Time-lapse monitoring has enabled quantitative embryo morphokinetic descriptors, yet the predictive value of such parameters for embryo ploidy remains debated. We present an applied machine-learning workflow that combines time-lapse morphokinetics with reproductive history variables to support pre-implantation embryo ploidy prediction. The study is a double-blind, retrospective longitudinal cohort including 383 blastocysts from 193 patients undergoing PGT-A/PGT-SR (two IVF clinics, 2017–2019). Following data pre-processing and augmentation, model development used feature selection (e.g., Pearson correlation and tree-based ranking) and stratified five-fold validation/testing. The most informative predictors included maternal age, PGT indication, and cleavage-to-morulation time intervals (e.g., tM–t5…tM–t9, tPNf, t4, t7–tPNf). The resulting classifier achieved ROC AUC ≈ 0.70–0.71, with mean accuracy ≈ 0.68 and precision ≈ 0.73 across folds. While the model cannot replace genetic testing, it can act as an optimization-oriented decision support component for embryo prioritization when PGT is not performed, motivating future work on larger multi-clinic datasets and calibration for clinical deployment.
Tsakalis et al. (Sun,) studied this question.