EUVIMEDEUVIMED entdecken
Crossref

L26/O-135 A real-time predictive alert system for monitoring IVF laboratory performance using time-lapse–derived embryo features

L. Bori, S. Rajendran, S. Marqueño, N. Zaninovic, I. Hajirasouliha, M. Meseguer

Human Reproduction · 01.07.2026 · 41 (Supplement_1)

DOI: 10.1093/humrep/deag083.135 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Human Reproduction, online erschienen am 08.07.2026. Verfasst von L. Bori und 5 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Study question Can a real-time, stage-aware predictive system informed by embryo morphokinetics anticipate deviations in IVF laboratory key performance indicators (KPIs) during embryo culture? Summary answer Using morphokinetic inputs, a real-time system forecasts usable blastocyst and implantation rates in daily cohort of embryos in culture, enabling early alerts of KPI underperformance. What is known already IVF laboratory performance is traditionally monitored using retrospective KPIs, limiting the ability to intervene during ongoing cycles. Consensus frameworks such as the Vienna and Maribor documents define standardized KPIs, but their application is mainly post hoc. Time-lapse incubators and artificial intelligence have shown associations with outcomes such as implantation and ploidy at the embryo level; however, few approaches address real-time monitoring of laboratory performance at the incubator-population level, i.e., in daily cohort of embryos in culture. Predictive quality control systems are widely used in other biomedical laboratories but remain underexplored in assisted reproduction. Study design, size, duration This retrospective cohort study included 70,456 embryos cultured at a single IVF center between January 2016 and December 2021. Embryos from autologous and donor ICSI cycles were cultured in time-lapse incubators and followed from fertilization to the blastocyst stage and, if transferred, implantation. Participants/materials, setting, methods A stage-aware system combined multiple morphokinetic submodels trained with gradient-boosted trees to estimate usable blastocyst and implantation probabilities using only features available at each developmental stage (t2, t3, t4, t5, tB, automatic score at 44 hours, at 66 hours and final). Submodel outputs were aggregated across all embryos simultaneously in culture to forecast incubator-population KPIs in real time. In parallel, an exploratory surveillance layer quantified day-level morphokinetic variability using outcome-agnostic percentile-based reference bands (Normal/Warning/Critical). Main results and the role of chance Predictive performance of individual submodels increased as additional developmental information became available: for usable blastocyst, discrimination improved from early cleavage timings alone (AUC 0.739, accuracy 0.731) to extended morphokinetics and automated scores (up to AUC 0.787, accuracy 0.785); for implantation, later-stage features yielded higher performance, reaching AUC 0.562 and accuracy 0.583 with inclusion of time to blastocyst. Importantly, submodel discrimination reflects partial information available at each developmental stage and was not intended for embryo-level clinical outcome prediction, but to support population-level KPI forecasting in real time. At the incubator-population level, the full system integrating all active submodels forecasted usable blastocyst and implantation rates with mean absolute error (proportion units) of 0.148 ± 0.099 and 0.147 ± 0.182 at the day level, improving to 0.103 ± 0.056 and 0.086 ± 0.099 for 7-day windows. Forecasts below pre-specified thresholds (usable blastocyst <51%, implantation <50%) generally coincided with observed sub-threshold periods, and aggregation over 7-day windows reduced forecast variability by approximately 30% for usable blastocyst and 42% for implantation. An exploratory alert layer based on outcome-agnostic percentile thresholds identified day-level variability in incubator-population morphokinetic parameters, defining Normal (P25–P75), Warning (P10–P25 and P75–P90), and Critical (<P10 or >P90) ranges. Limitations, reasons for caution This study represents a preliminary evaluation of a real-time monitoring system and currently targets two laboratory KPIs. Performance depends on the availability of real-time morphokinetic annotations, additional endpoints (fertilization, post-thaw survival) are not yet included, and alert thresholds may require local calibration. A multicenter validation is needed to assess generalizability. Wider implications of the findings Real-time KPI forecasting at the level of the incubator embryo population introduces a proactive quality-control layer in IVF laboratories. By complementing retrospective audits with early alerts during embryo culture, this approach may support timely intervention, enhance laboratory consistency, and provide a scalable framework for extending surveillance to additional KPIs. Trial registration number No

Zitieren

APA

Bori, L., Rajendran, S., Marqueño, S., Zaninovic, N., Hajirasouliha, I., & Meseguer, M. (2026). L26/O-135 A real-time predictive alert system for monitoring IVF laboratory performance using time-lapse–derived embryo features. Human Reproduction, 41(Supplement_1). https://doi.org/10.1093/humrep/deag083.135

Vancouver

Bori L, Rajendran S, Marqueño S, Zaninovic N, Hajirasouliha I, Meseguer M. L26/O-135 A real-time predictive alert system for monitoring IVF laboratory performance using time-lapse–derived embryo features. Human Reproduction. 2026;41(Supplement_1). doi:10.1093/humrep/deag083.135

BibTeX
@article{Bori2026,
  author = {Bori, L. and Rajendran, S. and Marqueño, S. and Zaninovic, N. and Hajirasouliha, I. and Meseguer, M.},
  title = {L26/O-135 A real-time predictive alert system for monitoring IVF laboratory performance using time-lapse–derived embryo features},
  journal = {Human Reproduction},
  year = {2026},
  volume = {41},
  number = {Supplement_1},
  doi = {10.1093/humrep/deag083.135}
}
Kostenlos starten →