P-166 An artificial intelligence-powered tool to score fresh donor oocytes and predict blastulation: interim analysis of a prospective investigation conducted at 3 laboratories

Abstract

Abstract Study question Is the Magenta-score associated with fresh donor oocytes’ blastulation competence, and can it be used to predict blastocyst rates per cohort of inseminated oocytes (BRpC)? Summary answer Developmentally-competent fresh donor oocytes showed significantly higher Magenta-score, although the number of blastocysts obtained was within the predicted range in only 63% of the cycles. What is known already Currently, many couples undergo donor oocyte treatments. However, despite donors being theoretically fertile, not all oocytes are developmentally competent, and many do not reach the blastocyst stage. Even if subjective and ineffective to predict oocyte competence, visual oocyte morphological assessment is still routinely used to evaluate oocytes. Lately, the rise of artificial intelligence (AI) resulted in promising tools to assess oocytes more objectively in a standardized method, possibly providing novel insights into the prediction of their competence. In this context, Magenta-score (Future Fertility), an AI-based image analysis tool, can potentially prove useful to optimize the management of oocyte donation treatments. Study design, size, duration A Blinded cohort study (June-2023/December-2023) was conducted to assess the primary outcome: association between fresh donor-oocytes’ Magenta-Score and their subsequent blastocyst development (assumed to be 50%). A sample size of 779 achieves 80%-power to detect the difference between the null-hypothesis point-biserial-correlation of 0 and the alternative hypothesis of 0.1 using a two-sided test with 5%-significance. Secondary outcomes were differences (i) predicted - true BRpCs, and (ii) predicted - true numbers of blastocysts obtained. Participants/materials, setting, methods Interim-analysis of 514 fresh-oocytes obtained from 63 donors at 3 centers and allocated (7.7±3, range:3-18) to 67 recipients. Pictures of denuded mature oocytes were acquired before ICSI. Magenta-scores were generated blindly (providing only donor age and total number of mature oocytes in cohort to the software). The software also estimated BRpCs and range of blastocysts obtainable from each cohort. Donors’ characteristics, ovarian stimulation/treatment cycle parameters, and sperm analyses were tested as confounders. Main results and the role of chance The only visual oocyte anomaly associated with lower Magenta-score was irregular shape (N = 484,5.4±2.2 vs N = 30,3.3±2.1, Mann-Whitney-U<0.01). Oocytes that developed into blastocysts (N = 194) had significantly higher Magenta-scores versus oocytes that did not reach this milestone (N = 320) (5.8±2.1 vs 4.9±2.3, Mann-Whitney-U<0.01; Odds-Ratio adjusted for sperm motility and incubator [standard/time-lapse]=1.26, 95%CI:1.13-1.34, p < 0.01; power=99.6%). No association was noted with day of blastulation or morphological quality. Data were concordant across centers. The average predicted-BRpC was 39.4±11.4% versus a true-BRpC of 40.6±22.6% (Person’s correlation: -0.24; mean difference: -1.1±27.6%, 95%CI from -7.8% to + 5.6%, p = 0.74). The average predicted number of blastocysts per oocyte cohort was 3.3±1.6 versus a true average of 2.9±1.7 (Person’s correlation: 0.25; mean difference: +0.4±2.0, 95%CI from -0.04 to + 0.9, p = 0.07). In 21% (N = 14) and 31% (N = 21) of cycles, the tool predicted an equal or lower number of blastocysts, respectively. In 5 cycles, no blastocyst was obtained, but the tool did not predict this unexpected outcome. When testing the predicted range of blastocysts obtained, the true number was within the range in 63% (N = 42) of cases and higher than the maximum predicted number in 13% (N = 9) of cases. No confounder was identified on the BRpC in this population of fresh donor cycles. Limitations, reasons for caution Of note, this is an interim analysis of an ongoing study. Not all mature oocytes recovered from a donor were utilized, thus not all oocytes in a cohort were imaged and included in analysis. Other putative confounders on BRpC should be considered in future studies with a larger sample size. Wider implications of the findings Effective management of an oocyte donation program is critical to comply with expected high success rates while minimizing the number of surplus blastocysts produced. The integration of genomic data, recipient couples’ characteristics, and the benefits of objective standardized AI-powered oocyte scoring is the most promising workflow to optimize this task. Trial registration number Not applicable