Comprehensive assessment of pronuclear morphological pattern prognostic value using machine learning approaches in IVF programs

Objective To comprehensively evaluate the prognostic value of pronuclear Z-score morphological assessment in IVF programs, and to explore whether integrating machine learning (ML) approaches improves prediction of embryo development and implantation outcomes. Design A retrospective analysis of 5,524 zygotes from 742 IVF/ICSI cycles. Standard regression analysis was applied to assess associations between Z-score categories and implantation. Advanced ML techniques—including principal component analysis (PCA), K-means clustering, and ensemble neural networks—were implemented to identify hidden developmental patterns and construct a prognostic implantation model. Results Conventional regression confirmed the limited predictive value of Z-score categories for implantation potential. Cluster analysis identified distinct subgroups of embryos, with Z1 and Z2 patterns associated with superior blastocyst formation and quality. Neural network modeling demonstrated that merging Z-score categories into broader groups (e.g., Z1/Z2/Z3 vs. Z4/Z5) improved alignment between predicted and observed clinical pregnancy rates. Notably, model predictions tended to overestimate outcomes for common Z-patterns, underscoring the limitations of single-timepoint morphology. Conclusions While Z-score morphology provides useful descriptive information, its stand-alone prognostic power for implantation is limited. Integrating morphological assessment with ML-based clustering and neural networks allows more nuanced embryo classification and may enhance predictive accuracy. These findings highlight the need for simplified Z-score groupings and future prospective studies incorporating live birth outcomes, time-lapse morphokinetics, and molecular markers to optimize embryo selection strategies in IVF.