JMIR Publications

ABSTRACT

Background:

Kidney cancer is the most common malignant cancer of the urinary system, and nephrectomy is the standard treatment for locoregional kidney cancer

Objective:

Aimed to establish a survival prediction model for postoperative patients with kidney cancer based on CT radiomics.

Methods:

Radiomic features were extracted from ROI in CT images of 207 postoperative patients with kidney cancer. The eigenvalue data of all radiomics features were processed using z-score standardization and R software package ‘GLMNet’. We integrated survival time, survival status, and radiomics features and screened these features using LASSO-Cox regression method. We conducted 10-fold cross-validation to obtain optimal model to obtain five radiomics features. Multivariate Cox regression hazard models were established to analyze patients’ overall survival. The predictive ability of the nomogram (relative operating characteristic curve [ROC] and calibration curve) was evaluated in a training cohort and validated in a validation cohort. Patients were divided into high- and low-risk groups based on the Rad-score cutoff value, and the Kaplan–Meier method was conducted to identify established models’ forecasting ability. Five radiomic features were screened for predictive model construction.

Results:

ROC curve and area under curve showed that the predictive model performed well. The calibration curve of nomogram and radiomic features in cohort study set indicated an overall net benefit. KM curves indicated that OS time was dramatically shorter in high-risk group.

Conclusions:

Our model, combining CT-extracted radiomic features, had good potential for evaluating OS in patients treated with nephrectomy and may facilitate clinical management and the prognostic evaluation of postoperative patients with kidney cancer.