In Placenta ; h5-index 40.0
INTRODUCTION : Fetal growth restriction (FGR) is associated with placental abnormalities, and its precise diagnosis is challenging. This study aimed to explore the role of radiomics based on placental MRI in predicting FGR.
METHODS : A retrospective study using T2-weighted placental MRI data were conducted. A total of 960 radiomic features were automatically extracted. Features were selected using three-step machine learning methods. A combined model was constructed by combining MRI-based radiomic features and ultrasound-based fetal measurements. The receiver operating characteristic curves (ROC) were conducted to assess model performance. Additionally, decision curves and calibration curves were performed to evaluate prediction consistency of different models.
RESULTS : Among the study participants, pregnant women who delivered from January 2015 to June 2021 were randomly divided into training (n = 119) and test (n = 40) sets. Forty-three other pregnant women who delivered from July 2021 to December 2021 were used as the time-independent validation set. After training and testing, three radiomic features that were strongly correlated with FGR were selected. The area under the ROC curves (AUCs) of the MRI-based radiomics model reached 0.87 (95% confidence interval [CI]: 0.74-0.96) and 0.87 (95% CI: 0.76-0.97) in the test and validation sets, respectively. Moreover, the AUCs for the model comprising MRI-based radiomic features and ultrasound-based measurements were 0.91 (95% CI: 0.83-0.97) and 0.94 (95% CI: 0.86-0.99) in the test and validation sets, respectively.
DISCUSSION : MRI-based placental radiomics could accurately predict FGR. Moreover, combining placental MRI-based radiomic features with ultrasound indicators of the fetus could improve the diagnostic accuracy of FGR.
Song Fuzhen, Li Ruikun, Lin Jing, Lv Mingli, Qian Zhaoxia, Wang Lisheng, Wu Weibin
2023-Feb-24
Fetal growth restriction, Machine learning, Magnetic resonance imaging, Radiomics, Ultrasound
