Xiao-Cong Wang

Purpose: To determine whether day 2 to day 3 developmental trajectories improve day 3 embryo ranking and selection compared with day 3 cell number alone, thereby offering better prognostic information for cleavage-stage embryos, including fast-cleaving embryos whose developmental potential remains uncertain. Design: This retrospective multicenter study (2016–2024) developed a ranking framework using 27,002 embryos from one center and validated it in 1691 fresh day 3 single embryo transfers across three centers (maternal age ≤ 38 years; embryos derived from two pronuclei, with fragmentation ≤ 10%, symmetric, and no multinucleation for confounder control). Developmental trajectories combined day 2 and day 3 cell numbers. Mixed-effects logistic models assessed trajectory–blastulation associations; post hoc comparisons consolidated trajectories into four ranking categories. Validation used trend tests between ranking levels and pregnancy rates, multivariable mixed-effects logistic model to estimate adjusted odds ratios (aORs) between ranking levels, and AUC for pregnancy prediction. Results: Trajectory analysis reversed the advantage of day 3 8-cell over 9–16-cell embryos (30.7% vs. 24.2% high-quality blastulation; P < 0.001), with Normal_Fast (4 → 9–16) achieving the highest rate (40.7%), significantly higher than Normal_Normal (4 → 8; 35.1%; P < 0.001). Trajectories were consolidated into a four-category ranking: (1) Normal_Fast (4 → 9–16); (2) Normal_Normal (4 → 8); (3) Normal_Slow (4 → 6–7), Slow_Normal (< 4 → 8), Fast_Normal and Fast_Fast (> 4 → 8–16); and (4) all others. Ranking 1 had a higher likelihood of clinical pregnancy than ranking 2 (aOR = 2.06 [1.05–6.82]), although this validation was limited by sample size ( n = 24). Trajectory outperformed day 3 cell number in predicting high-quality blastocyst (AUC 0.776 vs. 0.729; P < 0.001) and clinical pregnancy (AUC original 0.619 vs. 0.596, P = 0.02; down-sampled 0.682 vs. 0.618, P = 0.016). Conclusions: For women ≤ 38 years undergoing fresh day 3 single embryo transfer, developmental trajectories from day 2 to day 3 improve embryo ranking and selection with other morphological features controlled (two pronuclei, fragmentation ≤ 10%, symmetry, no multinucleation). The 4 → 9–16 pathway emerges as potentially optimal, though requiring further validation. This dual-day framework treats both cell numbers as interdependent markers, where day 2 status redefines the priority of day 3 cell number, rather than serving only as a secondary criterion.

Background: Clarifying the impact of day-3 cell number on blastulation and subsequent pregnancy outcomes is essential for optimizing blastocyst selection criteria and refining embryo assessment protocols. While slow cleavage on day 3 is well-recognized as detrimental, the prognostic value of rapid cleavage (>8 cells) remains ambiguous. Methods: This retrospective cohort study (January 2015–April 2024) included 64,853 embryos undergoing blastocyst culture (Cohort 1) and 2,669 single-blastocyst frozen embryo transfer (FET) cycles (Cohort 2) at a large tertiary assisted reproduction center. Cohort 1 examined the association between day-3 cell number and blastulation potential. Cohort 2 evaluated clinical pregnancy, live birth, and miscarriage rates following single-blastocyst transfer using multivariable logistic regression adjusted for confounders. Embryos were stratified by maternal age ( 1.4; 95% CIs excluded 1). This pattern was consistent across female age subgroups. In Cohort 2, however, the advantage of 11–16-cell embryos translated into superior pregnancy and live birth rates only in younger women receiving top-quality blastocysts versus 8-cell embryos (76.5% vs . 63.0%, P = 0.002, aOR = 1.95 [1.30–2.96]; 61.2% vs . 51.2%, P = 0.034, aOR = 1.58 [1.10–2.30]). Conversely, in older women with non-top-quality blastocysts, 11–16-cell embryos predicted lower pregnancy and live birth rates (26.5% vs . 51.0%, P = 0.023, aOR = 0.40 [0.15–0.97]; 14.7% vs . 38.5%, P = 0.019, aOR = 0.32 [0.10–0.89]). The 9–10-cell embryos generally showed outcomes comparable to 8-cell embryos, except for a reduced live birth rate in the older, non-top-quality blastocyst subgroup (23.9% vs . 38.5%, P = 0.047, aOR = 0.51 [0.26–0.98]). Conclusion: Day-3 cell number serves as a context-dependent prognostic indicator for optimizing blastocyst selection. For young women with top-quality blastocysts, ≥11-cell embryos are the strongest candidates; conversely, 8-cell embryos appear optimal for older women with non-top-quality blastocysts.

Predicting blastocyst formation poses significant challenges in reproductive medicine and critically influences clinical decision-making regarding extended embryo culture. While previous research has primarily focused on determining whether an IVF cycle can produce at least one blastocyst, less attention has been given to quantifying blastocyst yields. This study aims to develop and validate such a quantitative predictive tool for IVF cycles. We employed three machine learning models—SVM, LightGBM, and XGBoost—which demonstrated comparable performance and outperformed traditional linear regression models (R²: 0.673–0.676 vs. 0.587, Mean absolute error: 0.793–0.809 vs. 0.943). Ultimately, LightGBM emerged as the optimal model, due to utilizing fewer features (8 vs. 10–11 in SVM/XGBoost) and offering superior interpretability. We then stratified predictions and actual yields into three categories (0, 1–2, and ≥ 3 blastocysts) to evaluate the model’s discriminative performance. In this multi-classification task, LightGBM demonstrated robust accuracy (0.675–0.71) with fair-to-moderate agreement (kappa coefficients: 0.365–0.5) across both the overall cohort and poor-prognosis subgroups. Feature importance analysis identified three critical predictors: the number of extended culture embryos, the mean cell number on Day 3, and the proportion of 8-cell embryos. By leveraging the potential of machine learning, this research provides clinicians with valuable insights for making individualized decisions regarding extended embryo culture.

The aim of this proof-of-concept study is to develop a predictive model based on deep learning algorithms to predict working alliances after the first therapeutic session and to provide a basis for...