ERICA's (Embryo Ranking Intelligent Classification Assistant) ranking, based on ploidy prediction, is strongly correlated with pregnancy outcomes

Abstract Study question How does ERICA perform when ranking the most suitable embryos for transfer in terms of clinical pregnancy, and the presence of a fetal heartbeat (FHB)? Summary answer ERICA’s Artificial Intelligence ranking system was positively correlated with outcomes defined as implantation and presence of FHB. Best-ranking embryos outperformed lower-ranking embryos by statistical significance. What is known already ERICA, the Embryo Ranking Intelligent Classification Assistant, is a deep learning AI system trained to rank embryos based on their ploidy status, which is highly correlated with successful treatments. ERICA ranks the embryos according to their prognosis predictions and labels them into four quality categories: optimal, good, fair, and poor. ERICA’s performance in the clinic remains to be tested. Study design, size, duration Retrospective analysis on ERICA’s performance over 4 consecutive months after quality assurance and fine-tuning processes. We compared both the ranking and prognosis of the AI algorithm against clinical outcomes in IVF cycles and subsequent embryo transfers. For this study, all cycles where ERICA was used to assist embryologists during the embryo selection process were included. Double embryo transfers with a single FHB where excluded. Participants/materials, setting, methods Total 77 cycles with 81 transfers of 98 embryos (17 cases underwent a double embryo transfer) from two IVF clinics. Evaluated clinical outcomes included biochemical pregnancy test (defined as beta human chorionic gonadotropin >20 mUI/ml), and presence/absence of FHB. We compared the ERICA rankings and predictions against outcome and a sub-analysis was performed on transferred embryos with known ploidy status (14 embryos). Main results and the role of chance The distribution of embryos within the ERICA categories are 42% for optimal, 38% for good, 19% for fair, and 6% for poor. The observed biochemical pregnancy rate was 51%, 25%, 47% and 33% respectively, and 39%, 22%, 42%, 17% for FHB. We found statistical significance (Z = 1.78; p = 0.0378) for the proportion of biochemical pregnancy between transfers labelled by ERICA as optimal (51%) and all lower rankings (33%). The proportion of transfers with presence of FHB within the optimal group was 39%, compared with 29% for the rest of the embryos. This did not show statistical significance (Z = 1.141; p = 0.127). Additionally, we observed that the proportion of biochemical pregnancy and presence of FHB in the group of transfers with known ploidy (n = 14) was 50% and 36% respectively, and the transfers with unknown ploidy and labelled as optimal by ERICA (n = 35) was 54% and 43% respectively. Limitations, reasons for caution This is the first report on ERICA’s performance on real clinical data, and despite being a relatively small dataset, we observed statistical significance of the embryos labelled by ERICA as having optimal quality. Further studies should be conducted with larger datasets and more clinics included to strengthen the evidence. Wider implications of the findings: This is the first report on ERICA’s performance on real clinical data, and despite being a relatively small dataset, we observed statistical significance of the embryos labelled by ERICA as having optimal quality. Further studies should be conducted with larger datasets and more clinics included to strengthen the evidence. Trial registration number Not applicable