A Novel Radiomics-based Interpretable Model for Bladder Cancer Grade Prediction Using White-Light Cystoscopy Images.

Area of Science:

• Urology
• Medical Imaging
• Artificial Intelligence

Background:

• White-light cystoscopy (WLC) is standard for bladder cancer diagnosis.
• Preoperative grading of bladder tumors using WLC remains challenging.
• Accurate grading is crucial for treatment planning and patient management.

Purpose of the Study:

• To develop a multichannel radiomics model for predicting bladder tumor grade (low-grade vs. high-grade) from WLC images.
• To identify imaging biomarkers associated with tumor grade.
• To assess the model's performance and interpretability.

Main Methods:

• Retrospective collection of WLC images from 423 patients across two centers.
• Extraction of radiomic features from greyscale and RGB channels.
• Training and validation of five machine-learning classifiers, including support vector machine (SVM).
• Performance evaluation using discrimination, calibration, and decision curve analysis (DCA).
• Interpretability analysis using SHapley Additive exPlanations (SHAP).

Main Results:

• The SVM-based radiomics model demonstrated robust performance with an AUC of 0.87 for internal validation and 0.79 for external validation.
• SHAP analysis identified distinct radiomic patterns differentiating low-grade from high-grade tumors.
• The green channel was identified as a significant contributor to grade prediction.

Conclusions:

• The developed multichannel radiomics model accurately predicts bladder tumor grade from WLC images.
• This approach can aid in identifying imaging biomarkers for intraoperative risk stratification.
• Potential for developing real-time, filter-based tools for improved bladder cancer management.