Artificial intelligence-derived oocyte morphology and follicular fluid biomarkers in donors

In brief: Accurate assessment of oocyte competence remains challenging in assisted reproduction. Artificial intelligence morphology analysis is a promising noninvasive tool. This study demonstrates that oocyte morphology scores are associated with distinct molecular signatures in follicular fluid from young oocyte donors, supporting a biological basis for computational oocyte assessment. Abstract: Oocyte quality is a key determinant of reproductive success, yet its assessment in assisted reproduction largely relies on subjective morphological criteria. Artificial intelligence-based image analysis has introduced greater objectivity into oocyte evaluation; however, the biological features captured by artificial intelligence-derived morphological scores remain incompletely defined. In this study, we integrated artificial intelligence-based oocyte morphology with molecular profiling of follicular fluid (FF) to identify biological correlates of oocyte competence. Reproductive outcomes were analysed in 49 young oocyte donors (20-33 years), while FF samples pooled per woman from a subset of 25 donors were analysed for metabolic (glucose, total cholesterol, triglycerides, high-density lipoprotein [HDL], low-density lipoprotein [LDL], apolipoprotein A1 [APOA1]), extracellular matrix-related (heparan sulfate proteoglycan 2 [HSPG2]/Perlecan), signaling-related (Gremlin-1), and fertility-related (anti-Müllerian hormone [AMH], LH, FSH) biomarkers. Artificial intelligence-derived oocyte quality scores were positively associated with specific intrafollicular markers, including glucose, total cholesterol, HDL, AMH, and HSPG2, while no associations were observed with triglycerides, LDL, Gremlin-1, LH, or FSH. APOA1 showed a positive trend with the artificial intelligence score. These findings provide a biological context for artificial intelligence-based oocyte morphological assessment by linking digital image-derived scores with metabolic and structural features of the follicular microenvironment. The integration of artificial intelligence-driven morphology with donor FF biomarker profiling may contribute to the development of more objective and biologically informed approaches for oocyte quality evaluation in assisted reproduction.