A Predictive Imaging Model for Cervical PAS in Complete Placenta Previa Using Combined Ultrasound and MRI

To develop and validate a combined ultrasound-MRI predictive imaging model for diagnosing cervical placenta accreta spectrum (PAS) in women with complete placenta previa.

This retrospective study included a total of 350 patients with complete placenta previa who underwent comprehensive prenatal ultrasound and MRI examinations; independent imaging predictors were identified using multivariate logistic regression; Two internal validation methods were employed: a 7:3 cohort split and 1000 bootstrap resamples. The model's diagnostic performance was comprehensively evaluated using a combination of ROC curves, calibration curves, and decision curve analysis (DCA).

Five independent risk factors were identified: abundant cervical blood flow, cervical sinus, lacunae over the cervix, cervical length <3.0 cm, and placental heterogeneity. The combined model yielded an AUC of 0.892, sensitivity of 87.603%, and specificity of 81.489%. The AUC on the internal validation set reached 0.889; the model demonstrated good discriminatory power and calibration, and showed considerable clinical net benefit across a wide range of risk thresholds. Patients with cervical PAS had significantly higher rates of adverse maternal-fetal outcomes.

A predictive model combining five key ultrasound and MRI features shows high diagnostic performance for cervical PAS in complete placenta previa, supporting preoperative risk stratification and surgical planning. This study is a single-center retrospective analysis; it is subject to selection bias and does not include a multicenter external cohort to support its generalizability. Further validation of the model's clinical utility requires large-sample, multicenter prospective cohort studies.