Related Experiment Video
Updated: Sep 17, 2026
Establishment of an Embryo Implantation Model In Vitro
Published on: June 21, 2024
Clinical evaluation and predictive model development of AI-based non-invasive embryo selection in in vitro
Pengfei Zhu1, Xingyu Bi1, Dan Su1
1Center of Reproductive Medicine, Shanxi Children's Hospital (Shanxi Women and Children Health Hospital), Taiyuan, Shanxi, China.
Background:
Embryo selection remains one of the most critical determinants of IVF success. Conventional embryo grading relies primarily on morphological assessment and is subject to inter-observer variability and limited predictive accuracy. Recent advances in AI have enabled non-invasive systems to analyze embryological and morphokinetic features, improving embryo selection and reproductive outcomes.
Objective:
To evaluate the clinical effectiveness of an AI-based non-invasive embryo selection system and develop predictive models for IVF outcomes among patients undergoing assisted reproductive treatment.
Methods:
A retrospective cohort study was conducted at a tertiary fertility center involving 300 IVF cycles from January 2020 to December 2024. Clinical, demographic, embryological, and AI-derived data were extracted from electronic medical records and laboratory databases. Primary outcome was clinical pregnancy; secondary outcomes included implantation, ongoing pregnancy, live birth, and model performance. Machine learning algorithms (logistic regression, random forest, SVM, gradient boosting, and neural networks) were evaluated using accuracy, sensitivity, specificity, precision, F1-score, and AUC-ROC.
Results:
A total of 300 IVF cycles and 742 embryos were analyzed. Implantation, clinical pregnancy, ongoing pregnancy, and live birth rates were 60.7%, 56.0%, 51.3%, and 48.7%, respectively. Higher AI-derived embryo scores were significantly associated with improved outcomes. Clinical pregnancy rates increased progressively across AI score categories, from 18.5% (scores <55) to 74.0% (scores ≥85, p < 0.001). Multivariable logistic regression identified maternal age (AOR = 0.92; 95% CI: 0.88-0.97), AMH (AOR = 1.18; 95% CI: 1.04-1.34), high-quality blastocyst (AOR = 2.41; 95% CI: 1.54-3.78), and AI score ≥85 (AOR = 3.67; 95% CI: 2.12-6.35) as independent predictors of clinical pregnancy. The developed AI model demonstrated excellent performance, with accuracy 84.3%, sensitivity 82.1%, specificity 79.8%, and AUC-ROC 0.89.
Conclusion:
AI-based non-invasive embryo selection showed significant predictive value for reproductive outcomes and outperformed conventional assessment in identifying embryos with high implantation potential. AI-derived embryo scores were independently associated with clinical pregnancy success, and the model exhibited excellent discrimination and calibration. These findings support integrating AI-assisted embryo assessment into contemporary IVF practice and highlight its potential to enhance personalized reproductive care.
