OBJECTIVE:

To compare the predictive performance of an AI-based embryo ranking model (Vita Embryo) with a Gardner grade-based model evaluated by lab directors (>20 years of experience) and to assess the consistency of the AI model’s performance across different IVF clinics.

MATERIALS AND METHODS:

This retrospective study analyzed 497 blastocyst transfer cycles from two IVF clinics in Korea: Clinic A (n=101; March 2023 May 2024) and Clinic B (n=396; January September 2024). Both fresh and frozen-thawed embryo transfers (ET) were included, limited to cases with available fetal heart tone (FHT) outcomes. Blastocysts at stage 1 and stage 6 were excluded because there were fewer than five cases.

Two LightGBM-based predictive models were developed:

- Model 1: Gardner grading (blastocyst stage, ICM grade, TE grade) + maternal age.

- Model 2: Vita Embryo scores (blastocyst image + maternal age)

Vita Embryo is an AI-based software that predicts the likelihood of fetal heart tone (FHT) development based on day-5 blastocyst microscope static images. The dataset was randomly split 70:30 into training and testing sets, both overall and within each clinic. Model performance was assessed by AUROC, with statistical comparisons using the DeLong test.

RESULTS:

Across all datasets, model 2 (image with age) achieved higher AUROC values compared to model 1(Gardner grade with age), although differences were not statistically significant by DeLong test. Vita Embryo demonstrated more consistent predictive performance across clinics (0.65-0.69), while the Gardner grade-based model showed greater variability depending on the clinical environment (0.59-0.67)

CONCLUSIONS:

The AI-based Vita Embryo model demonstrated pregnancy prediction performance comparable to that of lab directors using Gardner grading, with greater consistency across clinics. This suggests its potential to reduce inter-clinic variability in embryo evaluation.

IMPACT STATEMENT:

Vita Embryo, an AI-based embryo evaluation software, predicts blastocyst developmental potential toward FHT based on day-5 microscope images. By providing standardized and reproducible predictive performance across clinical environments, Vita Embryo supports the broader adoption of AI technology to enhance objectivity and consistency in reproductive medicine.