Ultrasound texture-based radiomics of placental and myometrial tissue for predicting postpartum blood loss in women with placenta previa and low-lying placenta.

INTRODUCTION

This study aimed to evaluate the role of radiomic analysis applied to ultrasound images in predicting postpartum blood loss at birth in women affected by low-lying placenta or placenta previa.

MATERIAL AND

METHODS

In this retrospective, single-center study, we analyzed singleton pregnancies with placenta previa or a low-lying placenta, initially diagnosed at the second-trimester ultrasound examination. Data were collected from ultrasound examinations conducted in the second and third trimesters, along with birth outcomes. Radiomic analysis was conducted on archival ultrasound images to extract quantitative features. Predictive models were constructed utilizing multivariable generalized linear modeling (Gamma regression with a log link), encompassing radiomics-only, clinical/sonographic-only, and an integrated model.

RESULTS

In the final analysis of 107 women, 51 exhibited postpartum blood loss exceeding 500 mL. A prior cesarean delivery was recognized as a notable clinical risk factor. Multiple radiomic features identified in second- and third-trimester ultrasound scans correlated with a heightened risk of significant blood loss during birth. The integrated predictive model exhibited superior accuracy for blood loss exceeding 500 mL, achieving an AUC of 82.32% (95% CI: 74.18%-90.45%). This performance surpassed that of the clinical ultrasound model, which had an AUC of 71.27% (95% CI: 62.27%-80.27%), with a statistically significant difference (p = 0.001). Additionally, it demonstrated a nonsignificant improvement over the radiomics-only model, which recorded an AUC of 77.17% (95% CI: 68.25%-86.09%).

CONCLUSIONS

Radiomic analysis of ultrasound images enhances risk prediction for postpartum major blood loss in pregnancies affected by placenta previa and low-lying placenta. Integrating radiomics with clinical and sonographic data improves predictive accuracy, offering a promising tool for personalized obstetric risk assessment and management.