Study of the Risk Prediction Model of Preeclampsia

Preeclampsia is the main cause of increased maternal and perinatal mortality during pregnancy. Preeclampsia is mainly manifested as hypertension, urine protein, or damage symptoms of other target organs after 20 weeks of pregnancy. In preeclampsia high-risk group, early intervention and prevention of aspirin treatment can reduce preeclampsia or reduce its complications. Some serological biomarkers, such as placental protein 13 and placental growth factor, are closely related to preeclampsia. The clinical manifestations of preeclampsia are diverse, and the biomarkers distribution of early and late preeclampsia is also different. Multivariate models will be the trend for the prediction of risk of preeclampsia. The deep learning model can train the algorithm layer by layer by unsupervised learning method, and then use the supervised back propagation algorithm for tuning. It has strong capability and flexibility, and has been successfully applied in medical fields, such as the diagnosis of skin cancer.

In this study, maternal clinical data, routine laboratory indicators and biological markers in early pregnancy will be combined, and a deep learning method based on multiple models will be adopted to establish a risk prediction model for early preeclampsia, so as to improve the clinical ability for early diagnosis of preeclampsia. The deep learning method reduces the number of parameters by using spatial relative relation, which can improve the prediction ability of the model. Multi-model method is a less commonly used modeling method, and the models established by this method generally have better stability.

This project combines the above two methods to establish a risk prediction model for preeclampsia, and the research is of great significance.