Retrospective analysis of predicted pregnancy rate improvements using an AI model for embryo ranking.

Introduction: Embryo evaluation and selection is one of the most important steps of an in vitro fertilization (IVF) procedure. In recent years, artificial intelligence (AI) models have been developed to automate embryo analysis and reduce the subjectivity of manual grading. A key question of these models is what improvement they can offer in terms of pregnancy rates. In this study, we evaluate a previously developed model on a new test dataset using a novel bootstrapped analysis of virtual patient pregnancy rates.

Methods: Historical, de-identified images of blastocyst-stage embryos were collected from a single IVF center in the United States for cycles started between 2015-2020. Images were captured on day 5, 6, or 7 using the inverted microscope prior to biopsy or freeze. A total of 229 images and their corresponding manual grades were collected and matched to fetal heartbeat outcomes. The embryos were sorted by age, PGT status, and race, to create 16 distinct patient categories. Virtual patient panels were created within each category using a random selection of 3-5 embryos. For each virtual patient, the top embryo was selected using the manual grading system as well as the AI model. The average pregnancy rates of the top-ranked embryos were compared between manual grading and the AI model. The analysis was then repeated 1000 times to generate a distribution of results.

Results: On average, 422 virtual patient panels were constructed from the 229 embryos. The average pregnancy rate of the top-ranked embryo using manual grading was 55.2%, and the average pregnancy rate of the top-ranked embryo using the AI model was 62.7%. The average improvement from using the AI model was 7.5% with a standard deviation of 2.0% measured across the 1000 simulations.

Conclusions: This study demonstrates the potential of using an AI model in terms of improved pregnancy rates. Results from this retrospective analysis will help inform the design of future clinical validation studies.