Role of DECT-Based Imaging Biomarkers and Machine Learning to Predict Renal Cell Carcinoma Subtypes

Area of Science:

• Radiology
• Oncology
• Medical Imaging

Background:

• Renal cell carcinoma (RCC) is a common malignancy.
• Accurate differentiation between clear cell RCC (ccRCC) and non-ccRCC is crucial for treatment planning.
• Dual-energy CT (DECT) offers quantitative imaging biomarkers that may aid in tumor characterization.

Purpose of the Study:

• To assess and compare quantitative parameters derived from DECT for differentiating ccRCC from non-ccRCC.
• To evaluate the accuracy of machine learning (ML) models in classifying ccRCC subtypes and predicting tumor grade using DECT data.

Main Methods:

• Retrospective analysis of 112 RCC patients who underwent DECT prior to surgery.
• Measurement of iodine concentration (IC) and iodine ratio (IR) by two independent readers.
• Development and evaluation of ML models (Random Forest, AdaBoost) for classification and grading tasks.

Main Results:

• Individual DECT parameters (IC, IR) showed moderate accuracy (77.7%, 77.5%) in distinguishing ccRCC from non-ccRCC.
• ML models, particularly AdaBoost, achieved higher accuracy, reaching 100% when combined, for ccRCC classification.
• DECT parameters and ML models demonstrated similar accuracy (around 77%) for predicting ccRCC grade, with combined ML models showing improved performance.

Conclusions:

• DECT-based quantitative imaging biomarkers provide moderate diagnostic accuracy for ccRCC differentiation.
• Machine learning significantly enhances the diagnostic performance of DECT for ccRCC classification and grading.
• ML integration with DECT holds promise for improved non-invasive characterization of renal tumors.