Multimodal ensemble machine learning pipeline for oocyte fertilization prediction using first polar body morphology and clinical factors

While machine learning has seen growing applications in in-vitro fertilization (IVF), existing research primarily focuses on embryo-level data, which represents the developmental stage following the fertilization of the oocyte and sperm. However, the study of such application on oocyte, which is the major contributing factor to the success of IVF, is limited. This study introduces an ensemble-learning pipeline designed to predict successful fertilization—defined as two-pronuclei (2PN) development—following Intracytoplasmic Sperm Injection (ICSI) by integrating morphological image data with clinical features. Furthermore, this study examines how the first polar body (PBI) influences the established analytical pipeline. Our pipeline input consists of a complete metaphase of meiosis II (MII) oocyte image, its PBI segmentation, and numerical clinical cellular features. We evaluated two prominent machine learning models: Random Forest and Extreme Gradient Boosting (XGBoost), to predict outcomes based on numerical features. Our experimental results indicate that XGBoost demonstrates superior performance and is more suitable for this specific numerical dataset. Furthermore, our results demonstrate that integrating PBI imagery significantly enhances the pipeline’s predictive performance, achieving a peak AUC of 0.82. Subsequent analysis suggests that the morphological characteristics of the PBI provide a valuable predictive association within the ensemble framework.