MRI-based deep learning-radiomics ensemble model for predicting postpartum hemorrhage in high-risk pregnancies(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2025年第11期
- Page:
- 1523-1531
- Research Field:
- 医学人工智能
- Publishing date:
Info
- Title:
- MRI-based deep learning-radiomics ensemble model for predicting postpartum hemorrhage in high-risk pregnancies
- Author(s):
- ZHANG Qi1; 2; WANG Haijie3; LIANG Xiaoyun3; ZHU Hao4; YANG Guang1; 2
- 1. Shanghai Key Laboratory of Magnetic Resonance, School of Physics and Electronic Science, East China Normal University, Shanghai 200062, China 2. Institute of Magnetic Resonance and Molecular Imaging in Medicine, East China Normal University, Shanghai 200062, China 3. Innovation Institute, Shanghai Neusoft Medical Systems Co., Ltd., Shanghai 201100, China 4. Obstetrics and Gynecology Hospital of Fudan University, Shanghai 200011, China
- Keywords:
- Keywords: postpartum hemorrhage placenta accreta deep learning radiomics blood loss
- PACS:
- R318；R714.46
- DOI:
- DOI:10.3969/j.issn.1005-202X.2025.11.018
- Abstract:
- Abstract: Objective To develop a predictive model integrating clinical features, deep learning (DL), and radiomics based on T2-weighted imaging for prenatal assessment of postpartum hemorrhage (PPH) risk in high-risk pregnant women. Methods A total of 538 pregnant women with ultrasound-reported high-risk placenta accrete were retrospectively enrolled and divided into training, internal test, and external test cohorts. A nnUNet model was trained for automatic placental segmentation. Univariate and multivariate analyses were conducted on clinical features to identify those associated with PPH. Quantitative radiomic features were extracted from the placental region, and a random forest model was developed to predict estimated blood loss (EBL) and PPH risk. A DenseNet-based multi-task DL model was trained to predict PPH risk, EBL, and placenta previa status. Finally, a DL-radiomics ensemble (DRE) model was constructed by integrating clinical features, DL outputs, and radiomics scores. Diagnostic performance was evaluated using the area under the receiver operating characteristic curve (AUC) and DeLong test. Results The DRE model achieved AUC values of 0.874 (95% CI: 0.792-0.951) and 0.836 (95% CI: 0.648-0.974) in the internal and external test cohorts, respectively, significantly outperforming the standalone clinical, DL, and radiomics models. Incorporation of EBL regression improved the performance of the PPH classification model, with the external test AUC increasing from 0.261-0.788 to 0.836. Conclusion The DRE model integrating DL and radiomics can efficiently predict PPH risk and assist in the clinical management of high-risk pregnancies.
Last Update: 2025-12-01