Importance of phase enhancement for machine learning classification of solid renal masses using texture analysis...

Area of Science:

• Radiology
• Medical Imaging
• Machine Learning

Background:

• Accurate classification of solid renal masses is crucial for patient management.
• Texture analysis (TA) offers quantitative features for image-based diagnosis.
• Machine learning (ML) can leverage TA features for improved diagnostic accuracy.

Purpose of the Study:

• To compare the efficacy of ML-based TA features for classifying solid renal masses across different CT phases.
• To evaluate the diagnostic performance for differentiating renal cell carcinoma (RCC) from benign tumors and clear cell RCC (cc-RCC) from other subtypes.

Main Methods:

• Retrospective analysis of 177 solid renal masses using CT (NCCT, CM-CECT, NG-CECT).
• Extraction of 25 TA features from each CT phase.
• XGBoost model used for classification of RCC vs. benign and cc-RCC vs. other tumors.

Main Results:

• ML-TA achieved comparable diagnostic accuracy across NCCT, CM-CECT, and NG-CECT phases.
• No significant improvement in AUC when combining data from multiple phases.
• Models trained on one CT phase showed decreased performance when tested on a different phase, particularly between NCCT and CECT.

Conclusions:

• CT phase is not a critical factor for ML-based TA classification of renal masses.
• Phase-dependent model performance highlights challenges for multi-institutional studies using varied CT protocols.
• Standardization of CT protocols may be necessary for robust ML applications in renal mass analysis.