Noninvasive renal mass diagnostic models based on radiomic features and clinical data: a multi-center study

Area of Science:

• Medical Imaging
• Oncology
• Data Science

Background:

• Approximately 20% of surgically treated renal masses are benign.
• Accurate preoperative differentiation of renal masses is crucial for appropriate patient management.

Purpose of the Study:

• To develop and validate a radiomics-based strategy for the differential diagnosis of renal masses.
• To distinguish between benign and malignant renal masses and identify subtypes of renal cell carcinoma.

Main Methods:

• A retrospective study involving 726 renal masses (54 benign, 672 malignant) from multiple institutions and public databases.
• Development of two diagnostic models: Kidney Mass Diagnosis Model (KMDM) and Kidney Cancer Subtypes Diagnosis Model (KCSDM).
• Each model comprised radiomics, clinical, and integrated units, evaluated on training and independent validation sets.

Main Results:

• The integrated KMDM demonstrated superior performance (AUC 0.937 training, 0.942 validation) compared to individual models and expert assessment.
• A risk of malignant index (RMI) cutoff of 1.661 achieved 92.9% sensitivity and 83.3% specificity for distinguishing benign from malignant masses in validation sets.
• Similar high performance was observed for the integrated KCSDM in classifying renal cell carcinoma subtypes.

Conclusions:

• The integrated radiomics model provides a noninvasive and accurate method for differentiating benign from malignant renal masses.
• This strategy also effectively distinguishes between different subtypes of renal cell carcinoma.
• The developed models offer promising tools for improving the diagnostic accuracy of renal masses.