CT radiomics for predicting postoperative disease specific survival in clear cell renal cell carcinoma: a...

Clear cell renal cell carcinoma (ccRCC) is biologically heterogeneous, and existing clinical models provide limited precision for postoperative risk stratification. This multi-center study aimed to develop a computed tomography (CT)-based radiomics model to predict postoperative disease-specific survival (DSS) in ccRCC patients and to explore the correlations between radiomics and pathomics features.

A total of 530 ccRCC patients from four datasets were enrolled and divided into one training and three validation datasets. A total of 1,502 radiomics features were extracted from CT images using Pyradiomics, screened, and utilized to develop a radiomics model for predicting DSS with least absolute shrinkage and selection operator (LASSO) Cox regression. Clinical and fusion models were constructed and comprehensively assessed using the concordance index (C-index), Akaike information criterion (AIC), integrated Brier score (IBS), and time-dependent receiver operating characteristic analysis. Pathomic features were extracted from whole slide images (WSIs) using a pre-trained ResNet50. The top 20 prognostic features underwent correlation analysis with radiomics features. A multi-scale radiomics and pathomics (radiopathomics) model was then developed for prognostic evaluation.

The radiomics model identified three key features and demonstrated good predictive ability for ccRCC DSS. This model strengthened clinical prognostic evaluations, and the fusion model offered superior prediction accuracy and fit across four datasets, with C-indices of 0.804, 0.855, 0.830, and 0.814, respectively. Correlation analysis showed significant correlations between radiomics and pathomics features. The multi-scale radiopathomics model achieved exceptional predictive performance on the training dataset, with a C-index of 0.884.

This multi-center study developed a CT-based radiomics model for predicting DSS in ccRCC patients, enhancing the prognostic evaluations of the clinical model. Additionally, it revealed a significant correlation and complementarity between radiomics and pathomics features.