EUVIMEDEUVIMED entdecken
Crossref

L26/P-283 Predicting pregnancy and live-birth using embryo segmentation dynamics across multi- centre IVF lab datasets

H. Misaghi, L. Cree, V. Padmanabhan, A. Storr, N. Knowlton

Human Reproduction · 01.07.2026 · 41 (Supplement_1)

DOI: 10.1093/humrep/deag083.619 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Human Reproduction, online erschienen am 08.07.2026. Verfasst von H. Misaghi und 4 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Study question Can objectively quantified embryo segmentation predict pregnancy across multiple clinics and time-lapse imaging platforms? Summary answer Of the 10 features segmented, higher blastocyst expansion rates correspond to higher pregnancy and live birth rates. What is known already Blastocyst architecture is fundamental to embryo grading and has been linked to IVF outcomes, but assessment is typically categorical and subjective. Time-lapse incubators enable objective measurement, yet comprehensive segmentation-based pipelines remain limited by restricted annotations and narrow structural coverage. As a result, quantitative and explainable embryo selection approaches that (i) learn from multi-object embryo morphology and (ii) yield clinically interpretable predictors with cross-site reproducibility are still emerging. There remains a need for scalable methods that transform routine time-lapse imaging into reproducible, validated, and clinically meaningful quantitative biomarkers. Study design, size, duration Retrospective multi-centre cohort study with 2,162 transferred blastocysts from six IVF clinics imaged on EmbryoScope (+) systems. A newly developed multi-object segmentation model was applied to generate a quantitative dataset. Univariate screening combined with multivariable feature selection were performed to identify the most informative predictors of pregnancy and live birth. The top-ranked features were evaluated in a focused expansion analysis using quintile stratification and correlation testing with follow-up to confirmed outcomes. Participants/materials, setting, methods Transferred blastocysts with segmentation-derived measurements and clinical labels were included. A deep-learning segmentation model (Anthesis®) which was developed to cover a broad set of embryonic ‘objects’ including cells, vacuoles, pronuclei, blastomeres, trophectoderm, inner cell mass, inner/outer zona, degenerated cells, and fragmentation was used to compute ‘object’ masks. Standard machine learning techniques were applied across ‘objects’ in both a univariate and multivariable analysis resulting in a biologically grounded and explainable interpretation of what is driving live-births. Main results and the role of chance Across 2,162 transferred blastocysts, 49.6% resulted in pregnancy, and 32.8% in live birth. Following dataset construction using the segmentation pipeline, univariate and multivariable analyses converged on a single, interpretable quantitative feature capturing blastocyst expansion dynamics: the ratio of maximal to minimal inner zona area during development (up to 116 hours). We observed a monotonic increase in the relationship between blastocyst expansion and success rates when examined by quintiles. The lowest quintile had a success rate of 37.0% and rose to 60.7% in the highest. Live birth rates increased similarly from 25.4% to 43.2%. This observed association between expansion dynamics and binary success was statistically significant (point-biserial r = 0.143, p < 0.0001), supporting a modest but consistent effect. The reproducible pattern of increasing success across pregnancy and live birth endpoints underscores the clinical interpretability of this AI-derived biomarker and illustrates how explainable embryo features can be incorporated into clinical practice. Limitations, reasons for caution Retrospective design and lack of adjustment for confounders (e.g., maternal age, stimulation, genetic testing, transfer context) limit causal interpretation. Wider implications of the findings Multi-object embryo segmentation enables scalable extraction of interpretable quantitative biomarkers from routine time-lapse imaging. Expansion dynamics may complement conventional grading and strengthen explainable decision support for embryo selection aimed at improving live birth outcomes. Trial registration number No

Zitieren

APA

Misaghi, H., Cree, L., Padmanabhan, V., Storr, A., & Knowlton, N. (2026). L26/P-283 Predicting pregnancy and live-birth using embryo segmentation dynamics across multi- centre IVF lab datasets. Human Reproduction, 41(Supplement_1). https://doi.org/10.1093/humrep/deag083.619

Vancouver

Misaghi H, Cree L, Padmanabhan V, Storr A, Knowlton N. L26/P-283 Predicting pregnancy and live-birth using embryo segmentation dynamics across multi- centre IVF lab datasets. Human Reproduction. 2026;41(Supplement_1). doi:10.1093/humrep/deag083.619

BibTeX
@article{Misaghi2026,
  author = {Misaghi, H. and Cree, L. and Padmanabhan, V. and Storr, A. and Knowlton, N.},
  title = {L26/P-283 Predicting pregnancy and live-birth using embryo segmentation dynamics across multi- centre IVF lab datasets},
  journal = {Human Reproduction},
  year = {2026},
  volume = {41},
  number = {Supplement_1},
  doi = {10.1093/humrep/deag083.619}
}
Kostenlos starten →