Ultrasound radiomics for distinguishing early and middle-late stage endometrial cancer
10.13929/j.issn.1003-3289.2024.11.022
- VernacularTitle:基于超声影像组学鉴别早期与中晚期子宫内膜癌
- Author:
Xiaoli PENG
1
;
Xueying WANG
;
Lu ZHAO
;
Shichun WANG
;
Menglin LUO
;
Lin REN
;
Maochun ZHANG
Author Information
1. 川北医学院附属医院超声科,四川 南充 637000
- Keywords:
endometrial neoplasms;
neoplasm staging;
ultrasonography;
machine learning;
radiomics
- From:
Chinese Journal of Medical Imaging Technology
2024;40(11):1739-1744
- CountryChina
- Language:Chinese
-
Abstract:
Objective To observe the value of ultrasound radiomics for distinguishing early and middle-late stage endometrial cancer(EC).Methods A total of 294 women with EC were retrospectively enrolled,including 196 in early stage and 98 in middle-late stage.The patients were randomly divided into training set(n=206)and validation set(n=88)at the ratio of 7∶3.Clinical data were compared between different stages,and a clinical model was constructed.Radiomics features were extracted and screened based on ultrasound data,and radiomics models were constructed with logistic regression(LR),random forest(RF),support vector machine(SVM),Gaussian naive Bayes(GNB)and extreme gradient boosting(XGBoost),respectively.Finally,a clinical-radiomics model was constructed.The value of each model for distinguishing early and middle-late stages EC was observed.Results Significant differences of age of consultation,menstrual disorders,abdominal pain and proportion of menopause were found between patients with early and middle-late stage EC(all P＜0.05).Among these 5 radiomics models,RF model had the highest area under the curve(AUC)for distinguishing early and middle-late stage EC.Pairwise comparison of clinical model,RF radiomics model and clinical-RF radiomics model showed that significant differences of AUC were found between each 2 models(all P＜0.05),and clinical-RF radiomics model had the highest AUC.Conclusion Ultrasound radiomics based on RF were helpful for distinguishing early and middle-late stage EC,and better diagnostic efficacy could be obtained through combining with clinical data.
