Multidimensional CT radiomics for preoperative prediction of TFE3-rearranged renal cell carcinoma
10.3760/cma.j.cn112330-20250325-00117
- VernacularTitle:基于多维度特征的CT影像组学术前诊断TFE3重排肾细胞癌的模型研究
- Author:
Bin XIA
1
;
Chengwei CHEN
1
;
Na LI
1
;
Yun BIAN
1
;
Chengwei SHAO
1
;
Jianping LU
1
;
Qinqin KANG
1
Author Information
1. 海军军医大学第一附属医院放射诊断科，上海 200433 夏斌为上海理工大学研究生，上海 200093
- Publication Type:Journal Article
- Keywords:
Carcinoma，renal cell;
TFE3-rearranged;
Habitat imaging;
Peritumoral radiomics;
Radiomics
- From:
Chinese Journal of Urology
2025;46(5):343-348
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To develop a preoperative CT-based radiomics model integrating multidimensional features for the accurate prediction of TFE3-rearranged renal cell carcinoma（TFE3-rRCC）.Methods:This study retrospectively enrolled 865 pathologically confirmed renal cell carcinoma（RCC）patients in The First Affiliated Hospital of Naval Medical University from June 2013 to June 2023，including 60 cases of TFE3-rRCC and 805 cases of non-TFE3 RCC（comprising clear cell RCC，papillary RCC，and chromophobe RCC）. Among them，627 were male and 238 were female，with a mean age of（54.1 ± 12.7）years（range：14?82 years）. The median maximum tumor diameter was 4.0（2.6，6.0）cm. Based on the chronological order of CT examinations，the patients were divided into training（ n=478），validation（ n=206），and test（ n=181）sets in an approximate 6∶2∶2 ratio. Using precontrast and corticomedullary phase CT images，we extracted peritumoral imaging features，habitat features，3D radiomic features，and 2.5D deep learning radiomic features. A deep learning radiomics score（DLR-SCORE）prediction model was constructed using least absolute shrinkage and selection operator（LASSO）regression. The diagnostic performance of the model was evaluated by receiver operating characteristic（ROC）curve analysis，with the area under the curve（AUC）as the primary metric. Additionally，sensitivity，specificity，and accuracy were calculated based on the confusion matrix. Results:A total of 12 442 features were extracted from non-contrast and corticomedullary phase CT images，from which eight key features were selected to construct the DLR-SCORE model. The model demonstrated diagnostic accuracies for TFE3-rRCC of 98.5%（471/478）in the training set，81.6%（168/206）in the validation set，and 86.2%（156/181）in the test set. The AUC of ROC curve was 0.98（95% CI 0.96?1.00）in the training set，0.83（95% CI 0.71?0.94）in the validation set，and 0.88（95% CI 0.76?1.00）in the test set. In the test set，the DLR-SCORE model achieved a sensitivity of 88.9%（16/18）and a specificity of 85.9%（140/163）for detecting TFE3-rRCC. Conclusions:The DLR-SCORE model integrating multidimensional CT radiomics features demonstrated favorable predictive performance for TFE3-rRCC，offering a promising noninvasive tool to assist preoperative diagnosis.
