Predicting renal-cell carcinoma recurrence after partial nephrectomy: a CT-radiomics approach

Area of Science:

• Medical Imaging
• Oncology
• Machine Learning

Background:

• Partial nephrectomy is a common treatment for renal cell carcinoma (RCC).
• Despite surgery, 30-40% of RCC patients experience recurrence within 5 years.
• Current prognostic systems have limitations in predicting recurrence risk.

Purpose of the Study:

• To identify radiomic and clinical prognostic features for RCC recurrence post-partial nephrectomy.
• To develop an integrated machine learning model for enhanced recurrence risk stratification.

Main Methods:

• Retrospective analysis of 190 RCC patients undergoing laparoscopic partial nephrectomy.
• Preoperative contrast-enhanced CT scans analyzed for radiomic features.
• Machine learning models (Random Forest, Gradient Boosting, Logistic Regression) trained to predict recurrence, with SHAP analysis for feature importance.

Main Results:

• Gradient Boosting model showed highest predictive performance (ROC-AUC 0.744).
• Top predictive features included radiomic parameters (Energy, Max, Median, RMS, Kurtosis) and clinical factors (RENAL score, PADUA score, Centrality index).

Conclusions:

• CT-based radiomic features show potential for predicting RCC recurrence after partial nephrectomy.
• Combined radiomics-clinical machine learning models offer moderate prognostic accuracy.
• Further multicenter studies with external validation are needed for clinical implementation.