Impact of artificial intelligence in embryo selection to improve pregnancy rates per transfer in IVF cycles: a systematic review and meta-analysis.

BACKGROUND

Embryo selection is one of the most critical determinants of success in in vitro fertilization (IVF). Conventional approaches based on static morphological grading and time-lapse imaging (TLI) remain subjective and show only modest predictive accuracy. Artificial intelligence (AI) tools, including deep learning-based platforms such as iDAScore and KIDScore, have emerged as potentially more standardized alternatives, yet their true clinical impact on per-transfer outcomes remains uncertain.

OBJECTIVES

To determine whether AI-assisted embryo selection offers measurable clinical advantages over conventional morphology-based methods in IVF.

SEARCH STRATEGY: A systematic review and meta-analysis was conducted by searching four electronic databases (PubMed, Scopus, Embase, and the Cochrane Library) from January 2013 to April 2024.

SELECTION CRITERIA: Randomized controlled trials (RCTs) and observational cohort studies evaluating AI-assisted embryo selection versus conventional morphology in IVF cycles without preimplantation genetic testing for aneuploidy (PGT-A), reporting at least one of clinical pregnancy rate (CPR), implantation rate (IR), or live birth rate (LBR).

DATA COLLECTION AND ANALYSIS: Two reviewers independently screened records and extracted data. Study quality was assessed using the Cochrane Risk of Bias 2 tool for the RCT and the Newcastle-Ottawa Scale for observational studies. Where arm-level data were available, risk ratios (RR) with 95% confidence intervals (CI) were pooled using a random-effects model.

RESULTS

Ten studies were included. Only two reported arm-level outcomes suitable for meta-analysis. The pooled estimate showed no significant difference in clinical pregnancy rate between AI-assisted and morphology-based selection (RR = 0.99, 95% CI 0.87-1.12; I2 = 26.5%), with no statistically significant difference between the two methods. The remaining eight studies reported predictive performance metrics with area under the receiver operating characteristic curve values of 0.85-0.96, though these did not consistently translate into improved per-cycle outcomes. Implantation and live birth rates were inconsistently reported and could not be pooled.

CONCLUSION

AI-based embryo scoring showed no statistically significant difference compared with conventional morphology for per-transfer clinical pregnancy rate, with potential advantages in standardization and inter-observer consistency. However, superiority remains unconfirmed and key patient-centered outcomes including cumulative live birth rate, miscarriage rate, and time-to-pregnancy were not reported in available studies. Larger multicenter RCTs with standardized outcome reporting are required.