Establishment of a prediction model combined CT-radiomics and clinical features for differentiating benign and malignant renal tumors
10.3760/cma.j.cn112330-20240607-00258
- VernacularTitle:基于CT影像组学和临床特征构建的预测模型对肾肿瘤良恶性的预测能力
- Author:
Yafeng FAN
1
;
Shuanbao YU
;
Zeyuan WANG
;
Haoke ZHENG
;
Wendong JIA
;
Meng WANG
;
Xuepei ZHANG
Author Information
1. 郑州大学第一附属医院泌尿外科，郑州 450052
- Publication Type:Journal Article
- Keywords:
Kidney neoplasms;
Renal cell carcinoma;
Benign renal lesion;
Radiomics;
Machine learning
- From:
Chinese Journal of Urology
2025;46(2):91-96
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the efficacy of a predictive model for differentiating benign and malignant renal tumors based on CT radiomic features and clinical features.Methods:A retrospective study was conducted on 1 395 patients with renal tumors admitted to the First Affiliated Hospital of Zhengzhou University from December 2011 to December 2021, including 842 males and 553 females. The median age was 55 (44, 59) years, and the median tumor diameter was 3.6 (2.7, 4.6) cm. All patients underwent contrast-enhanced CT scaning before surgery, and radiomic features were extracted from non-contrast, arterial, and venous phase images. Prediction models for distinguishing benign and malignant renal tumors were constructed using five machine learning algorithms (logistic regression, support vector machine, neural network, random forest, and extreme gradient boosting), and these models were then ensembled to construct a stacking classifier. All patients underwent partial nephrectomy, and they were divided into a training group (941 cases, December 2011 to June 2020) and a validation group (454 cases, July 2020 to December 2021) based on the date of surgery. A clinical-radiomic model was developed by combining the result of stacking classifier, clinical features and CT report results, and its predictive performance was evaluated in the validation group.Results:The radiomic signature based on the combined features and five machine learning algorithms(AUC 0.835-0.844) showed higher accuracy in predicting benign and malignant renal tumors compared to single phases (AUC 0.744-0.831). After integrating the five machine learning algorithms, the AUC of the three-phase combined radiomic model in the validation group improved to 0.847(95% CI 0.802-0.892). The clinical-radiomic model, incorporating radiomic features, clinical features, and CT report results, achieved a significantly higher AUC in the validation group compared to radiologists [0.919(95% CI 0.889-0.950)vs. 0.835(95% CI 0.786-0.883), P<0.01]. Conclusions:The predictive model integrating CT radiomics features, clinical characteristics, and CT report results demonstrates excellent discriminative ability in distinguishing benign and malignant renal tumors.
