P-115 A deep learning-based system for oocyte image assessment predicts the outcomes of intracytoplasmic sperm injection and morphokinetic fate in human preimplantation embryos

Abstract

Study question How is the deep learning-based oocyte assessment system, Magenta, related to the outcomes of intracytoplasmic sperm injection (ICSI) and biological events during the preimplantation period? Summary answer The Magenta score is predictive of delayed fertilization events in the pronuclei (PNs) and cytoplasm, abnormal blastomere cleavage, compaction errors, and impaired blastulation and expansion. What is known already Morphological assessment of oocyte quality remains challenging because of the lack of objective criteria. Artificial intelligence (AI) has recently been implemented in assisted reproductive technologies, leading to the development of several AI-based evaluation systems for oocytes and embryos. Magenta is a deep learning-based tool that predicts the embryonic development of mature oocytes to the blastocyst stage. It provides individual oocyte scores where higher scores indicate greater likelihood of blastocyst development. However, the specific relationship between Magenta score and morphokinetic parameters is unclear. Additionally, Magenta was developed using data from controlled-ovarian stimulation cycles; therefore, its effectiveness for minimal-stimulation cycles remains uncertain. Study design, size, duration This retrospective study included 2,950 images of mature oocytes from 1,487 oocyte retrieval cycles (average cohort size=1.99±1.11; 1,231 patients; mean age 38.7±4.0 years). Patients underwent clomiphene citrate-based minimal ovarian stimulation followed by ICSI between October 2019 and December 2020. Images were obtained immediately post-ICSI. Surgical sperm retrieval and/or recurrent implantation failure cases were excluded. Oocytes unsuitable for observation (i.e. poor image quality) were excluded. Magenta analysed individual mature oocytes, providing scores from 0 to 10. Participants/materials, setting, methods The microinjected oocytes were cultured for 4–7 days in a time-lapse incubator (Embryoscope+/Flex). Fertilization and embryo development at each stage were assessed with EmbryoViewer software, and key phenomena, including meiotic resumption, PN/cytoplasmic dynamics, cleavage patterns, blastomere compaction, and embryo quality were manually annotated. The relationship between the Magenta score, embryonic and pregnancy outcomes, and morphokinetics was analysed using generalized estimating equations, adjusting for bias using covariates and confounders to verify the statistical significance. Main results and the role of chance The mean Magenta score was 5.3±2.9 (median: 5.4; quartiles: 2.6, 8.0). A lower Magenta score correlated with a higher degeneration rate (P = 0.0006) and a lower normal fertilization rate (P = 0.0004) per ICSI. Additionally, the Magenta score positively correlated with the incidence of the first cleavage, development to the 8-cell stage, compaction, blastulation, and blastocyst expansion (P < 0.0001–0.0333). Although no correlation was observed between the Magenta score and the incidence of fertilization events, a low score was associated with delays in each fertilization event, including second polar body extrusion, cytoplasmic waves and halos, PN appearance and breakdown (P < 0.0001–0.023). Furthermore, a low score was associated with delayed first cleavage and a higher incidence of abnormal cleavage, aberrant blastomere movement, and multinucleation (P < 0.0001–0.0016). A lower Magenta score was also associated with a higher incidence of early compaction (P = 0.0133) and blastomere exclusion (P = 0.0002). No correlation was observed between the Magenta score and the time of blastocyst expansion; however, a low score was associated with poor morphology of the inner cell mass and trophectoderm (P < 0.0001–0.0031). Magenta score was not associated with pregnancy outcomes after single blastocyst transfers. Limitations, reasons for caution This study has some limitations owing to its retrospective design. Independent validation from other research groups is necessary, as data were obtained from a single centre. Wider implications of the findings We demonstrated a significant correlation between the Magenta score and embryonic outcomes, including morphokinetic and morphological changes during the preimplantation period. Our findings contribute to accumulating knowledge on oocyte assessment using deep learning models, which can enhance standardization and provide greater transparency to patients throughout the IVF process. Trial registration number No