Objective This study investigates whether a model constructed based on clinical-MRI morphological indica-tors can provide a robust prediction for adverse clinical outcomes of placenta accreta spectrum disorders(placenta accreta spectrum,PAS).Methods Retrospective analysis was conducted on data from 125 pregnant women with placenta accreta spectrum disorders(PAS)from two medical centers.The data were divided into a training set(60 cases from center 1),a validation set(25 cases from center 1),and an external testing set(40 cases from center 2).Adverse clinical outcomes of PAS were defined as intraoperative bleeding exceeding 1000 milliliters and/or hysterectomy.Five MRI morphological indica-tors(cervical length,lower uterine segment bulging ratio,area of T2(T2 weighted image,T2 WI)low signal band in the placenta,diameter and area of thickened blood vessels in the placenta)and six clinical high-risk factors of PAS(age,ges-tational week,parity,number of previous cesarean sections,and number of previous curettages)were extracted as model inputs.A predictive model for adverse clinical outcomes of PAS(low risk vs.high risk)was constructed using XGBoost as the algorithm based on these inputs.Furthermore,we employed a five-fold cross-validation and a multi-center external tes-ting strategy to verify the stability and generalization performance of the model.Finally,we also calculated the SHAP value of each individual feature to explore their contributions to the model decisions.Results Regarding the predictive model for adverse clinical outcomes of PAS(low risk vs.high risk),the AUROC values obtained from the five-fold cross-valida-tion were 0.94,0.86,0.87,0.93,and 0.92,and the corresponding accuracies were 0.92,0.81,0.81,0.85,and 0.85,respectively.The optimal model achieved an AUROC of 0.94 on the validation set and 0.85 on the testing set,both falling within the 95％confidence interval.The corresponding accuracies were 0.92 and 0.78,the sensitivities were 0.93 and 0.80,and the specificities were 0.92 and 0.80,respectively.Conclusion This study demonstrates that a predictive model for adverse clinical outcomes of PAS in pregnant women,constructed based on clinical-MRI morphological indices,can provide a strong basis for preoperative risk assessment of adverse clinical outcomes of PAS.
NETL
NSTL
万方数据
