Early research of applying contrast-enhanced ultrasound radiomics model to forecast pathological grades in bladder urothelial carcinoma
10.3760/cma.j.cn131148-20250428-00241
- VernacularTitle:基于超声造影影像组学模型预测膀胱尿路上皮癌病理分级的初步研究
- Author:
Wen LI
1
;
Hua HONG
;
Qian LIU
;
Yang LIU
;
Danyan LIANG
;
Senlin BAO
;
Heyang LIU
Author Information
1. 内蒙古医科大学研究生院，呼和浩特 010110
- Publication Type:Journal Article
- Keywords:
Contrast-enhanced ultrasound;
Bladder urothelial carcinoma;
Radiomics;
Super-resolution reconstruction
- From:
Chinese Journal of Ultrasonography
2025;34(11):999-1006
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the predictive value of a machine learning model combining contrast-enhanced ultrasound（CEUS）parameters，radiomics features of ultrasound images，and clinical data for pathological grading in bladder urothelial carcinoma（BUC）.Methods:A retrospective analysis was conducted on 174 BUC patients from Inner Mongolia Autonomous Region People 's Hospital and the First Affiliated Hospital of Baotou Medical College from December 2017 to March 2024. One hundred and thirteen BUC patients from the former hospital were randomly divided into training group and internal test group in a ratio of 7 to 3，while 61 BUC patients from the latter hospital served as an external test group. The patients were stratified into low-grade bladder urothelial carcinoma（LGBUC）and high-grade bladder urothelial carcinoma（HGBUC）groups based on pathology. Two-dimensional grayscale ultrasound images were subjected to super-resolution（SR）reconstruction，followed by extraction and screening of radiomics features in comparison with CEUS video sequences. Selected features were input into a support vector machine（SVM）to build the radiomics model. CEUS parameters，conventional ultrasound metrics and clinical data with statistical significance between LGBUC and HGBUC groups were input into SVM to construct the clinical model. The radiomics and clinical model outputs were fused via multivariate Logistic regression to form a combined model. Model performances were evaluated using ROC curves，calibration curves，and clinical decision curves. Results:Seven radiomics features from SR images were used to build the radiomics model，while CEUS parameters（peak intensity and time-to-peak half），age，tumor-wall interface and tumor-wall angle formed the clinical model. The combined model integrated these outputs. All 3 models exhibited respective strengths，the combined model showed superior robustness. The AUCs of the combined model in the training，internal test and external test groups were 0.92，0.84 and 0.82，respectively.Conclusions:The combined model combining CEUS parameters，ultrasound radiomics features，and clinical data accurately predicts BUC pathological grade，providing a potential tool for clinical diagnosis and treatment.
