were all included, and machine-learning and deep-learning approaches were used to arrive at the algorithm.

'The algorithm turns tens of thousands of embryo images into AI models that may ultimately be used to help improve IVF efficacy and further democratise access by reducing costs,' said co-author Professor Olivier Elemento.

Initial validation tests of STORK-A, showed a 69.3 percent accuracy at predicting aneuploid versus euploid embryos and 77.6 percent at predicting complex aneuploidy versus euploid embryos. They later tested the algorithm on independent datasets and labelled images, including one from an IVI Valencia in Spain, and found comparable results.

The research builds on the team's existing work from 2019, where they developed an algorithm that could assess embryo quality on par with IVF clinic staff (see BioNews 994). The study provides initial proof of concept for an AI approach. Still, standardising use of STORK-A in clinics would require clinical trials to compare it to the current standard: PGT-A.

'This technology is being optimised with the hope that at some point its accuracy will be close to genetic testing, which is the gold standard and is more than 90 percent accurate,' said co-author Professor Zev Rosenwaks, 'But we realise that this goal is aspirational.'