Value of ultrasound-based radiomics combined with clinical parameters in predicting lymph node metastasis of endometrial cancer / 中华超声影像学杂志

ABSTRACT

Objective:To explore the value of ultrasound-based radiomics combined with clinical parameters in predicting lymph node metastasis of endometrial cancer.Methods:The clinical data and preoperative ultrasound images of 255 patients with endometrial cancer diagnosed after operation in the First Affiliated Hospital of Guangxi Medical University from January 2017 to February 2021 were analyzed retrospectively. The boundary of the tumor, that is, the region of interest(ROI), was determined manually by ITK-SNAP software, and the features were extracted from it. Through random sampling, the data were divided into training group used to build the model and verification group used to evaluate the reliability of the model at the proportion of 7∶3.In the training group, the spearman correlation coefficient was used as the threshold of 0.95 to remove the features with high correlation. The importance of the features were ranked by the decision tree model, the features of the top 6% importance were selected to establish a logical regression model, and then calculated the Radiomics score of each patient. On the basis of logical regression analysis, a comprehensive prediction model was established, the line diagram was drawn, and the prediction effect was evaluated by the analysis of the area under the subject working curve (AUC), calibration curve and the decision curve.Results:After dimensionality reduction and logical regression analysis, 20 features and 2 clinical parameters (CA125, CA153) were obtained to build the model. The AUC of training group and verification group in radiomics model and comprehensive prediction model were 0.81, 0.78 and 0.89, 0.85, respectively. The decision curve analysis showed that both of them had good clinical utility.Conclusions:The ultrasound-based radiomics model and the comprehensive model combined with relevant clinical parameters are effective in predicting lymph node metastasis of endometrial cancer.