Exploratory study of WHO/ISUP classification of renal clear cell carcinoma pre-scholarly prediction based on ultrasonographic radiomics
10.3760/cma.j.cn131148-20230301-00121
- VernacularTitle:基于超声影像组学对肾透明细胞癌WHO/ISUP分级的探索研究
- Author:
Dai ZHANG
1
;
Lihui ZHAO
;
Hailing WANG
;
Jie MU
;
Fan YANG
;
Yiran MAO
;
Wenjing HOU
;
Xi WEI
Author Information
1. 天津医科大学肿瘤医院超声诊疗科 国家恶性肿瘤临床医学研究中心 天津市肿瘤防治重点实验室 天津市恶性肿瘤临床医学研究中心，天津 300060
- Keywords:
Ultrasonography;
Clear cell renal cell carcinoma;
Radiomics;
World Health Organization/International Society of Urological Pathology (WHO/ISUP) classific
- From:
Chinese Journal of Ultrasonography
2023;32(9):801-806
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To predict the clinical value of World Health Organization/International Society of Urological Pathology (WHO/ISUP) grading of clear cell renal cell carcinoma (ccRCC) pre-scholarly based on ultrasound imaging group.Methods:Clinical and ultrasound imaging data of patients with surgically pathologically confirmed ccRCC at Tianjin Medical University Cancer Institue and Hospital from January 2021 to October 2022 were retrospectively collected and divided into a low grade group (grade Ⅰ and Ⅱ, 105 cases) and a high grade group (grade Ⅲ and Ⅳ, 70 cases) using WHO/ISUP pathological grading criteria. The clear image of the largest diameter of the tumor was selected and imported into ITK-SNAP software for manual segmentation of the image and extraction of ultrasonographic radiomics features. The patients were randomly divided into a training group and a test group in the ratio of 7∶3, with 122 cases in the training group and 53 cases in the test group. Stable radiomics features were obtained by dimensionality reduction. The support vector machines (SVM) algorithm was applied to predict the pathological grading of ccRCC. Finally, a clinical-ultrasound imaging model, an ultrasonographic radiomics model and a comprehensive model combining the two were constructed. The predictive effects of the three models were analyzed by the area under the ROC curve (AUC). The performance of each model was evaluated by applying the calibration curve. The net benefit of patients was obtained by applying the decision curve.Results:A total of 873 radiomics features were extracted, and 10 features were finally obtained for model construction after dimensionality reduction. Final test results showed that the AUC, sensitivity, specificity and accuracy of the clinical-ultrasound imaging model were 0.68, 0.47, 0.78, 0.66. The AUC, sensitivity, specificity and accuracy of the ultrasonographic radiomics model were 0.74, 0.53, 0.88, 0.74. The AUC, sensitivity, specificity and accuracy of the comprehensive model were 0.84, 0.63, 0.86, 0.77. The AUC of the comprehensive model being larger than that of the clinical-ultrasound imaging model ( Z=-3.224, P=0.001) and ultrasonographic radiomics model ( Z=-2.594, P=0.009). The calibration curves showed that the comprehensive model was more stable than the other two models. The decision curve showed a higher net clinical benefit for the comprehensive model than for the other two models within a threshold of 0.1-1.0. Conclusions:The preoperative prediction of ccRCC pathological grading by the radiomics model based on ultrasound images is effective. The comprehensive model constructed by combining relevant clinical and ultrasound parameters has better performance, which can help predict ccRCC pathological grading preoperatively to a certain extent. It is crucial to help physicians choose the best management plan in the era of personalized medicine.
