Artificial intelligence-driven oocyte assessment for predicting blastulation and high-quality blastocyst formation in severe male factor infertility.

Objective: To study whether AI-driven oocyte evaluation is associated with blastocyst development and quality in couples with severe male factor infertility (SMF) undergoing intracytoplasmic sperm injection (ICSI) cycles.

Design: Cohort study SUBJECTS: 14,602 oocyte images from 2,156 ICSI cycles performed between January 2020 and May 2024 in a private university-affiliated in vitro fertilization center. Cycles were categorized into two groups: SMF (n=200 cycles, 1,478 embryos) and non-SMF (n=1,956 cycles, 13,124 embryos). SMF was defined as <5 million sperm in the ejaculate.

Exposure: Oocyte images were captured before ICSI and scored using the AI tool MAGENTA™. The predictive value of Magenta Scores (MS) on embryonic development was assessed. The association between MS and oocyte fertilization and blastocyst formation was analyzed.

Main outcome measure(s): Oocyte fertilization, blastulation rate, blastocyst quality.

Results: MS were significantly lower in oocytes that failed to fertilize compared to those that successfully fertilized (5.00 ± 0.04 vs. 6.44 ± 0.03, p<0.001). Blastulation rate was lower in the SMF group (46.61% vs. 50.80%, p=0.003), and blastocysts exhibited higher MS than non-blastocysts (5.12 ± 0.3 vs. 6.69 ± 0.3, p<0.001). The top-quality blastocyst rate was lower in SMF (56.6% vs. 65.2%, p<0.001), and high-quality blastocysts had higher MS than lower-quality ones (7.2 ± 0.6 vs. 6.8 ± 0.5, p<0.001). Among SMF cycles, MS were lower in oocytes that failed to fertilize (4.91 ± 0.12 vs. 6.34 ± 0.10, p<0.001). MS also differed between embryos that reached the blastocyst stage and those that did not (6.70 ± 0.11 vs. 4.96 ± 0.10, p<0.001). Top-quality blastocysts had significantly higher MS than others (7.00 ± 0.21 vs. 6.39 ± 0.19, p<0.001). Paternal age negatively correlated with fertilization, blastulation, and blastocyst quality; however, differences remained significant after adjusting for paternal age.

Conclusion: AI-based oocyte evaluation is associated with fertilization, blastulation, and blastocyst quality in SMF couples undergoing ICSI cycles. MS values were consistently higher for blastocysts than non-blastocysts, demonstrating the AI tool's utility in identifying oocytes with greater developmental potential, regardless of male infertility factors. However, the absence of sperm-specific factors in the MAGENTA™ algorithm may limit its ability to fully account for male infertility.