Prediction of reproductive outcomes in unicornuate uterus women based on Three-dimensional MRI radiomic features

Purpose: To develop and validate a radiomics nomogram integrating three-dimensional (3D) MRI features for predicting ≥35 weeks' gestation in women with unicornuate uterus. Methods: This retrospective study enrolled 170 unicornuate uterus patients who underwent pelvic 3D MRI between 2005-2022. Patients were categorized into four groups based on reproductive outcomes: primary infertility, <24 weeks, 24-35 weeks, and ≥35 weeks gestation. The uterus was segmented and reconstructed in 3D using ITK-SNAP. Radiomics features (n=1,834), including volume, surface area, sphericity, and diameters, were extracted. Feature selection employed LASSO regression, and eight machine learning algorithms were compared for model construction. A radiomics nomogram integrating selected features with clinical variables was developed. Predictive performance was assessed using ROC and decision curve analyses. Results: Uterine volume was significantly larger in the ≥35 weeks gestation group compared to infertility and <24 weeks groups (P<0.05). Surface area was also greater in the ≥35 weeks group versus the infertility group (P<0.05). No significant differences in axis lengths were observed among groups. The radiomics nomogram demonstrated robust discrimination for predicting ≥35 weeks gestation, achieving AUCs of 0.86 (95% CI: 0.79-0.92) in the training cohort and 0.84 (95% CI: 0.69-0.98) in the validation cohort. Decision curve analysis confirmed favorable clinical utility. Conclusion: Uterine volume measured by 3D MRI reconstruction serves as a reliable prognostic factor for predicting term delivery in unicornuate uterus patients. The developed radiomics nomogram integrating radiomics signatures with clinical indicators enables individualized prediction of reproductive outcomes.