Predicting renal-cell carcinoma recurrence after partial nephrectomy: a CT-radiomics approach.
Authors
Affiliations (4)
Affiliations (4)
- Sechenov University, Moscow, Russian Federation.
- National Research Nuclear University MEPhI (Moscow Engineering Physics Institute), Moscow, Russian Federation.
- Design Information Technologies Center Russian Academy of Sciences, Moscow, Russian Federation.
- Sechenov University, Moscow, Russian Federation. [email protected].
Abstract
About 30-40% of patients with renal-cell carcinoma experience recurrence within 5 years after partial nephrectomy despite current prognostic scoring systems. Radiomic analysis of preoperative CT images has the potential to offer improved risk stratification when compared with conventional clinicopathological variables. To identify radiomic and clinical prognostic features associated with RCC recurrence after partial nephrectomy and to develop an integrated machine learning model for enhanced recurrence risk stratification. The retrospective study included 190 RCC patients who underwent laparoscopic partial nephrectomy from 2011 to 2022. Preoperative contrast-enhanced CT scans were analyzed using radiomic feature extraction. Analyzed clinical variables included age, BMI, nephrometry scores (RENAL, PADUA, Centrality index), and tumor volume. Three machine learning algorithms (Random Forest, Gradient Boosting, and Logistic Regression) were trained to predict recurrence, with SHAP analysis for feature importance evaluation. The Gradient Boosting model achieved the highest predictive performance with ROC-AUC of 0.744, followed by Random Forest (0.722) and Logistic Regression (0.689). SHAP analysis revealed that 50% of the top 10 predictive features were radiomic parameters (Energy, Max, Median, RMS, and Kurtosis), while key clinical predictors included RENAL score, PADUA score, and Centrality index. The use of CT-based radiomic features has demonstrated initial potential in predicting recurrence of renal cell carcinoma (RCC) following partial nephrectomy. Combined radiomics-clinical machine learning models demonstrate moderate prognostic accuracy and may be utilized in future multicenter studies involving larger patient cohorts and external validation prior to clinical implementation.