JBRA Assist. Reprod. 2025;29(Suppl 1):32-32
POSTER PRESENTATION
doi: 10.5935/1518-0557.20250086
1Research and Development Department - Huntington Medicina Reprodutiva - Eugin Group, São Paulo, SP, Brazil
2Embryology Department - Huntington Medicina Reprodutiva - Eugin Group, São Paulo, SP, Brazil
3Clinical Department - Huntington Medicina Reprodutiva - Eugin Group, São Paulo, SP, Brazil
Objective: Artificial intelligence (AI) tools are increasingly used to assist embryologists in selecting the most viable embryo for transfer and optimizing gamete selection in assisted reproductive technology (ART) procedures. AI algorithms have been integrated into embryology laboratories to enhance ART outcomes and support personalized patient decision-making. This study presents a multicenter validation of Magenta, an AI software by Future Fertility (Fjeldstad et al., 2024), designed to evaluate oocytes and predict blastocyst development. Magenta incorporates all morphological grades in its predictions, including early-stage blastocysts (grades 1 and 2) and those with Gardner scores of A, B, or C. This study aims to assess how Magenta’s blastocyst formation predictions align with our center’s standard protocol for identifying usable blastocysts (expansion grade ≥3) for clinical use.
Methods: This prospective multicenter cohort study included 608 autologous metaphase II (MII) oocytes from 67 patients undergoing in vitro fertilization (IVF) between November 2024 and February 2025 across four ART centers. Blastocyst formation rates per fertilized oocyte was analyzed. Magenta defines a positive blastocyst as any embryo reaching at least Gardner grade 1CC between 4 and 7, while a negative blastocyst includes cases of failed fertilization or embryo arrest. Usable blastocysts were defined as those reaching expansion grade ≥3CC. Blastocyst formation rates and usable blastocyst rates were analyzed across Magenta score ranges: 0 - 2.5 (18.3%), 2.6 - 5 (38.2%), 5.1 - 7.5 (48.8%), and 7.6 - 10 (56.8%). Model performance was assessed using accuracy, sensitivity, specificity and the area under the curve (AUC) from receiver operating characteristic (ROC) analysis. Statistical tests included Kruskal-Wallis, Fisher’s test and Chi-squared tests, with significance set at p<0.05.
Results: The mean maternal age was 37.7±3.7 years. Maternal age was significantly lower in the Magenta 7.6-10 group (35.6±3.4 years) compared to the 0-2.5 (36.9±3.3), 2.6-5.0 (37.5±3.9) and 5.1-7.5 (37.8±3.6) groups (p<0.0001). The mean number of MII oocytes per patient was 9.2±5.6 [range: 1 - 28], with a similar distribution across the Magenta score groups (02.5: 25.5%; 2.6-5.0: 25.8%; 5.1-7.5: 23.5% and 7.6-10: 25.2%). Blastocyst formation per fertilized oocyte was 59%, while per MII oocyte; it was 45.8%, closely matching Magenta’s validation results (58.5% and 43.6%, respectively, Fjeldstad et al., 2024). Usable and all blastocyst formation rates per Magenta score range was 0 - 2.5: 26.5% and 28.4% (p=0.8); 2.6 - 5.0: 50.3% and 62.4% (p=0.04); 5.1 - 7.5: 48.3% and 53.8% (p=0.4); 7.6 - 10: 55.6% and 61.4% (p=0.35). The proportion of usable and all blastocysts formed achieved was lower in the 0-2.5 group compared to all other groups (p<0.0001), while no significant differences were observed among the higher-score groups. The mean Magenta score of MII oocytes that formed blastocysts (5.7±2.7) was significantly higher than those that did not (4.5±2.9, p<0.0001). Oocytes that developed into high-quality blastocysts (A/B grades) had significantly higher scores (5.8±2.6) than those failing to develop (4.5±2.9, p<0.0001) and similar scoring than those forming low-quality (C-grade) blastocysts (5.4±3.0, p=0.9). Performance analysis for predicting usable and all blastocyst formation yielded accuracy of 0.57 and 0.56, sensitivity of 0.56 and 0.55, specificity of 0.57 and 0.58, and AUC of 0.62 and 0.63, aligning with the developer’s reported validation results.
Conclusion: It is essential to clarify to patients that the Magenta algorithm predicts blastocyst formation based on broad morphological criteria, not exclusively usable blastocysts suitable for transfer. Oocytes with low Magenta scores (0-2.5) had less than a 30% chance of developing into blastocysts, while those scoring 2.6 to 10 had approximately a 50% probability. Our validation suggests that Magenta provides informative predictions, particularly for lower-scoring oocytes, offering.
Reference
Fjeldstad et al. Reprod Biomed Online. 2024;48:103842.