L26/P-827 Automated antral follicle count for prediction of low and high ovarian response: a prospective study

Le résumé fourni par la source

Abstract Study question Is AI-based automated antral follicle count as effective as manual AFC and AMH in predicting poor and high ovarian response to stimulation? Summary answer Automated AFC showed comparable predictive performance to manual AFC and AMH for identifying low and high ovarian responders. What is known already Antral follicle count (AFC) and anti-Müllerian hormone (AMH) are established biomarkers of ovarian reserve and widely used to predict ovarian response to controlled ovarian stimulation in IVF. Manual AFC assessment is operator-dependent and associated with inter- and intra-observer variability, which may affect clinical consistency. Recent advances in artificial intelligence have enabled automated ultrasound-based follicle detection and measurement, aiming to improve standardization and efficiency. While automated AFC systems have shown promising technical performance, prospective clinical evidence directly comparing their predictive value with conventional manual AFC and AMH for clinically relevant ovarian response outcomes remains limited. Study design, size, duration This prospective observational study included 334 IVF stimulation cycles performed at CEGYR (Argentina) clinic between September 2023 and November 2025. Predictive performance of automated AFC, manual AFC, and AMH was evaluated for low and high ovarian response outcomes. Participants/materials, setting, methods A total of 318 women (mean age 37±4 years) undergoing IVF were included. AMH, manual AFC, and automated AFC were available for 287, 243, and 149 cycles, respectively. Automated AFC measured follicles 2–10 mm using longest and mean perpendicular diameters. Low and high ovarian response were defined as < 4 and >15 oocytes retrieved, respectively. Predictive performance was assessed using ROC AUC, consistent with ESHRE ovarian stimulation guidelines. Main results and the role of chance Across the full dataset, automated AFC demonstrated strong predictive performance for ovarian response. For prediction of low ovarian response (<4 oocytes), AMH achieved a ROC AUC of 0.800 (95% CI 0.745–0.848), manual AFC 0.723 (0.654–0.792), automated AFC using longest diameter 0.840 (0.786–0.893), and automated AFC using mean perpendicular diameters 0.837 (0.788–0.888). For prediction of high ovarian response (>15 oocytes), AMH yielded a ROC AUC of 0.653 (0.571–0.736), manual AFC 0.669 (0.579–0.753), automated AFC longest diameter 0.657 (0.538–0.767), and automated AFC mean diameter 0.648 (0.524–0.768). In the subset of 100 cycles with complete data for AMH, manual AFC, and automated AFC, predictive performance remained consistent. ROC AUC for low response prediction was 0.817 (0.730–0.896) for AMH, 0.812 (0.704–0.906) for manual AFC, 0.822 (0.747–0.894) for automated AFC longest diameter, and 0.822 (0.747–0.894) for automated AFC mean diameter. For high response prediction in this subset, ROC AUC values were 0.575 (0.438–0.701) for AMH, 0.598 (0.476–0.722) for manual AFC, 0.611 (0.486–0.732) for automated AFC longest diameter, and 0.607 (0.466–0.729) for automated AFC mean diameter. Limitations, reasons for caution Not all cycles had complete data for AMH, manual AFC, and automated AFC, reducing sample size for combined analyses. The study was conducted at a single center, which may limit generalizability. Wider implications of the findings AI-based automated AFC may offer a standardized, reproducible alternative to manual follicle counting while maintaining predictive accuracy for ovarian response. This could improve clinical efficiency and reduce operator-dependent variability in ovarian reserve assessment. Trial registration number No

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.