Tumor grade-titude: XGBoost radiomics paves the way for RCC classification

Area of Science:

• Radiology and Medical Imaging
• Machine Learning in Oncology
• Cancer Diagnostics

Background:

• Renal cell carcinoma (RCC) grading is crucial for treatment decisions.
• Accurate differentiation of high-grade from low-grade RCC non-invasively remains a challenge.
• Radiomics offers potential for quantitative image analysis in oncology.

Purpose of the Study:

• To develop and validate a non-invasive XGBoost machine learning model for differentiating grade 4 RCC from lower-grade tumors.
• To utilize radiomic features from pre-treatment CT images for this classification task.
• To assess the model's performance and potential clinical utility.

Main Methods:

• Extraction of radiomic features from contrast-enhanced CT scans of 102 RCC patients.
• Application of a two-step feature selection methodology to identify relevant features.
• Development and evaluation of an XGBoost-based machine learning model.

Main Results:

• The XGBoost model achieved high performance, with an AUC of 0.87 in training and 0.92 in testing sets.
• No significant difference was observed between training and testing performance (p=0.521).
• The model demonstrated high sensitivity, specificity, and predictive values, with selected features capturing intensity and spatial information.

Conclusions:

• The developed XGBoost radiomic model shows significant potential for non-invasively differentiating high-grade RCC.
• This tool could aid in personalized adjuvant immunotherapy decision-making and improve patient outcomes.
• Further validation in multicenter cohorts and integration with other data types are warranted.