End-to-End Prediction of Massive Hemorrhage During Dilatation and Curettage in Cesarean Scar Pregnancy Patients Using a 3D VGG Model

Motivation: Dilation and Curettage (D&C) is the standard treatment for cesarean scar pregnancy (CSP). However, existing quantitative approaches for preoperative assessment of intraoperative bleeding require manual feature selection, introducing uncertainty and hindering automation. Goal(s): To develop an end-to-end deep-learning model for automatic and preoperative prediction of massive hemorrhage risk. Approach: 109 CSP patients were included. A 3D VGG model was introduced based on high-resolution T2-weighted MR images. Results: The prediction model achieved promising performance, with AUC, accuracy, specificity, and sensitivity at 0.9375, 0.9167, 0.9583, and 0.9167 on the test dataset, respectively. Impact: This study presents an effective deep-learning solution for predicting hemorrhage in CSP patients, reducing manual intervention, and enhancing automation in clinical practice. The 3D VGG model's high accuracy and minimal preprocessing may streamline clinical decision-making and improve patient safety.