EUVIMEDEUVIMED entdecken
Crossref

L26/P-228 AI-Powered Embryo Selection And Retrospective Study, Evaluation Of AI In Embryo By Comparing Implantation Of Single Embryo Transfer Of Patient With Confirmed Positive Pregnancy Outcomes

C. N, M. Sourirajan, R. Segu Mohamed, A. Ekambarakrishnan, P. Senthil

Human Reproduction · 01.07.2026 · 41 (Supplement_1)

DOI: 10.1093/humrep/deag083.564 ↗

Kurz gefasst

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

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Study question The primary objective of this research is to evaluate the dependability of current AI capabilities and identify existing flaws within these systems. Summary answer AI and embryologists showed 77% concordance; however, the 23% discordance risks discarding embryos potentially capable of achieving a successful pregnancy. What is known already Embryo selection is a critical factor in the success of assisted reproductive technology. While traditional methods rely on subjective morphological assessments by embryologists, artificial intelligence (AI) has emerged as an objective decision-support tool. Utilizing deep learning and neural networks, these AI models show promise in predicting implantation and pregnancy outcomes. However, the real-world concordance between AI-derived scores and manual embryologist grading remains insufficiently explored, particularly in cycles involving single embryo transfers with confirmed positive outcomes. Further research is required to evaluate the integration of these technologies into routine clinical practice. Study design, size, duration Trial study started from 2023 to 2024 This was a retrospective pilot concordance study conducted at a multiple ARC fertility center. 120 embryos transferred as single embryo transfer cycles with confirmed positive pregnancy outcomes were included. Embryo images were retrospectively analyzed using an AI-based embryo assessment system after clinical outcomes were known to find out the accuracy of AI scoring for embryo Selection Participants/materials, setting, methods The study included 120 patients who underwent single embryo transfer at ARC Fertility Hospital. Embryos were graded morphologically as per Gardners Grading system by experienced embryologists prior to transfer. Corresponding embryo images were subsequently analyzed using AI software. AI scores were categorized as poor (<5), moderate (5–6.9), or good (≥7) and compared with embryologist-derived quality categories. Concordance was assessed using categorical agreement analysis. Main results and the role of chance Total 120 patients and 120 embryos was conducted to evaluate the concordance between AI-based assessment and manual embryologist grading. AI system categorized embryos as follows: 7 patients with a score of < 5, 13 patients with a score between 5 and 7, and the remaining patients with a score between 7 and 10. This yielded a concordance rate of 76.66%. The remaining 23.33% of embryos were classified by the AI as poor quality. In comparison, embryologists utilized the Gardner Grading System to categorize embryos based on expansion (grades 3 to 5): AA, AB, and BA were classified as “Good”; BB as “Moderate”; and BC as “Poor.” The study focused on patients who underwent single embryo transfers and had already achieved confirmed positive pregnancies. An analysis of the 23.33% discrepancy revealed that the AI predictive model categorized embryos as low-grade despite them resulting in successful pregnancies. Further investigation indicates that static image AI models may lack dependability and accuracy because embryo grading varies significantly based on the viewing angle. Due to the 3D structure of embryos, static images cannot consistently focus on the depth of the Inner Cell Mass (ICM) and Trophectoderm. Consequently, the AI misinterprets these structures, leading to inaccurate scoring. Limitations, reasons for caution The study is limited by selection bias, as the cohort included only confirmed pregnancies, precluding specificity analysis. Additionally, the reliance on static 2D AI imaging failed to account for volumetric depth (ICM/Trophectoderm) and angular variations. This led to 23.33% discordance rate, where the AI incorrectly categorized viable embryos as low-grade.” Wider implications of the findings The study demonstrates that static 2D AI risks discarding viable embryos, evidenced by a 23.33% false-negative rate. This finding necessitates a shift toward dynamic 3D imaging and maintains the requirement for human oversight, proving static models are currently insufficient for autonomous clinical decision-making. Trial registration number No

Zitieren

APA

N, C., Sourirajan, M., Mohamed, R. S., Ekambarakrishnan, A., & Senthil, P. (2026). L26/P-228 AI-Powered Embryo Selection And Retrospective Study, Evaluation Of AI In Embryo By Comparing Implantation Of Single Embryo Transfer Of Patient With Confirmed Positive Pregnancy Outcomes. Human Reproduction, 41(Supplement_1). https://doi.org/10.1093/humrep/deag083.564

Vancouver

N C, Sourirajan M, Mohamed RS, Ekambarakrishnan A, Senthil P. L26/P-228 AI-Powered Embryo Selection And Retrospective Study, Evaluation Of AI In Embryo By Comparing Implantation Of Single Embryo Transfer Of Patient With Confirmed Positive Pregnancy Outcomes. Human Reproduction. 2026;41(Supplement_1). doi:10.1093/humrep/deag083.564

BibTeX
@article{N2026,
  author = {N, C. and Sourirajan, M. and Mohamed, R. Segu and Ekambarakrishnan, A. and Senthil, P.},
  title = {L26/P-228 AI-Powered Embryo Selection And Retrospective Study, Evaluation Of AI In Embryo By Comparing Implantation Of Single Embryo Transfer Of Patient With Confirmed Positive Pregnancy Outcomes},
  journal = {Human Reproduction},
  year = {2026},
  volume = {41},
  number = {Supplement_1},
  doi = {10.1093/humrep/deag083.564}
}
Kostenlos starten →