How Often Does Artificial Intelligence System (Ivy) Select the Same Embryo for Transfer Compared to Selection Based on Traditional Blastocyst Morphology or Time-Lapse Algorithm (KIDScore™D5)?

Abstract

Background: When patients have multiple blastocysts following IVF/ICSI, selection of the most viable blastocyst for transfer can reduce time taken for patients to conceive. An artificial intelligence system, Ivy, has been developed to predict pregnancy from embryo images. Aim: To determine the frequency that Ivy selects the same embryo for transfer as morphological assessment or time-lapse algorithm (KIDScore™D5v2). Method: Data was collected from 100 patients that had [Formula: see text] 2 blastocysts for day 5 fresh single embryo transfer (total n=1319 embryos, cultured in EmbryoScope+). The blastocyst transferred was selected by clinic protocol: a combination of morphology and morphokinetics. Ivy analysed time-lapse images to generate an IVYScore. Four embryologists assessed morphology of one image per embryo using different focal planes. Morphokinetics parameters were annotated to calculate KIDScore. Results: Embryologists chose the same blastocyst for transfer in 51 out of 100 patients, and only 35.3% of these embryos had the highest IVYscore compared to sibling blastocysts. Compared to KIDScore, Ivy would have chosen the same blastocyst in 61% of the cases. Agreement between all methods was lower when patients had morphologically similar blastocysts (18.2% vs. 50.0%, p[Formula: see text]0.001). Clinical pregnancy rate was 48.0% (n=100), which was similar for blastocysts selected by Ivy (50.0%, n=64), morphology (n=29, 41.4%) or KIDScore (n=74, 45.9%). Further, 44 frozen blastocyst transfers resulted in 14 pregnancies. If IvyScore were to be used instead of clinic protocol, 11 cases would have resulted in an equal or shorter time to pregnancy. Conclusion: There is limited agreement between methods when patients have multiple blastocysts available for transfer, especially when morphology is similar. Pregnancy rates of Ivy-selected blastocysts were similar to other methods, indicating that Ivy could be used with comparable success, and could even lead to reduction in time to pregnancy. This study is limited by its retrospective nature, and unknown pregnancy potential of blastocysts that were not transferred.