[Application of the blastomere count variations "skip value" in the embryo AI assessment].

Objective: To explore the correlation between blastomere count variations "skip value" which extracted from by time-lapse technology (TLT) combined with artificial intelligence (AI) and morphological features of in vitro fertilization (IVF) embryo, and to test its feasibility in clinical applications. Methods: This study was a diagnostic experiment (AI reassessment of embryo transferred patients), a total of 6 545 embryos from 1 226 patients who underwent IVF at the Women and Children's Hospital of Chongqing Medical University from December 2020 to December 2021 were retrospectively analyzed, of which 2 869 embryos were attempted to cultured to blastocyst stage by TLT. The embryo dynamic map (EDM) was drawn by Embryo Viewer, a TLT recording software, based on embryo developmental kinetics. The self-developed AI embryo evaluation software identified and recorded the number of cleavages in real time during embryonic development, and compared with the EDM, the correlation between the skip value formed by the change of cleavage sphere counts and the outcomes of the embryos was analyzed. The correlation among skip value, morphological score of embryo, implantation rate and live birth rate were performed by Spearman and step-up logistic regression. The receiver operating characteristic (ROC) curve was selected for reporting there relationship of skip value and morphology. Finally, predicting power of skip value for implantation and live birth rate were performed by ROC analysis. Results: The total skip values extracted from the blastomere count of embryos (72 hours post-fertilization) were negatively correlated with abnormal cleavage, blastocyst formation rate, day 3 (D3)-cell score, uneven size and fragmentation (the β values were -0.268, -0.116, -0.213, -0.159 and -0.222, respectively; all P <0.001); positively correlated with D3-cell number ( β =0.034; P <0.001); negatively correlated with blastocyst formation rate and implantation rate ( OR =0.97, 95% CI : 0.93-0.99, P =0.034; OR =0.96, 95% CI : 0.93-0.98, P =0.044). The power of predicting implantation were similar between the order selection of skip values and traditional morphology criteria [area under curve (AUC): 0.679 vs 0.620]. Live birth rate were negatively correlated with female age ( OR =0.91, 95% CI : 0.88-0.93; P <0.001), D3 general score ( OR =0.77, 95% CI : 0.59-0.99; P =0.045) and order selection of skip values ( OR =0.98, 95% CI : 0.96-0.99; P =0.038), while positively correlated with retrieved oocyte number and endometrial thickness in embryo transferred ( OR =1.08, 95% CI :1.05-1.11, P <0.001; OR =1.09, 95% CI :1.06-0.12, P <0.001, respectively) from multivariate regression analysis, and the power of predicting live birth was 0.666 for AUC. Conclusions: The skip value and its order form is a systematic quantification of embryo development, correlated with embryo developmental quality and clinical outcome. It could be an addition parameter for embryo culture and selection.