Machine learning-based radiomics from multiparametric MRI for predicting aggressive pathology in clear cell renal...

Area of Science:

• Radiology
• Oncology
• Artificial Intelligence

Background:

• Clear cell renal cell carcinoma (ccRCC) presents significant heterogeneity, with aggressive subtypes posing a poor prognosis.
• Accurate preoperative differentiation between aggressive and indolent ccRCC is crucial for tailored patient management but remains a clinical challenge.

Purpose of the Study:

• To evaluate the efficacy of machine learning models utilizing multiparametric MRI radiomics for distinguishing aggressive from indolent ccRCC.
• To identify the most effective MRI sequences and radiomic features for this classification task.

Main Methods:

• A retrospective analysis of 157 ccRCC patients (114 indolent, 43 aggressive) was conducted.
• Regions of interest were delineated on five MRI sequences, and 31 feature combinations were extracted.
• 168 classification models were built and compared using various classifiers and feature selection methods.

Main Results:

• Aggressive ccRCC tumors were significantly larger than indolent ones (8.3 cm vs. 3.0 cm).
• Radiomic features from T2-weighted imaging (T2WI) showed the highest performance.
• The optimal model (RF+ICAP) achieved an AUC of 0.960, with 86.1% accuracy, 86.4% sensitivity, and 86.0% specificity.

Conclusions:

• Renal T2WI radiomics features offer superior discriminative power for differentiating aggressive from indolent ccRCC compared to T1WI and contrast-enhanced T1WI.
• An optimized classification model integrating multiple algorithms demonstrates potential for distinguishing aggressive ccRCC pathology preoperatively.