Predicting Value of Multi-b-value DWI Models for Pathological Grading of Clear Cell Renal Cell Carcinoma: A Multi-scanner Study

Motivation: This study evaluates the diagnostic value of multi-b-value diffusion models for pathological grading of ccRCC across different MRI scanners. Goal(s): To assess the diagnostic value of multi-b-value DWI models and evaluate if advanced models offer significant improvements. Approach: A multi-scanner, retrospective study with MRI data from 285 patients, divided into training, testing, and validation sets. Diffusion maps were created using six models, with radiomic features selected via statistical methods. Predictive models were built using LASSO regression and evaluated using ROC curves, DeLong tests, and SHAP analysis. Results: Diffusion models showed effective diagnostic value for ccRCC grading, with no significant advantage over mono-exponential DWI. Impact: This study provides valuable insights into the utility of multi-b-value DWI models for grading ccRCC, suggesting that advanced models offer limited additional benefits over mono-exponential DWI. It enables more efficient imaging choices and informs clinical decision-making in ccRCC management.