AI “co-pilot” in the PMA laboratory: identifies embryos that will become blastocysts

Three studies coordinated by the Italian research team of the Genera group presented at the 41st congress of the European Society of Human Reproduction and Embryology ( Eshre ) underway in Paris, demonstrate how the integration between artificial intelligence (AI) and time-lapse technology is radically changing clinical practices in reproductive medicine. And AI is increasingly establishing itself as an 'assistant', 'co-pilot' of the embryologist in assessing the quality of oocytes, the development of embryos and the ideal time for embryonic biopsy, promising a more objective, personalized, reproducible and efficient management of assisted fertilization treatments.

Predicting embryo development The first study, carried out in collaboration with the University of Pavia and thanks to an Eshre grant, involved over 6,000 oocytes inseminated via ICSI and cultivated in EmbryoScope time-lapse incubators, analyzing cytoplasmic movements in the first 3 days of development. The researchers trained three AI models – ROCKET, LSTM-FCN and ConvTran – using algorithms to transform some videos into analyzable temporal data. The models achieved a predictive accuracy of blastocyst development , therefore of the embryo's ability to mature correctly before transfer to the uterus (day 5-7 after insemination), of 63% already on day 1 and up to 70% on day 3. "The models were developed without any manual intervention, using exclusively time-lapse data and without additional annotations - comments Danilo Cimadomo , Research Manager of the Genera group - it is a significant step forward towards earlier and more objective embryonic assessments. AI, integrating cytoplasmic dynamics, development timing and patient metadata, will be able to provide increasingly personalized and reliable clinical support tools in the future. The advantage of not being determined by manual interventions by operators is positive because it saves time to invest in more cognitively active and important functions, and aims at AI-operators collaboration based on observations of different characteristics".

AI and oocyte quality: the Magenta-score in donations The second multicenter study, conducted in Spain, evaluated 1,275 oocytes from egg donors using the Magenta-score algorithm, developed by Future Fertility. Applied blindly before insemination, the system predicted with good accuracy the probability of developing into blastocysts. Oocytes that developed into blastocysts had significantly higher Magenta scores than those that did not, demonstrating the predictive efficacy of the tool. Furthermore, in 72% of cases, the number of blastocysts obtained fell exactly within the range predicted by the algorithm or was even higher. Among completed cycles, the Magenta score also showed a good association with the cumulative live birth rate.

Biopsy Timing: Greater Safety and Standardization The third study addressed one of the most delicate procedures in the IVF process: embryo biopsy for preimplantation genetic diagnosis. By analyzing 1,943 blastocysts cultured between 2013 and 2020 and biopsied over several years and by seven operators, researchers demonstrated how the use of TLM combined with AI (in particular with Fairtility's CHLOE system) allows identifying the optimal moment for the biopsy in a uniform and reproducible way. "The moment in which the biopsy is performed is crucial for the quality of the genetic result and for the safety of the embryo," Cimadomo emphasized . "AI can help monitor the expansion in real time and guide the embryologist in choosing the most appropriate moment for the procedure."

But AI won't replace embryologists “These three studies – comments Laura Rienzi , embryologist and Scientific Director of the Genera centers – demonstrate how artificial intelligence and time-lapse technology can provide integrated solutions to optimize oocyte selection, predict embryonic competence early and standardize laboratory practices, reducing variability and subjectivity. From the initial analysis of oocytes to preimplantation biopsy, reproductive medicine enters a new era, based on data, algorithms and personalized medicine”.

Even in a recent editorial in the journal RBMO (Reproductive BioMedicine Online) with the provocative title: “ Do we still need embryologists? ”, Prof. Rienzi highlighted that “if on the one hand AI algorithms can predict the implantation potential of embryos, as demonstrated by several comparative studies, on the other hand data interpretation , management of complex cases and the creation of new performance indicators remain irreplaceable skills of the embryologist. The literature cited in the editorial suggests that AI has a performance comparable, but not superior, to human experience , especially in decision-making activities. Even experiences such as robotic vitrification or automated ICSI show promising results, but require further large-scale clinical validation”.