AUTHOR=Jiang Yi , Li Wuchao , Huang Chencui , Tian Chong , Chen Qi , Zeng Xianchun , Cao Yin , Chen Yi , Yang Yintong , Liu Heng , Bo Yonghua , Luo Chenggong , Li Yiming , Zhang Tijiang , Wang Rongping
TITLE=Preoperative CT Radiomics Predicting the SSIGN Risk Groups in Patients With Clear Cell Renal Cell Carcinoma: Development and Multicenter Validation
JOURNAL=Frontiers in Oncology
VOLUME=10
YEAR=2020
URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2020.00909
DOI=10.3389/fonc.2020.00909
ISSN=2234-943X
ABSTRACT=Objective: The stage, size, grade, and necrosis (SSIGN) score can facilitate the assessment of tumor aggressiveness and the personal management for patients with clear cell renal cell carcinoma (ccRCC). However, this score is only available after the postoperative pathological evaluation. The aim of this study was to develop and validate a CT radiomic signature for the preoperative prediction of SSIGN risk groups in patients with ccRCC in multicenters.
Methods: In total, 330 patients with ccRCC from three centers were classified into the training, external validation 1, and external validation 2 cohorts. Through consistent analysis and the least absolute shrinkage and selection operator, a radiomic signature was developed to predict the SSIGN low-risk group (scores 0–3) and intermediate- to high-risk group (score ≥ 4). An image feature model was developed according to the independent image features, and a fusion model was constructed integrating the radiomic signature and the independent image features. Furthermore, the predictive performance of the above models for the SSIGN risk groups was evaluated with regard to their discrimination, calibration, and clinical usefulness.
Results: A radiomic signature consisting of sixteen relevant features from the nephrographic phase CT images achieved a good calibration (all Hosmer–Lemeshow p > 0.05) and favorable prediction efficacy in the training cohort [area under the curve (AUC): 0.940, 95% confidence interval (CI): 0.884–0.973] and in the external validation cohorts (AUC: 0.876, 95% CI: 0.811–0.942; AUC: 0.928, 95% CI: 0.844–0.975, respectively). The radiomic signature performed better than the image feature model constructed by intra-tumoral vessels (all p < 0.05) and showed similar performance with the fusion model integrating radiomic signature and intra-tumoral vessels (all p > 0.05) in terms of the discrimination in all cohorts. Moreover, the decision curve analysis verified the clinical utility of the radiomic signature in both external cohorts.
Conclusion: Radiomic signature could be used as a promising non-invasive tool to predict SSIGN risk groups and to facilitate preoperative clinical decision-making for patients with ccRCC.