Association between a deep learning-based scoring system with morphokinetics and morphological alterations in human embryos

Abstract

Research questionWhat is the association between the deep learning-based scoring system, iDAScore, and biological events during the pre-implantation period? DesignRetrospective observational study of patients (n = 925) who underwent oocyte retrieval in a clomiphene citrate-based minimal stimulation cycle and obtained expanded blastocysts between October 2019 and December 2020. The association between iDAScore with morphokinetics and morphological alteration during fertilization, cleavage stage, compaction and blastocyst stage was analysed. ResultsThe duration of the cytoplasmic halo was significantly prolonged in low-scoring blastocysts (P < 0.0001). The timing of female and male pronuclei breakdown was significantly delayed in low-scoring blastocysts compared with high-scoring blastocysts (P < 0.0001 in both). Embryos with either trichotomous, multi-chotomous, rapid or reverse cleavage or asymmetric division had a lower score than embryos with normal cleavage (P < 0.0001–0.0098). The cell number and amount of blastomere fragmentation on days 2 and 3 were significantly associated with iDAScore (P < 0.0001–0.0008). Delayed compaction, blastulation and blastocyst expansion were observed in low-scoring embryos (P < 0.0001 in all). The incidence of blastomere exclusion and extrusion during embryonic compaction was significantly higher in low-scoring embryos than in high-scoring embryos (P ≤ 0.0001 in both). Blastocyst morphology was significantly associated with iDAScore (P < 0.0001). Multiple linear regression analysis revealed that, during the transformation to blastocyst stage, morphokinetic and morphological events were strongly associated with iDAScore (P < 0.0001–0.0116). ConclusionsiDAScore was significantly correlated with morphokinetics and morphological alterations of pre-implantation embryos, especially during the late pre-implantation period. Our findings contribute to research on deep learning model-based embryo selection, which may provide patients with a compelling explanation of blastocyst selection.