{"paper_id":"8d73de94-b1a3-4a26-9b62-e74ddfc92e9d","body_text":"OP11.06: The predictive value of MUSA features of adenomyosis on live birth is poor, using a machine learning algorithm\nObjectives\nTo determine the predictive value of Morphological Uterus Sonographic Assessment (MUSA) group features of adenomyosis on cumulative live birth (CLB) after the first in vitro fertilization (IVF) or intracytoplasmic sperm injection (ICSI) treatment, using a machine learning algorithm.\nMethods\nWe included 1037 women aged 25 - ≤ 39 years, undergoing their first IVF/ICSI treatment between January 2019 and October 2022. The presence of MUSA features was confirmed by ultrasonography, using GE Voluson 10 Expert ultrasound equipment with a 5-9 MHz transvaginal transducer. To assess the prognostic capability of clinical variables and MUSA features on live birth, an Extreme Gradient Boosting (XGBoost) algorithm was used to develop a prediction model, presented with the area under receiver operating characteristics (ROC) curve. Clinical variables included in the model were age, BMI, ovarian reserve parameters, presence of endometriosis, and subjective symptoms such as dysmenorrhea, dyspareunia, pelvic pain, dyschezia, dysuria, hematuria, and hematochezia. The importance of each variable on the model was illustrated with the Shapley additive explanations algorithm (SHAP) variable importance.\nResults\nThe XGBoost model had an area under the ROC curve of 0.69. The variables with the highest mean of absolute SHAP values were serum antimüllerian hormone (s-AMH) (0.21) and a regular junctional zone (JZ) (0.13). The location of MUSA features (inner myometrium) had a mean SHAP value of 0.077.\nConclusions\nA regular JZ was the best ultrasonographic variable in predicting live birth, whereas s-AMH was the best clinical variable, confirmed by the XGboost model. The predictive ability of MUSA features in relation to live birth after IVF/ICSI treatment was poor. Considering that many factors may interact to impact fertility treatment outcomes, additional variables should be included in a clinically useful prediction model for IVF/ICSI treatment outcome.","source_license":"CC0","license_restricted":false}