L26/P-166 AI-based automation of Gardner blastocyst grading to improve standardisation across IVF centres
Human Reproduction · 01.07.2026 · 41 (Supplement_1)
DOI: 10.1093/humrep/deag083.502 ↗Kurz gefasst
Fachartikel in der Zeitschrift Human Reproduction, online erschienen am 08.07.2026. Verfasst von P. Pawlik und 8 weiteren Autor:innen.
Abstract (Originalsprache)
Study question Can an artificial intelligence model reliably automate Gardner blastocyst grading across multiple international IVF centres with consistent performance? Summary answer The AI model achieved high discriminative performance and consistent accuracy across geographically diverse centres, supporting standardized automated Gardner grading. What is known already The Gardner blastocyst grading system is widely used for embryo assessment but is inherently subjective, leading to substantial inter- and intra-observer variability among embryologists. Differences in experience, workload, laboratory protocols, and imaging conditions further contribute to inconsistent grading across clinics. Previous artificial intelligence approaches have shown promise for embryo assessment; however, many studies have been limited by single-centre data, narrow population diversity, or focus on selected grading components rather than the full Gardner scale. Robust multicentre validation of AI-based automated grading systems remains limited. Study design, size, duration This retrospective multicentre validation study included 6,299 blastocyst images collected from four international IVF centres. A deep learning model was developed and evaluated using five-fold cross-validation to predict all Gardner scale components simultaneously. Participants/materials, setting, methods Blastocyst images were obtained from four IVF centres in Europe and Latin America. Expert embryologists independently annotated expansion, inner cell mass (ICM), and trophectoderm (TE) grades. A deep learning model was trained to predict combined Gardner grades. Performance was assessed using multiclass area under the receiver operating characteristic curve (AUROC) and top-5 accuracy with 95% confidence intervals (CI). Main results and the role of chance The AI model demonstrated strong overall discriminative performance, achieving a multiclass AUROC of 0.913 (95% CI: 0.902–0.919). Among individual components, performance was highest for expansion grading (AUROC 0.948; 95% CI: 0.943–0.952), followed by trophectoderm grading (AUROC 0.866; 95% CI: 0.858–0.872) and inner cell mass grading (AUROC 0.842; 95% CI: 0.834–0.849). Across all centres, the model achieved an overall top-5 accuracy of 83.7% (95% CI: 82.8–84.6%), indicating that the correct combined Gardner grade was among the five most probable predictions out of 54 possible combinations. Performance remained consistent across centres despite geographic, equipment, and protocol differences, with top-5 accuracy ranging from 78.4% to 85.0%. Narrow confidence intervals across metrics suggest that observed performance was unlikely due to chance and reflects robust generalisability. Limitations, reasons for caution The study was retrospective and relied on expert annotations as reference standards. Clinical outcome associations were not assessed, and performance may vary in settings with substantially different imaging quality or laboratory practices. Wider implications of the findings AI-driven automated Gardner grading may support standardized, objective embryo assessment, reduce observer-related variability, and enable quality-controlled workflows, particularly in settings with limited access to highly experienced embryologists. Trial registration number No
Zitieren
APA
Pawlik, P., Siennicki, M., Kuśmierczyk-Kubiak, J., Rodriguez, A. V. P., Wojtasik, B., Bălănescu, M., Andrade, G. M., Sankowska, U., & Wygocki, P. (2026). L26/P-166 AI-based automation of Gardner blastocyst grading to improve standardisation across IVF centres. Human Reproduction, 41(Supplement_1). https://doi.org/10.1093/humrep/deag083.502
Vancouver
Pawlik P, Siennicki M, Kuśmierczyk-Kubiak J, Rodriguez AVP, Wojtasik B, Bălănescu M, et al. L26/P-166 AI-based automation of Gardner blastocyst grading to improve standardisation across IVF centres. Human Reproduction. 2026;41(Supplement_1). doi:10.1093/humrep/deag083.502
BibTeX
@article{Pawlik2026,
author = {Pawlik, P. and Siennicki, M. and Kuśmierczyk-Kubiak, J. and Rodriguez, A. Vidal Pascual and Wojtasik, B. and Bălănescu, M. and Andrade, G. Mamede and Sankowska, U. and Wygocki, P.},
title = {L26/P-166 AI-based automation of Gardner blastocyst grading to improve standardisation across IVF centres},
journal = {Human Reproduction},
year = {2026},
volume = {41},
number = {Supplement_1},
doi = {10.1093/humrep/deag083.502}
}
EUVIMED