Evaluation of Artificial Intelligence for Embryo Selection During IVF: A Prospective, Randomized Controlled Trial

Objective

To determine whether embryo selection using artificial intelligence (AI) is non-inferior to selection based solely on traditional morphological grading in terms of clinical pregnancy rate, through a prospective randomized control trial (RCT).

Materials and Methods

This RCT (NCT05483985) was conducted at seven sites in the United States. Eligible participants were women 21-43 years old undergoing autologous IVF treatment who had at least 8 eggs after retrieval. Patients were randomized 1:1 into two study arms: (1) the Traditional Morphology (TM) arm, where embryos were selected based solely on morphology grading, and (2) the AI arm, where embryos were selected using an automated AI score. The AI tool1, developed by Alife Health, analyzes images of day 5-7 blastocysts to generate a score from 0 to 1. In the AI arm, embryologists could override the AI-based selection if clinically justified. The primary endpoint was clinical pregnancy, defined as a fetal heartbeat at 6-8 weeks. The secondary endpoint was the incidence of AI score overrides.

Results

A total of 444 patients were enrolled, with 442 randomized between June 2022 - October 2024. The intention-to-treat (ITT) group (n=359) included all patients who completed an embryo transfer. The modified intention-to-treat (mITT) group (n=283) was limited to patients with multiple embryos available for selection. The randomized adherence to protocol (RAP) group (n=248) excluded patients in the AI arm where the embryo selected was not the highest AI-ranked, which occurred in 25/133 cases (18%). There were no differences in age, BMI, race, ethnicity, baseline hormones, protocols, laboratory outcomes, transfer type, or PGT utilization. In the mITT group, clinical pregnancy occurred in 102/150 (68%) participants in the TM arm and 97/133 (72.9%) participants in the AI arm, a 4.9% difference. The pre-specified 10% non-inferiority margin was met across all analysis groups (Table 1).

Conclusions

Embryo selection with AI support was non-inferior to traditional morphology-based selection and showed no safety concerns. Although the trial was not powered to detect superiority, the AI arm showed higher clinical pregnancy rates across all analysis groups.

Impact Statement

This is the first prospective RCT to demonstrate that AI-assisted embryo selection is non-inferior to traditional morphological grading, supporting its safe and effective integration into clinical practice.