P-180 Enhancing pregnancy prediction performance by combining pregnancy prediction AI with euploid prediction AI

Abstract Study question Can the combination of two different AI models, Life Whisperer “Viability” and Life Whisperer “Genetics” improve the performance of pregnancy prediction of embryos for transfer? Summary answer Combining the two AI models, improves pregnancy prediction performance, with optimal results achieved when the weight ratio of the two models is set to 8:2. What is known already Recent advancements in AI-based embryo image analysis have significantly improved pregnancy prediction. Life Whisperer “Viability,” (Via) predict pregnancy from a single blastocyst image and correlates with pregnancy rates. Similarly, Life Whisperer “Genetics” (Gen) predicts euploid embryos and correlates with euploidy rates. For successful pregnancy, both embryo morphology and euploidy status are crucial. Therefore, combining the pregnancy prediction AI, Viability, with the euploidy prediction AI, Genetics, could improve the performance of pregnancy predictions. Study design, size, duration This retrospective cohort study, involved embryos cultured in EmbryoScope+/8 (Vitrolife) that reached blastocyst stage (Gardner classification ≥3) on Day 5, vitrified, and then used for single blastocyst transfer (between 2018-2021). Images from Day 5 (110-120 hours of culture) were scored using two AI models: Gen and Via, and fetal heartbeat was used as the outcome variable. Study 1: 148 embryos (1st embryo transfer cycle) analyzed. Study 2: 693 embryos (all embryo transfer cycles) analyzed. Participants/materials, setting, methods Study 1: Various weight ratios of these two models (10:0, 9:1, 8:2, etc.) were analyzed to determine the optimal prediction performance based on the ROC Area Under Curve (AUC). Study2: Using the ratios obtained in Study 1, the pregnancy prediction performance of Via and the combination scores (Combine) were compared in all single blastocyst transfer cycles. The two-tailed DeLong’s test was used to compare AUCs. The Cochran-Armitage test was then used to analyze the trend. Main results and the role of chance Study 1: The AUC for the individual models was 0.654 for Via. The highest prediction performance (AUC = 0.7) was achieved when the weight ratio of Gen:Via was set to 8:2 (Combine). Comparing the AUCs of Via and Combine, there was a trend toward improved pregnancy prediction performance with Combine (P = 0.052). Study2: The AUC for pregnancy prediction performance in Via was 0.623 compared to 0.657 in Combine, which was significantly higher (P = 0.002).The scores were divided into four categories as recommended by the manufacturer (Low: 0-2.4, Medium: 2.5-7.4, High: 7.5-8.9, Very High:9.0-10.0), and a trend test for scores and pregnancy rates was conducted. The results showed that in both methods, there was a tendency for the pregnancy rate to increase as the score increased, with the trend being stronger in the Combine method. (Via: Low: 27.4% (2/18), Medium: 41.2% (34/131), High: 56.0% (67/187), Very High: 50.6% (194/357); P > 0.0001, Combine: Low: 14.3%, (3/21) Medium: 26.7% (46/172), High: 41.6% (92/221), Very High: 55.9% (156/279); P > 0.0001). Limitations, reasons for caution This is a retrospective study performed in a single clinic. In addition, the study was limited to transfer cycles of embryos where blastocyst formation was reached on day 5 of culture. Wider implications of the findings This study suggests that combining two AI models, each focused on different aspects of embryo analysis, can enhance pregnancy prediction performance. This approach could contribute to improving the decision-making process for embryo transfer, potentially leading to higher success rates in assisted reproduction. Trial registration number No