Non-Invasive Prediction of Embryo Ploidy from Time-Lapse Videos Using Video Vision Transformers (ViViT)

Accurate selection of viable embryos is crucial in in vitro fertilisation (IVF) to improve clinical outcomes. Traditional embryo assessment relies on subjective visual evaluation or invasive genetic testing, each with inherent limitations. To address these limitations, we propose a non-invasive approach employing time-lapse videos and the Video Vision Transformer (ViViT) architecture to predict embryo ploidy. We analysed 2,388 embryo videos derived from non-donor IVF cycles. The ViViT-based model achieved a validation ROC-AUC of 0.85, demonstrating its ability to discriminate between eligible (euploid, mosaic and carrier) and ineligible (aneuploid) embryos. Our findings suggest that transformer-based video models can capture dynamic developmental patterns, offering a scalable, automated, and non-invasive decision-support tool for embryo selection in IVF.