EUVIMEDEUVIMED entdecken
Crossref

L26/P-148 General versus oocyte-specific artificial intelligence for predicting blastocyst formation and embryo euploidy: a comparative diagnostic accuracy study

M. Montiveros, M. Schapira, F. Di Biase, J. Pennino Martinez, S. Papier, D. Glujovsky

Human Reproduction · 01.07.2026 · 41 (Supplement_1)

DOI: 10.1093/humrep/deag083.484 ↗

Kurz gefasst

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

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Study question How does an oocyte-specific artificial intelligence (IA) model compare with a general generative AI model in predicting blastocyst yield and embryo euploidy across IVF cycles? Summary answer A general generative AI showed predictive accuracy comparable to an oocyte-specific AI for blastocyst yield and euploidy, while exhibiting distinct and clinically relevant misclassification patterns. What is known already Artificial intelligence is increasingly used in assisted reproduction to support embryo selection and reduce subjectivity. Most available tools are domain-specific algorithms trained on oocyte or embryo images and validated within restricted clinical settings. These systems are generally assumed to outperform non-specialised approaches. However, little is known about how their predictive performance compares with that of general-purpose generative AI models when both are provided with identical clinical information. Direct comparative evidence evaluating oocyte-specific AI against multimodal generative AI for predicting blastocyst formation and embryo euploidy remains scarce. Study design, size, duration This retrospective diagnostic accuracy study included 96 IVF/PGT-A cycles performed between January and June 2025 at a single private clinic. Predictions of blastocyst yield and euploid embryo number were generated at the cycle level by two AI approaches: Magenta, an oocyte-specific AI, and ChatGPT, a general generative AI. Predicted outcomes were compared with observed blastocyst development and PGT-A results. Participants/materials, setting, methods Cycles with autologous oocytes, complete embryological outcomes and PGT-A results were included. ChatGPT received structured clinical data and oocyte images, mirroring the information provided to embryologists in the PREDO study. Magenta analysed oocyte images using its proprietary algorithm. Accuracy, agreement, misclassification profiles and age-stratified performance were assessed at the cycle level. Main results and the role of chance 96 IVF/PGT-A cycles were analysed. Predictive accuracy for euploid embryo number was comparable between the general generative AI and the oocyte-specific AI (77.1% [74/96] vs 69.8% [67/96],p=0.25), with an overall agreement of 69.8% (p = 0.01). Despite similar overall accuracy, the two models demonstrated distinct misclassification patterns. Among cycles correctly predicted by the general generative AI, the oocyte-specific AI misclassified 24.3% (18/74). Conversely, among cycles incorrectly predicted by the general generative AI, the oocyte-specific AI correctly classified 50% (11/22). Overestimation of euploid embryo number was less frequent with the general generative AI (2.1%) than with the oocyte-specific AI(19.0%), whereas underestimation occurred more frequently with the general generative AI (20.8% vs 10.5%). Mean maternal age was 38.4 ± 3.5 years, with 90% of patients having undergone two or fewer previous IVF cycles. The median number of retrieved MII oocytes was 4 (IQR 2–8), resulting in a median of 2 blastocysts (IQR 1–4) and 0 euploid embryos (IQR 0–2). The oocyte-specific AI classified oocyte quality as medium or high in 52% of cycles. Prediction accuracy was higher in women older than 38 years for both models, while prediction errors in younger women were mainly driven by overestimation. Age-stratified analyses should be interpreted as exploratory. Limitations, reasons for caution The relatively small sample size limited statistical power to detect modest differences between models. The retrospective, single-centre design and use of static oocyte images may further limit generalisability. Accordingly, findings should be interpreted as exploratory and require validation in larger, prospective multicentre studies. Wider implications of the findings These findings suggest that general generative AI may achieve predictive performance comparable to oocyte-specific models in IVF. This challenges current assumptions about the need for highly specialised AI and highlights the importance of further validation before clinical integration. Trial registration number No

Zitieren

APA

Montiveros, M., Schapira, M., Biase, F. D., Martinez, J. P., Papier, S., & Glujovsky, D. (2026). L26/P-148 General versus oocyte-specific artificial intelligence for predicting blastocyst formation and embryo euploidy: a comparative diagnostic accuracy study. Human Reproduction, 41(Supplement_1). https://doi.org/10.1093/humrep/deag083.484

Vancouver

Montiveros M, Schapira M, Biase FD, Martinez JP, Papier S, Glujovsky D. L26/P-148 General versus oocyte-specific artificial intelligence for predicting blastocyst formation and embryo euploidy: a comparative diagnostic accuracy study. Human Reproduction. 2026;41(Supplement_1). doi:10.1093/humrep/deag083.484

BibTeX
@article{Montiveros2026,
  author = {Montiveros, M. and Schapira, M. and Biase, F. Di and Martinez, J. Pennino and Papier, S. and Glujovsky, D.},
  title = {L26/P-148 General versus oocyte-specific artificial intelligence for predicting blastocyst formation and embryo euploidy: a comparative diagnostic accuracy study},
  journal = {Human Reproduction},
  year = {2026},
  volume = {41},
  number = {Supplement_1},
  doi = {10.1093/humrep/deag083.484}
}
Kostenlos starten →