Reassessment of pelvic radiographic measurements for delivery prediction using machine learning

Background and Objective: Pelvimetry has historically shown limitations in diagnosing cephalopelvic disproportion, yet recent evidence suggests potential predictive value. This study uses artificial intelligence to reassess pelvimetry's utility in predicting cesarean section. Methods: This single-center, retrospective case-control study included pregnant women at 37 weeks 0 days and 41 weeks 6 days of gestation, who underwent pelvic radiography for suspected cephalopelvic disproportion from January 2015 to August 2023. Pelvic radiographic measurements were obtained using the Guthmann-Sussmann method. Maternal characteristics, ultrasound examination data, and pelvimetric measurements were extracted from electronic medical records as potential predictors of delivery outcomes. In this study, the input data were analyzed using four machine learning models: Light Gradient Boosting Machine, Random Forest, Extreme Gradient Boosting, and Category Boosting. The primary outcome was the hierarchical importance of pelvic measurements in the predictive models. Results: Analysis included 355 participants. The strongest predictors were the differences between (1) the obstetric conjugate and biparietal diameter and (2) the interspinous diameter and biparietal diameter. The receiver operating characteristic curve for each model was Light Gradient Boosting Machine 0.74, Random Forest 0.85, Extreme Gradient Boosting 0.83, and Category Boosting 0.82. Conclusions: We developed high-performance machine learning models demonstrating that pelvimetric measurements— particularly, the differences between the obstetric conjugate and biparietal diameter, and between the interspinous diameter and biparietal diameter —combined with maternal and ultrasound factors, are strong predictors of cesarean section. The model’s ability to capture nonlinear associations may enhance predictive accuracy, and reassessing pelvimetric values could support delivery planning in clinical settings.