Texture analysis as a radiomic marker for differentiating renal tumors

Area of Science:

• Radiology
• Oncology
• Medical Imaging Analysis

Background:

• Renal tumors, including various subtypes of renal cell carcinoma and oncocytoma, present diagnostic challenges.
• Accurate differentiation is crucial for appropriate treatment planning and patient management.
• Non-invasive imaging techniques are sought to improve diagnostic accuracy and reduce the need for invasive procedures.

Purpose of the Study:

• To evaluate the utility of texture analysis for differentiating renal tumors.
• To assess the capability of texture analysis in distinguishing between renal cell carcinoma subtypes and oncocytoma.
• To determine if machine learning enhances the diagnostic performance of texture analysis.

Main Methods:

• Retrospective analysis of abdominal computed tomography (CT) examinations in patients with pathology-proven renal tumors.
• Manual segmentation of tumors followed by texture analysis of CT images.
• Application of a support vector machine (SVM) for tumor classification and comparison with texture analysis results.

Main Results:

• Histogram-based texture features, such as skewness and kurtosis, demonstrated high accuracy (AUCs 0.91-0.93) in differentiating clear cell subtype from oncocytoma.
• The median histogram feature showed excellent discrimination (AUC 0.99) between papillary subtype and oncocytoma.
• Machine learning significantly improved discrimination, achieving excellent AUCs (0.91-0.92) for distinguishing tumor subtypes from others.

Conclusions:

• Texture analysis is a promising non-invasive tool for distinguishing renal tumors on CT images.
• Machine learning integration further enhances the diagnostic performance of texture analysis.
• Texture analysis holds potential as a quantitative biomarker for differentiating various renal tumors.