Predicting Renal Cell Carcinoma Subtypes and Fuhrman Grading Using Multiphasic CT-Based Texture Analysis and Machine...

Area of Science:

• Radiology
• Medical Imaging
• Computational Pathology

Background:

• Clear cell renal cell carcinoma (ccRCC) is the most common subtype of kidney cancer.
• Accurate differentiation of ccRCC from non-ccRCC and grading are crucial for treatment planning.
• Current diagnostic methods may have limitations in precise classification and grading.

Purpose of the Study:

• To evaluate computed tomography texture analysis (CTTA) for differentiating ccRCC from non-ccRCC.
• To assess CTTA's ability to predict Fuhrman's grade in ccRCC.
• To utilize open-source Python libraries for these analyses.

Main Methods:

• Retrospective analysis of 144 patients with renal cell carcinoma (RCC).
• Multiphase CT scans analyzed using first- and second-order texture features computed with Python libraries (scipy, numpy, opencv).
• Machine learning models, including Support Vector Machine (SVM), applied for classification and prediction, with external validation.

Main Results:

• Entropy in the corticomedullary (CM) phase showed the best performance for ccRCC vs. non-ccRCC classification (F1 score: 0.83).
• An SVM model using CM phase features achieved the highest F1 score (0.87) for ccRCC differentiation, with external validation yielding 0.82 accuracy and 0.81 F1 score.
• Classification of ccRCC grades (low vs. high) showed lower accuracy, with a maximum F1 score of 0.76 for the CM phase SVM model.

Conclusions:

• CTTA using open-source Python tools is a valuable method for distinguishing ccRCC from non-ccRCC.
• CTTA demonstrates potential for predicting ccRCC tumor grade.
• The corticomedullary phase and SVM models are particularly effective for ccRCC classification.