Embryo classification beyond pregnancy: Early prediction of first trimester miscarriage using machine learning

Abstract

ABSTRACTObjectiveDevelop a machine learning classifier for predicting the risk of cleavage-stage embryos to undergo first trimester miscarriage based on time-lapse images of preimplantation development.DesignRetrospective study of a 4-year multi-center cohort of women undergoing intra-cytoplasmatic sperm injection (ICSI). The study included embryos with positive indication of clinical implantation based on gestational sac visualization either with first trimester miscarriage or live birth outcome. Miscarriage was determined based on negative fetal heartbeat indication during the first trimester.SettingData were recorded and obtained in hospital setting and research was performed in university setting.Patient(s)Data from 391 women who underwent fresh single or double embryo transfers were included.Intervention(s)None.Main Outcome Measure(s)A minimal subset of six non-redundant morphodynamic features were screened that maintain high prediction capacity. Using this feature subset, XGBoost and Random Forest models were trained following a 100-fold Monte-Carlo cross validation scheme. Feature importance was scored using the SHapley Additive exPlanations (SHAP) methodology. Miscarriage versus live-birth outcome prediction was evaluated using a non-contaminated balanced test set and quantified in terms of the area under the receiver operating characteristic (ROC) curve (AUC), precision-recall curve, positive predictive value (PPV), and confusion matrices.Result(s)Features that account for the distribution of the nucleolus precursor bodies within the small pronucleus and pronuclei dynamics were highly predictive of miscarriage outcome. AUC for miscarriage prediction of validation and test set embryos using both models was 0.68-to-0.69. Clinical utility was tested by setting two classification thresholds accounting for high sensitivity 0.73 with 0.6 specificity and high specificity 0.93 with 0.33 sensitivity.Conclusion(s)We report the development of a decision-support tool for identifying the embryos with high risk of miscarriage. Prioritizing embryos for transfer based on their predicted risk of miscarriage in combination with their predicted implantation potential will improve live-birth rates and shorten time-to-pregnancy.CapsuleThe risk of first trimester miscarriage of cleavage stage embryos is predicted with AUC 68% by screening a minimal subset of six non-redundant morpho-dynamic features and training a machine-learning classifier.