Application of Renal Ultrasound Deep Learning in the Early Detection of Renal Impairment in Pregnant Women with Preeclampsia
10.3969/j.issn.1005-5185.2025.04.016
- VernacularTitle:肾脏超声深度学习早期预测子痫前期孕妇肾功能损害的应用价值
- Author:
Yingzi LIANG
1
;
Fangyi HUANG
1
;
Han YUAN
1
;
Qun HUANG
1
;
Yong GAO
1
Author Information
1. 广西医科大学第一附属医院超声科,广西 南宁 530021
- Publication Type:Journal Article
- Keywords:
Preeclampsia;
Ultrasonography;
Deep learning;
Kidney;
Impaired renal function;
Forecasting
- From:
Chinese Journal of Medical Imaging
2025;33(4):416-421,427
- CountryChina
- Language:Chinese
-
Abstract:
Purpose To construct a comprehensive model of deep learning features and clinical features based on renal ultrasound for early identification of renal impairment in the pregnant women with preeclampsia.Materials and Methods The information of 279 pregnant women in the First Affiliated Hospital of Guangxi Medical University from January 2018 to June 2023 were retrospectively collected,and all pregnant women were divided the into preeclampsia group(151 cases)and normal group(128 cases).The dataset was randomly divided into a training set(195 samples)and a testing set(84 samples)at a ratio of 7∶3.Based on ultrasound images,the deep learning convolutional neural networks Resnet152 was used to extract deep learning features.The non-zero coefficient features were selected from the deep learning features by the least absolute shrinkage and selection operator,and the K-nearest neighbor algorithm was used to establish the deep learning model.Then,the same classifier model was used to construct a comprehensive model based on clinical data.The receiver operating characteristic curve was used to evaluate the prediction effect.To address the interpretability visualization of models using gradient_weighted class activation mapping and SHapley Additive exPlanations(SHAP)values.Results The area under the curve of the composite model was 0.964(95%CI 0.940-0.988)in the training cohort and 0.899(95%CI 0.835-0.963)in the test cohort.SHAP analysis showed that deep learning features contributed the highest value in the prediction model.Conclusion The comprehensive model based on deep learning combined with clinical features of renal ultrasound can be used to identify renal impairment in normal pregnancy and preeclampsia pregnant women at an early stage,which is conducive to early clinical intervention.
